On a side note.. ya’ll must be prompt wizards if you can actually use the LLM code.
I use it for debugging sometimes to get an idea, or a quick sketch up of an UI.
As for actual code.. the code it writes is a huge mess of spaghetti code, overly verbose, with serious performance and security risks, and complete misunderstanding of pretty much every design pattern I give it..
I read AI coding negativity on Hacker News and Reddit with more and more astonishment every day. It's like we live in different worlds. I expect the breadth of tooling is partly responsible. What it means to you to "use the LLM code" could be very different from what it means to me. What LLM are we talking about? What context does it have? What IDE are you using?
Personally, I wrote 200K lines of my B2B SaaS before agentic coding came around. With Sonnet 4 in Agent mode, I'd say I now write maybe 20% of the ongoing code from day to day, perhaps less. Interactive Sonnet in VS Code and GitHub Copilot Agents (autonomous agents running on GitHub's servers) do the other 80%. The more I document in Markdown, the higher that percentage becomes. I then carefully review and test.
It’s not just you, I think some engineers benefit a lot from AI and some don’t. It’s probably a combination of factors including: AI skepticism, mental rigidity, how popular the tech stack is, and type of engineering. Some problems are going to be very straightforward.
I also think it’s that people don’t know how to use the tool very well. In my experience I don’t guide it to do any kind of software pattern or ideology. I think that just confuses the tool. I give it very little detail and have it do tasks that are evident from the code base.
Sometimes I ask it to do rather large tasks and occasionally the output is like 80% of the way there and I can fix it up until it’s useful.
> I also think it’s that people don’t know how to use the tool very well.
I think this is very important. You have to look at what it suggests critically, and take what makes sense. The original comment was absolutely correct that AI-generated code is way too verbose and disconnected from the realities of the application and large-scale software design, but there can be kernels of good ideas in its output.
* Code using sympy to generate math problems testing different skills for students, with difficulty values affecting what kinds of things are selected, and various transforms to problems possible (e.g. having to solve for z+4 of 4a+b instead of x) to test different subskills
(On this part, the LLM did pretty well. The code was correct after a couple of quick iterations, and the base classes and end-use interfaces are correct. There's a few things in the middle that are unnecessarily "superstitious" and check for conditions that can't happen, and so I need to work with the LLM to clean it up.
* Code to use IRT to estimate the probability that students have each skill and to request problems with appropriate combinations of skills and difficulties for each student.
(This was somewhat garbage. Good database & backend, but the interface to use it was not nice and it kind of contaminated things).
* Code to recognize QR codes in the corners of worksheet, find answer boxes, and feed the image to ChatGPT to determine whether the scribble in the box is the answer in the correct form.
(This was 100%, first time. I adjusted the prompt it chose to better clarify my intent in borderline cases).
The output was, overall, pretty similar to what I'd get from a junior engineer under my supervision-- a bit wacky in places that aren't quite worth fixing, a little bit of technical debt, a couple of things more clever that I didn't expect myself, etc. But I did all of this in three hours and $12 expended.
The total time supervising it was probably similar to the amount of time spent supervising the junior engineer... but the LLM turns things around quick enough that I don't need to context switch.
I think a lot of it is tool familiarity. I can do a lot with Cursor but frankly I find out about "big" new stuff every day like agents.md. If I wasn't paying attention or also able to use Cursor at home then I'd probably learn more inefficiently. Learning how to use rule globs versus project instructions was a big learning moment. As I did more LLM work on our internal tools that was also a big lesson in prompting and compaction.
Certain parts of HN and Reddit I think are very invested in nay-saying because it threatens their livelihoods or sense of self. A lot of these folks have identities that are very tied up in being craftful coders rather than business problem solvers.
People here work on all kinds of industries. Some of us are implementing JIT compilers, mission-critical embedded systems or distributed databases. In code bases like this you can't just wing it without breaking a million things, so LLM agents tend to perform really poorly.
Yes, it would be nice to have a lot more context (pun intended) when people post how many LoC they introduced.
B2B SaaS? Then can I assume that a browser is involved and that a big part of that 200k LoC is the verbose styling DSL we all use? On the other hand, Nginx, a production-grade web server, is 250k LoC (251,232 to be exact [1]). These two things are not comparable.
The point being that, as I'm sure we all agree, LoC is not a helpful metric for comparison without more context, and different projects have vastly different amounts of information/feature density per LoC.
On the other hand, fault-intolerant codebases are also often highly defined and almost always have rigorous automated tests already, which are two contexts where coding agents specifically excel in.
I am now making an emotional reaction based on zero knowledge of the B2B codebase's environment, but to be honest I think it is relevant to the discussion on why people are "worlds apart".
200k lines of code is a failure state. At this point you have lost control and can only make changes to the codebase through immense effort, and not at a tolerable pace.
Agentic code writers are good at giving you this size of mess and at helping to shovel stuff around to make changes that are hard for humans due to the unusable state of the codebase.
If overgrown barely manageble codebases are all a person's ever known and they think it's normal that changes are hard and time-consuming and needing reams of code, I understand that they believe AI agents are useful as code writers. I think they do not have the foundation to tell mediocre from good code.
I am extremely aware of the judgemental hubris of this comment. I'd not normally huff my own farts in public this obnoxiously, but I honestly feel it is useful for the "AI hater vs AI sucker" discussion to be honest about this type of emotion.
It really depends on what your use case is. E.g. of you're dealing with a lot of legacy integrations, dealing with all the edge cases can require a lot of code that you can't refactor away through cleverness.
Each integration is hopefully only a few thousand lines of code, but if you have 50 integrations you can easily break 100k loc just dealing with those. They just need to be encapsulated well so that the integration cruft is isolated from the core business logic, and they become relatively simple to reason about
My AI experience has varied wildly depending on the problem I'm working on. For web apps in Python, they're fantastic. For hacking on old engineering calculation code written in C/C++, it's an unmitigated disaster and an active hindrance.
And also ask: "How much money do you spend on LLMs?"
In the long run, that is going to be what drives their quality. At some point the conversation is going to evolve from whether or not AI-assisted coding works to what the price point is to get the quality you need, and whether or not that price matches its value.
There is the huge variance in prompt specificity as well as the subtle differences inherent to the models. People often don't give examples when they talk about their experiences with AI so it's hard to get a read on what a good prompt looks like for a given model or even what a good workflow is for getting useful code out of it.
Some gave. Some even recorded it, and showed it, because they thought that they are good with it. But they weren’t good at all.
They were slower than coding by hand, if you wanted to keep quality. Some were almost as quick as copy-pasting from the code just above the generated one, but their quality was worse. They even kept some bugs in the code during their reviews.
So the different world is probably what the acceptable level of quality means. I know a lot of coders who don’t give a shit whether it makes sense what they’re doing. What their bad solution will cause in the long run. They ignore everything else, just the “done” state next to their tasks in Jira. They will never solve complex bugs, they simply don’t care enough. At a lot of places, they are the majority. For them, LLM can be an improvement.
Claude Code the other day made a test for me, which mocked everything out from the live code. Everything was green, everything was good. On paper. A lot of people simply wouldn’t care to even review properly. That thing can generate a few thousands of lines of semi usable code per hour. It’s not built to review it properly. Serena MCP for example specifically built to not review what it does. It’s stated by their creators.
I deal with a few code bases at work and the quality differs a lot between projects and frameworks.
We have 1-2 small python services based on Flask and Pydantic, very structured and a well-written development and extension guide. The newer Copilot models perform very well with this, and improving the dev guidelines keep making it better. Very nice.
We also have a central configuration of applications in the infrastructure and what systems they need. A lot of similarly shaped JSON files, now with a well-documented JSON schema (which is nice to have anyway). Again, very high quality. Someone recently joked we should throw these service requests at a model and let it create PRs to review.
But currently I'm working in Vector and it's Vector remap language... it's enough of a mess that I'm faster working without any copilot "assistance". I think the main issue is that there is very little VRL code out in the open, and the remaps depend on a lot of unseen context, which one would have to work on giving to the LLM. Had similar experiences with OPA and a few more of these DSLs.
B2B SaaS in most cases are sophisticated masks over some structured data, perhaps with great ux, automation and convenience, so I can see LLMs be more successful there, even so because there is more training data and many processes are streamlined. Not all domains are equal, go try develop a serious game, not the yet another simple and broken arcade, with llms and you'll have a different take
I think its down to language and domain more than tools.
No model ive tried can write, usefully debug or even explain cmake. (It invents new syntax if it gets stuck, i often have to prompt multiple AI to know if even the first response in the context was made-up)
My luck with embedded c has been atrocious for existing codebase (burning millions of tolkens), but passable for small scripts. (Arduino projects)
My experience with python is much better. Suggesting relevant libraries and functions, debugging odd errors, or even making small script on its own. Even the original github copilot which i got access to early was excellent on python.
Alot of people that seem to have fully embraced agentic vibe-coding seem to be in the web or node.js domain. Which I've not done myself since pre-AI.
I've tried most (free or trial) major models or schemes in hope that i find any of them useful, but not found much use yet.
FWIW. Claude Code does great job for me on complex domain Rust projects, but I just use it one relatively small feature/code chunk at the time, where oftentimes it can pick up existing patterns etc. (I try to point it at similar existing code/feature if I have it). I do not let it write anything creative where it has to come up with own design (either high-level architectural, or low level facilities). Basically I draw the lines manually, and let it color the space between, using existing reference pictures. Works very, very well for me.
Is this meant to detract from their situation? These tech stacks are mainstream because so many use them... it's only natural that AI would be the best at writing code in contexts where it has the most available training data.
> These tech stacks are mainstream because so many use them
That's a tautology. No, those tech stacks are mainstream because it is easy to get something that looks OK up and running quickly. That's it. That's what makes a framework go mainstream: can you download it and get something pretty on the screen quickly? Long-term maintenance and clarity is absolutely not a strong selection force for what goes mainstream, and in fact can be an opposing force, since achieving long-term clarity comes with tradeoffs that hinder the feeling of "going fast and breaking things" within the first hour of hearing about the framework. A framework being popular means it has optimized for inexperienced developers feeling fast early, which is literally a slightly negative signal for its quality.
You are exactly right in my case - JavaScript and Python dealing with the AWS CDK and SDK. Where there is plenty of documentation and code samples.
Even when it occasionally gets it wrong, it’s just a matter of telling ChatGPT - “verify your code using the official documentation”.
But honestly, even before LLMs when deciding on which technology, service, or frameworks to use I would always go with the most popular ones because they are the easiest to hire for, easiest to find documentation and answers for and when I myself was looking for a job, easiest to be the perfect match for the most jobs.
They can choose jobs. Starting with my 3rd job in 2008, I always chose my employer based on how it would help me get my n+1 job and that was based on tech stack I would be using.
Once a saw a misalignment between market demands and current tech stack my employer was using, I changed jobs. I’m on job #10 now.
Fair: it was rude. Moderation is hard and I respect what you do. But it's also a sentiment several other comments expressed. It's the conversation we're having. Can we have any discussions of code quality without making assumptions about each others' code quality? I mean, yeah, I could probably have done better.
> "200k lines of code is a failure state ... I'd not normally huff my own farts in public this obnoxiously, but I honestly feel it is useful for the "AI hater vs AI sucker" discussion to be honest about this type of emotion." https://news.ycombinator.com/item?id=44976328
I understand the provocation, but please don't respond to a bad comment by breaking the site guidelines yourself. That only makes things worse.
Your GP comment was great, and probably the thing to do with a supercilious reply is just not bother responding (which is easier said than done of course). You can usually trust other users to assess the thread fairly (e.g.https://news.ycombinator.com/item?id=44975623).
> What makes you think I'm not "a developer who strongly values brevity and clarity"
Some pieces of evidence that make me think that:
1. The base rate of developers who write massively overly verbose code is about 99%, and there's not a ton of signal to deviate from that base rate other than the fact that you post on HN (probably a mild positive signal).
2. An LLM writes 80% of your code now, and my prior on LLM code output is that it's on par with a forgetful junior dev who writes very verbose code.
3. 200K lines of code is a lot. It just is. Again, without more signal, it's hard to deviate from the base rate of what 200K-line codebases look like in the wild. 99.5% of them are spaghettified messes with tons of copy-pasting and redundancy and code-by-numbers scaffolded code (and now, LLM output).
This is the state of software today. Keep in mind the bad programmers who make verbose spaghettified messes are completely convinced they're code-ninja geniuses; perhaps even more so than those who write clean and elegant code. You're allowed to write me off as an internet rando who doesn't know you, of course. To me, you're not you, you're every programmer who writes a 200k LOC B2B SaaS application and uses an LLM for 80% of their code, and the vast, vast majority of those people are -- well, not people who share my values. Not people who can code cleanly, concisely, and elegantly. You're a unicorn; cool beans.
Before you used LLMs, how often were you copy/pasting blocks of code (more than 1 line)? How often were you using "scaffolds" to create baseline codefiles that you then modified? How often were you copy/pasting code from Stack Overflow and other sources?
200kloc is what, 4 reams of paper, double sided? So, 10% of that famous Margaret Hamilton picture (which is roughly "two spaceships worth of flight code".) I'm not sure the intuition that gives you is good but at least it slots the raw amount in as "big but not crazy big" (the "9 years work" rather than "weekend project" measurement elsethread also helps with that.)
I use aider and your description doesn't match my experience, even with a relatively bad-at-coding model (gpt-5). It does actually work and it does generate "good" code - it even matches the style of the existing code.
Prompting is very important, and in an existing code base the success rate is immensely higher if you can hint at a specific implementation - i.e. something a senior who is familiar with the codebase somewhat can do, but a junior may struggle with.
It's important to be clear eyed about where we are here. I think overall I am still faster doing things manually than iterating with aider on an existing code base, but the margin is not very much, and it's only going to get better.
Even though it can do some work a junior could do, it can't ever replace a junior human... because a junior human also goes to meetings, drives discussions, and eventually becomes a senior! But management may not care about that fact.
I agree. AI is a wonderful tool for making fuzzy queries on vast amounts of information. More and more I'm finding that Kagi's Assistant is my first stop before an actual search. It may help inform me about vocabulary I'm lacking which I can then go successfully comb more pages with until I find what I need.
But I have not yet been able to consistently get value out of vibe coding. It's great for one-off tasks. I use it to create matplotlib charts just by telling it what I want and showing it the schema of the data I have. It nails that about 90% of the time. I have it spit out close-ended shell scripts, like recently I had it write me a small CLI tool to organize my Raw photos into a directory structure I want by reading the EXIF data and sorting the images accordingly. It's great for this stuff.
But anything bigger it seems to do useless crap. Creates data models that already exist in the project. Makes unrelated changes. Hallucinates API functions that don't exist. It's just not worth it to me to have to check its work. By the time I've done that, I could have written it myself, and writing the code is usually the most pleasurable part of the job to me.
I think the way I'm finding LLMs to be useful is that they are a brilliant interface to query with, but I have not yet seen any use cases I like where the output is saved, directly incorporated into work, or presented to another human that did not do the prompting.
Have you tried Opus? It's what got me past using LLMs only marginally. Standard disclaimers apply in that you need to know what it's good for and guide it well, but there's no doubt at this point it's a huge productivity boost, even if you have high standards - you just have to tell it what those standards are sometimes.
The one thing I've found AI is good at is parsing through the hundreds of ad ridden, barely usable websites for answers to my questions. I use the Duck Duck Go AI a lot to answer questions. I trust it about as far as I can throw the datacenter it resides in, but it's useful for quickly verifiable things. Especially stuff like syntax and command line options for various programs.
Nope, this only applies to a small percent of content, where a relatively small number of people needs access to it and the incentive to create derivative work based on it is low, or where there's a huge amount of content that's frequently changing (think airfares). But yes, they will protect it more.
For content that doesn't change frequently and is used by a lot of people it will be hard to control access to it or derivative works based on it.
I don't think you're considering the enshittification route here. I'm sure it will be: Ask ChatGPT a question -> "While I'm thinking, here's something from our sponsor which is tailored to your question" -> lame answer which requires you to ask another question. And on and on. While you're asking these questions, a profile of you is built and sold on the market.
It's one of those you get what you put in kind of deals.
If you spend a lot of time thinking about what you want, describing the inner workings, edge cases, architecture and library choices, and put that into a thoughtful markdown, then maybe after a couple of iterations you will get half decent code. It certainly makes a difference between that and a short "implement X" prompt.
But it makes one think - at that point (writing a good prompt that is basically a spec), you've basically solved the problem already. So LLM in this case is little more than a glorified electric typewriter. It types faster than you, but you did most of the thinking.
Right, and then after you do all the thinking and the specs, you have to read and understand and own every single line it generated. And speaking for myself, I am no where near as good at thinking through code I am reviewing as thinking through the code I am writing.
Other people will put up PRs full of code they don't understand. I'm not saying everyone who is reporting success with LLMs are doing that, but I hear it a lot. I call those people clowns, and I'd fire anyone who did that.
If it passes the unit tests I make it write and works for my sample manual cases I absolutely will not spend time reading the implementation details unless and until something comes up. Sometimes garbage makes its way into git but working code is better than no code and the mess can be cleaned up later. If you have correctness at the interface and function level you can get a lot done quickly. Technical debt is going to come out somewhere no matter what you do.
The trick is to not give a fuck. This works great in a lot of apps, which are useless to begin with. It may also be a reasonable strategy in an early-stage startup yet to achieve product-market fit, but your plan has to be to scrap it and rewrite it and we all know how that usually turns out.
This is an excellent point. Sure in an ideal world we should care very much about every line of code committed, but in the real world pushing garbage might be a valid compromise given things like crunch, sales pitches due tomorrow etc.
No, that's a much stronger statement. I'm not talking about ideals. I'm talking about running a business that is mature, growing and going to be around in five years. You could literally kill such a business running it on a pile of AI slop that becomes unmaintainable.
The bigger the task, the more messy it'll get. GPT5 can write a single UI component for me no problem. A new endpoint? If it's simple, no problem. The risk increases as the complexity of the task does.
I agree with the article but also believe LLM coding can boost my productivity and ability to write code over long stretches. Sure getting it to write a whole feature, high opportunity of risk. But getting it to build out a simple api with examples above and below it, piece of cake, takes a few seconds and would have taken me a few minutes.
It's just that being the dumbest thing we ever heard still doesn't stop some people from doing it anyway. And that goes for many kinds of LLM application.
I haven't had that experience, but I tend to keep my prompts very focused with a tightly limited scope. Put a different way, if I had a junior or mid level developer, and I wanted them to create a single-purpose class of 100-200 lines at most, that's how I write my prompts.
I think it has a lot to do with skill level. Lower skilled developers seem to feel it gives them a lot of benefit. Higher skilled developers just get frustrated looking at all the errors in produces.
Sounds like you are using it entirely wrong then...
Just yesterday I uploaded a few files of my code (each about 3000+ lines) into a gpt5 project and asked in assistance in changing a lot of database calls into a caching system, and it proceeded to create a full 500 line file with all the caching objects and functions I needed. Then we went section through section of the main 3000+ line file to change parts of the database queries into the cached version. [I didn't even really need to do this, it basically detected everything I would need changing at once and gave me most of it, but I wanted to do it in smaller chunks so I was sure what was going on]
Could I have done this without AI? Sure.. but this was basically like having a second pair of eyes and validating what I'm doing. And saving me a bunch of time so I'm not writing everything from scratch. I have the base template of what I need then I can improve it from there.
All the code it wrote was perfectly clean.. and this is not a one off, I've been using it daily for the last year for everything. It almost completely replaces my need to have a junior developer helping me.
How large is that code-base overall? Would you be able to let the LLM look at the entirety of it without it crapping out?
It definitely sounds nice to go and change a few queries, but did it also consider the potential impacts in other parts of the source or in adjacent running systems? The query itself here might not be the best example, but you get what I mean.
I know, I don't understand what problems people are having with getting usable code. Maybe the models don't work well with certain languages? Works great with C++. I've gotten thousands of lines of clean compiling on the first try and obviously correct code from ChatGPT, Gemini, and Claude.
I've been assuming the people who are having issues are junior devs, who don't know the vocabulary well enough yet to steer these things in the right direction. I wouldn't say I'm a prompt wizard, but I do understand context and the surface area of the things I'm asking the llm to do.
From my experience the further you get from the sort of stuff that easily accessible on Stack Overflow the worse it gets. I've had few problems having an AI write out some minor python scripts, but yield severely poorer results with Unreal C++ code and badly hallucinate nonsense if asked in general anything about Unreal architecture and API.
Does the Unreal API change a bit over versions? I've noticed when asking to do a simple telnet server in Rust it was hallucinating like crazy but when I went to the documentation it was clear the api was changing a lot from version to version. I don't think they do well with API churn. That's my hypothesis anyway.
You mean like it turned on Hibernate or it wrote some custom rolled in app cache layer?
I usually find these kinds of caching solutions to be extremely complicated (well the cache invalidating part) and I'm a bit curious what approach it took.
You mention it only updated a single file so I guess it's not using any updates to the session handling so either sticky sessions are not assumed or something else is going on. So then how do you invalidate the app level cache for a user across all machine instances? I have a lot of trauma from the old web days of people figuring this out so I'm really curious to hear about how this AI one shot it in a single file.
This is C# so basically just automatically detected that I had 4 object types I was working with that were being updated to the database that I want to keep in a concurrent dictionary type of cache. So it created the dictionaries for each object with the appropriate keys, created functions for each object type if I touch an object to get that one updated etc.
It created the function to load in the data, then the finalize where it writes to the DB what was touched and clears the cache.
Again- I'm not saying this is anything particularly fancy, but it did the general concept of what I wanted. Also this is all iterative; when it creates something I talk to it like a person to say "hey I want to actually load in all the data, even though we will only be writing what changed" and all that kind of stuff.
Also the bigger help wasn't really the creation of the cache, it was helping to make the changes and detect what needed to be modified.
End of the day even if I want to go a slightly different route of how it did the caching; it creates all the framework so I can simplify if needed.
A lot of times for me using this LLM approach is to get all the boilerplate out of the way.. sometimes just starting the process by yourself of something is daunting. I find this to be a great way to begin.
At least one CEO seems to get it. Anyone touting this idea of skipping junior talent in favor of AI is dooming their company in the long run. When your senior talent leaves to start their own companies, where will that leave you?
I’m not even sure AI is good for any engineer, let alone junior engineers. Software engineering at any level is a journey of discovery and learning. Any time I use it I can hear my algebra teacher telling me not to use a calculator or I won’t learn anything.
But overall I’m starting to feel like AI is simply the natural culmination of US economic policy for the last 45 years: short term gains for the top 1% at the expense of a healthy business and the economy in the long term for the rest of us. Jack Welch would be so proud.
> When your senior talent leaves to start their own companies, where will that leave you?
The CEO didn't express any concerns about "talent leaving". He is saying "keep the juniors" but he's implying "fire the seniors". This is in line with long standing industry trends and it's confirmed by the flowing quote from the OP:
>> [the junior replacement] notion led to the “dumbest thing I've ever heard” quote, followed by a justification that junior staff are “probably the least expensive employees you have” and also the most engaged with AI tools.
He is pushing for more of the same, viewing competence and skill as threats and liability to be "fixed". He's warning the industry to stay the course and keep the dumbing-down game moving as fast possible.
There are lots of personal projects that I have wanted to build for years but have pushed off because the “getting started cost” is too high, I get frustrated and annoyed and don’t get far before giving up. Being able to get the tedious crap out of the way lowers the barrier to entry and I can actually do the real project, and get it past some finish line.
Am I learning as much as I would had I powered through it without AI assistance? Probably not, but I am definitely learning more than I would if I had simply not finished (or even started) the project at all.
What was your previous approach? From what I've seen, a lot of people are very reluctant about picking a book or read through a documentation before they try stuff. And then they got exposed to "cryptic" error message and then throw the towel.
The biggest blocker for me would be that I would go through a "Getting Started" guide and that would go well until it doesn't. Either there would be an edge case that the guide didn't take into account or the guide would be out of date. Sometimes I would get an arcane error message that was a little difficult to parse.
There were also cases where the interesting part of what I'm working on (e.g. something with distributed computing) required a fair amount of stuff that I don't find interesting to get started (e.g. spinning up a Kafka and Zookeeper cluster), where I might have to spend hours screwing around with config files and make a bunch of mistakes before I get something more or less working.
If I was sufficiently interested, I would power through the issue, by either more thoroughly reading through documentation or searching through StackOverflow or going onto a project IRC, so it's not like I would never finish a project, but having a lower barrier of entry by being able to directly paste an error message or generate a working starter config helps a lot with getting past the initial hump, especially to get to the parts that I find more interesting.
In that case I’m not sure you really agree with this CEO, who is all-in on the idea of LLMs for coding, going so far as to proudly say 80% of engineers at AWS use it and that that number will only rise. Listen to the interview, you don’t even need ten minutes.
You also risk senior talent who stay but doesn't want to change or adopt, at least with any urgency. AI will accelerate that journey of discovery and learning, so juniors are going to learn super fast.
In the last few months we have worked with startups who have vibe coded themselves into an abyss. Either because they never made the correct hires in the first place or they let technical talent go. [1]
The thinking was that they could iterate faster, ship better code, and have an always on 10x engineer in the form of Claude code.
I've observed perfectly rational founders become addicted to the dopamine hit as they see Claude code output what looks like weeks or years of software engineering work.
It's overgenerous to allow anyone to believe AI can actually "think" or "reason" through complex problems. Perhaps we should be measuring time saved typing rather than cognition.
He wants educators to instead teach “how do you think and how do you decompose problems”
Ahmen! I attend this same church.
My favorite professor in engineering school always gave open book tests.
In the real world of work, everyone has full access to all the available data and information.
Very few jobs involve paying someone simply to look up data in a book or on the internet. What they will pay for is someone who can analyze, understand, reason and apply data and information in unique ways needed to solve problems.
Doing this is called "engineering". And this is what this professor taught.
In undergrad I took an abstract algebra class. It was very difficult and one of the things the teacher did was have us memorize proofs. In fact, all of his tests were the same format: reproduce a well-known proof from memory, and then complete a novel proof. At first I was aghast at this rote memorization - I maybe even found it offensive. But an amazing thing happened - I realized that it was impossible to memorize a proof without understanding it! Moreover, producing the novel proofs required the same kinds of "components" and now because they were "installed" in my brain I could use them more intuitively. (Looking back I'd say it enabled an efficient search of a tree of sequences of steps).
Memorization is not a panacea. I never found memorizing l33t code problems to be edifying. I think it's because those kinds of tight, self-referential, clever programs are far removed from the activity of writing applications. Most working programmers do not run into a novel algorithm problem but once or twice a career. Application programming has more the flavor of a human-mediated graph-traversal, where the human has access to a node's local state and they improvise movement and mutation using only that local state plus some rapidly decaying stack. That is, there is no well-defined sequence for any given real-world problem, only heuristics.
Memorizing is a super power / skill. I work in a ridiculously complex environment and have to learn and know so much. Memorizing and spaced repetition are like little islands my brain can start building bridges between. I used to think memorizing was anti-first principles, but it is just good. Our brains can memorize so much if we make them. And then we can connect and pattern matching using higher order thinking.
Recognizing the patterns and applying patterned solutions is where I see success in my niche of healthcare interoperability. So much of my time is spent watching people do things,process and how they use data. It's amazing how much people remember to do their job, but me coming in and be able to bridge the doctor and the lab to share data easier is like Im an alchemist. It's really not a problem I've been able to see ai solve without suggesting solutions that are too simple or too costly and in that goldilocks zone everyone will be happy with
What's even better about memorization is that you have an objective method to test your own understanding. It is so easy to believe you understand something when you don't! But, at least with math, I think if you can reproduce the proof from memory you can be very confident that you aren't deluding yourself.
Hmmm... It's the other way around for me. I find it hard to memorise things I don't actually understand.
I remember being given a proof of why RSA encryption is secure. All the other students just regurgitated it. It made superficial sense I guess.
However, I could not understand the proof and felt quite stupid. Eventually I went to my professor for help. He admitted the proof he had given was incomplete (and showed me why it still worked). He also said he hadn't expected anyone to notice it wasn't a complete proof.
During my elementary school years, there was a teacher who told me that I didn't need to memorize it as long as I understand them. I taught he was the coolest guy ever.
Only when I got late twenties, I realized how wrong he was. Memorization and understanding go hand in hand, but if one of them has to come first than it's memorization. He probably said that because that was what kids (who were forced to do rote memorization) wanted to hear.
You could argue this is just moving the memorization to meta-facts, but I found all throughout school that if you understand some slightly higher level key thing, memorization at the level you're supposed to be working in becomes at best a slight shortcut for some things. You can derive it all on the fly.
Sort of like how most of the trigonometric identities that kids are made to memorize fall out immediately from e^iθ = cosθ+isinθ (could be taken as the definitions of cos,sin), e^ae^b=e^(a+b) (a fact they knew before learning trig), and a little bit of basic algebraic fiddling.
Or like how inverse Fourier transforms are just the obvious extension of the idea behind writing a 2-d vector as a sum of its x and y projections. If you get the 2d thing, accept that it works the exact same in n-d (including n infinite), accept integrals are just generalized sums, and functions are vectors, and I guess remember that e^iwt are the basis you want, you can reason through what the formula must be immediately.
Probably. I hated memorization when I was a student too, because it was boring. But as soon as I did some teaching, my attitude changed to, "Just memorize it, it'll make your life so much easier." It's rough watching kids try to multiply when they don't have their times tables memorized, or translate a language when they haven't memorized the vocabulary words in the lesson so they have to look up each one.
There's things that you need to know (2*2 = 4) and there are things that you need to understand (multiplication rules). Both can happen with practice, but they're not that related.
Memorization is more like a shortcut. You don't need to go through the problem solving process to know the result. But with understanding, you master the heuristic factors needed to know when to take the shortcut and when to go through the problem solving route.
The Dreyfus Skill Model [0] is a good explanation. Novice typically have to memorize, then as they master the subject, their decision making becomes more heuristic based.
LLMs don't do well with heuristics, and by the times you've nailed down all the problems data, you could have been done. What they excels at is memorization, but all the formulaic stuff have been extracted into frameworks and libraries for the most popular languages.
I never memorized multiplication tables and was always one of those "good in math" kids. An attempt to memorize that stuff ended with me confusing results and being unable to guess when I did something wrong. Knowing "tricks" and understanding how multiplication works makes life easier.
> "Just memorize it, it'll make your life so much easier."
That is because you evaluate cost of memorization to 0, because someone else is paying it. And you evaluate the cost of making mistakes due to constantly forgetting and being unable to correct to 0, because simply the kid gets blamed for not having perfect memory.
> or translate a language when they haven't memorized the vocabulary words in the lesson so they have to look up each one
Teaching language by having people translate a lot is an outdated pedagogy - it simply did not produced people capable to understand and produce the language. If the kids are translating sentences word by word, there was something going on wrongly before.
I think we memorize the understanding. For me it also works better understanding how something works than memoryzing results. I remember in high school, in maths trigonometrics, there were a list of 20 something formulas derived from a single one. Everyboby was memorizing the whole list of formulas; i just had to memorize a simple formula and the underdtanding of how to derive the others from the fundamental one on the fly.
Probably the best example of this I can think of (for me at least) from mathematics is calculating combinations. I have it burned into my memory that (n choose r) = (n permute r) / (r permute r), and (n permute r) = n! / (n - r)!
Can I derive these from first principles? Sure, but after not seeing it for years, it might take me 10+ minutes to think through everything and correct any mistakes I make in the derivation.
But if I start with the formula? Takes me 5 seconds to sanity check the combination formula, and maybe 20 to sanity check the permutation formula. Just reading it to myself in English slowly is enough because the justification kind of just falls right out of the formula and definition.
So, yeah, they go hand in hand. You want to understand it but you sure as heck want to memorize the important stuff instead of relying on your ability to prove everything from ZFC...
As with most things, it depends. If you truly do understand something, then you can derive a required result from first principles. _Given sufficient time_. Often in an exam situation you are time-constrained, and having memorized a shortcut cut be beneficial. Not to mention retaining is much easier when you understand the topic, so memorization becomes easier.
Ish? I never ever memorized the multiplication tables. To this day, I don't think I know them fully. I still did quite well in math by knowing how to quiz the various equations. Not just know them, but how to ask questions about moving terms and such.
You don't need to memorize to understand. You can rederive it every time.
You need to memorize it to use it subconsciously while solving more complex problems. Other ways you won't fit more complex solutions into your working memory,vso whole classes of problems will be too hard for you.
It is waaaay easier to remember when you understand. The professor had it exactly right - if you learn to understand, you frequently end up remembering. But, memorization does not lead to understanding at all.
My controversial education hot take: Pointless rote memorization is bad and frustrating, but early education could use more directed memorization.
As you discovered: A properly structured memorization of carefully selected real world material forces you to come up with tricks and techniques to remember things. With structured information (proofs in your case) you start learning that the most efficient way to memorize is to understand, which then reduces the memorization problem into one of categorizing the proof and understanding the logical steps to get from one step to another. In doing so, you are forced to learn and understand the material.
Another controversial take (for HN, anyway) is that this is what happens when programmers study LeetCode. There’s a meme that the way to interview prep is to “memorize LeetCode”. You can tell who hasn’t done much LeetCode interviewing if they think memorizing a lot of problems is a viable way to pass interviews. People who attempt this discover that there are far too many questions to memorize and the best jobs have already written their own questions that aren’t out of LeetCode. Even if you do get a direct LeetCode problem in an interview, a good interview will expect you to explain your logic, describe how you arrived at the solution, and might introduce a change if they suspect you’re regurgitating memorized answers.
Instead, the strategy that actually works is to learn the categories of LeetCode style questions, understand the much smaller number of algorithms, and learn how to apply them to new problems. It’s far easier to memorize the dozen or so patterns used in LeetCode problems (binary search, two pointers, greedy, backtracking, and so on) and then learn how to apply those. By practicing you’re not memorizing the specific problems, you’re teaching yourself how to apply algorithms.
Side note: I’m not advocating for or against LeetCode, I’m trying to explain a viable strategy for today’s interview format.
Exactly. I agree with the leetcode part. A lot of problems in the world are composite of simpler smaller problems. Leetcode should teach you the basic patterns and how to combine them to solve real world problems. How will you ever solve a real world problem without knowing a few algorithms beforehand. For example, my brother was talking about how a Roomba would map a room. He was imagining 0 to represent free space and 1 as inaccessible points. This quickly reminded me of Number of Islands problem from leetcode. Yeah, there might be a lot of changes required to that problem but one could simple represent it as two problems.
1. Represent different objects in the room as some form of machine understandable form in a matrix
2. Find the number of Islands or find the Islands themselves.
Memorization of, like, multiplication tables gives us a poor view of the more interesting type of memorization. Remembering types of problems we’ve seen. Remembering landmarks and paths, vs just remembering what’s in every cell of a big grid.
> Memorization of, like, multiplication tables gives us a poor view of the more interesting type of memorization.
Memorizing multiplication tables is the first place many children encounter this strategy: The teacher shows you that you could try to memorize all of the combinations, or you could start learning some of the patterns and techniques. When multiplying by 5 the answer will end in 0 or 5. When multiplying by 2 the answer will be an even number, and so on.
I think there may have been a miscommunication somewhere on the chain of Mathematicians-Teachers-Students if that was the plan, when I was in elementary school.
Anecdotally (I only worked with math students as a tutor for a couple years), that math requires a lot of the boring type of memorization seems to be a really widespread misunderstanding.
My abstract algebra class had it exactly backwards. It started with a lot of needless formalism culminating in galois theory. This was boring to most students as they had no clue why the formalism was invented in the first place.
Instead, I wished it showed how the sausage was actually made in the original writings of galois [1]. This would have been far more interesting to students, as it showed the struggles that went into making the product - not to mention the colorful personality of the founder.
The history of how concepts were invented for the problems faced is far more motivating to students to build a mental model than canned capsules of knowledge.
> This was boring to most students as they had no clue why the formalism was invented in the first place.
> The history of how concepts were invented for the problems faced is far more motivating to students to build a mental model than canned capsules of knowledge.
That's something I really like about 3blue1brown, and he says it straight up [0]:
> My goal is for you to come away feeling like you could have invented calculus yourself. That is, cover all those core ideas, but in a way that makes clear where they actually come from, and what they really mean, using an all-around visual approach.
Interesting I had the same problem and suffered in grades back in school simply because I couldn't memorize much without understanding. However, I seemed to be the only one because every single other student, including those with top grades, were happy to memorize and regurgitate. I wonder how they're doing now.
I would wager some folks can memorize without understanding? I do think memorization is underrated, though.
There is also something to the practice of reproducing something. I always took this as a form of "machine learning" for us. Just as you get better at juggling by actually juggling, you get better at thinking about math by thinking about math.
Depends on the subject - I can remember multiple subjects where the teacher would give you a formula to memorise without explaining why or where it came from. You had to take it as an axiom. The teachers also didn't say - hey, if you want to know why did we arrive to this, have a read here, no, it was just given.
Ofc you could also say that's for the student to find out, but I've had other things on my mind
Fortunately that was not my experience in abstract algebra. The tests and homework were novel proofs that we hadn't seen in class. It was one of my favorite classes / subjects. Someone did tell me in college that they did the memorization thing in German Universities.
Code-wise, I spent a lot of time in college reading other people's code. But no memorization. I remember David Betz advsys, Tim Budd's "Little Smalltalk", and Matt Dillon's "DME Editor" and C compiler.
It is What you memorize that is important, you can't have a good discussion about a topic if you don't have the facts and logic of the topic in memory. On the other hand using memory to paper over bad design instead of simplifying or properly modularizing it, leads to that 'the worst code I have seen is code I wrote six months ago' feeling.
Is it the memorisation that had the desired effect or the having to come up with the novel proofs? Many schools seem to do the memorising part, but not the creating part.
Your comment about memorizing as part of understanding makes a lot of sense to me, especially as one possible technique to get get unstuck in grasping a concept.
If it doesn’t work for you on l33t code problems, what techniques are you finding more effective in that case?
I was part of an ACM programming team in college. We would review classes of problems based on the type of solution necessary, and learn those techniques for solving them. We were permitted a notebook, and ours was full of the general outline of each of these classes and techniques. Along with specific examples of the more common algorithms we might encounter.
As a concrete example, there is a class of problems that are well served by dynamic programming. So we would review specific examples like Dijkstra's algorithm for shortest path. Or Wagner–Fischer algorithm for Levenshtein-style string editing. But we would also learn, often via these concrete examples, of how to classify and structure a problem into a dynamic programming solution.
I have no idea if this is what is meant by "l33t code solutions", but I thought it would be a helpful response anyway. But the bottom line is that these are not common in industry, because hard computer science is not necessary for typical business problems. The same way you don't require material sciences advancements to build a typical house. Instead it flows the other way, where advancements in materials sciences will trickle down to changing what the typical house build looks like.
>If it doesn’t work for you on l33t code problems, what techniques are you finding more effective in that case?
Memorization of l33t code DOES work well as prep for l33t code tests. I just don't think l33t code has much to do with application programming. I've long felt that "computer science" is physics for computers, low on the abstraction ladder, and there are missing labels for the higher complexity subjects built on it. Imagine if all physical sciences were called "physics" and so in order to get a job as a biologist you should expect to be asked questions about the Schroedinger equation and the standard model. We desperately need "application engineering" to be a distinct subject taught at the university level.
That's a real major that's been around for a couple of decades which focuses on software development (testing, version control, design patterns) with less focus on the more theoretical parts of computer science? There are even specialties within the Software Engineering major that focus specifically on databases or embedded systems.
What I understand from the GP is that memorizing l33t code won't help you learn anything useful. Not that understanding the solutions won't help you memorize them.
> But an amazing thing happened - I realized that it was impossible to memorize a proof without understanding it!
This may be true of mathematical proofs, but it surely must not be true in general. Memorizing long strings of digits of pi probably isn’t much easier if you understand geometry. Memorizing famous speeches probably isn’t much easier if you understand the historical context.
> Memorizing famous speeches probably isn’t much easier if you understand the historical context.
Not commenting on the merits of critical thinking vs memorization either way, but I think it would be meaningfully easier to memorize famous speeches if you understand the historical context.
It's funny, because I had the exact opposite experience with abstract algebra.
The professor explained things, we did proofs in class, we had problem sets, and then he gave us open-book semi-open-professor take-home exams that took us most of a week to do.
Proof classes were mostly fine. Boring, sometimes ridiculously shit[0], but mostly fine. Being told we have a week for this exam that will kick our ass was significantly better for synthesizing things we'd learned. I used the proofs we had. I used sections of the textbook we hadn't covered. I traded some points on the exam for hints. And it was significantly more engaging than any other class' exams.
[0] Coming up with novel things to prove that don't require some unrelated leap of intuition that only one student gets is really hard to do. Damn you Dr. B, needing to figure out that you have to define a third equation h(x) as (f(x) - g(x))/(f(x) + g(x)) as the first step of a proof isn't reasonable in a 60 minute exam.
Mathematics pedagogy today is in a pretty sorrowful state due to bad actors and willful blindness at all levels that require public trust.
A dominant majority in public schools starting late 1970s seems to follow the "Lying to Children" approach which is often mistakenly recognized as by-rote teaching but are based in Paulo Freire's works that are in turn based on Mao's torture discoveries from the 1950s.
This approach contrary to classical approaches leverages torturous process which seems to be purposefully built to fracture and weed out the intelligent individual from useful fields, imposing sufficient thresholds of stress to impose PTSD or psychosis, selecting for and filtering in favor of those who can flexibly/willfully blind/corrupt themselves.
Such sequences include Algebra->Geometry->Trigonometry where gimmicks in undisclosed changes to grading cause circular trauma loops with the abandonment of Math-dependent careers thereafter, similar structures are also found in Uni, for Economics, Business, and Physics which utilize similar fail-scenarios burning bridges where you can't go back when the failure lagged from the first sequence, and you passed the second unrelated sequence. No help occurs, inducing confusion and frustration to PTSD levels, before the teacher offers the Alice in Wonderland Technique, "If you aren't able to do these things, perhaps you shouldn't go into a field that uses it". (ref Kubark Report, Declassified CIA Manual)
Have you been able to discern whether these "patterns" as you've called them aren't just the practical reversion to the classical approach (Trivium/Quadrivium)? Also known as the first-principles approach after all the filtering has been done.
To compare: Classical approaches start with nothing but a useful real system and observations which don't entrench false assumptions as truth, which are then reduced to components and relationships to form a model. The model is then checked for accuracy against current data to separate truth from false in those relationships/assertions in an iterative process with the end goal being to predict future events in similar systems accurately. The approach uses both a priori and a posteriori components to reasoning.
Lying to Children reverses and bastardizes this process. It starts with a single useless system which contains equal parts true and false principles (as misleading assumptions) which are tested and must be learned to competency (growing those neurons close together). Upon the next iteration one must unlearn the false parts while relearning the true parts (but we can't really unlearn, we can only strengthen or weaken) which in turn creates inconsistent mental states imposing stress (torture). This is repeated in an ongoing basis often circular in nature (structuring), and leveraging psychological blindspots (clustering), with several purposefully structured failings (elements) to gatekeep math through torturous process which is the basis for science and other risky subject matter. As the student progresses towards mastery (gnosis), the systems become increasingly more useful. One must repeatedly struggle in their sessions to learn, with the basis being if you aren't struggling you aren't learning. This mostly uses a faux a priori reasoning without properties of metaphysical objectivity (tied to objective measure, at least not until the very end).
If you don't recognize this, an example would be the electrical water pipe pressure analogy. Diffusion of charge in-like materials, with Intensity (Current) towards the outermost layer was the first-principled approach pre-1978 (I=V/R). The Water Analogy fails when the naive student tries to relate the behavior to pressure equations that ends up being contradictory at points in the system in a number of places introducing stumbling blocks that must be unlearned.
Torture being the purposefully directed imposition of psychological stress beyond a individuals capacity to cope towards physiological stages of heightened suggestability and mental breakdown (where rational thought is reduced or non-existent in the intelligent).
It is often recognized by its characteristic subgroups of Elements (cognitive dissonance, a lack of agency to remove oneself and coercion/compulsion with real or perceived loss or the threat thereof), Structuring (circular patterns of strictness followed by leniency in a loop, fractionation), and Clustering (psychological blindspots).
Wait, the electrical pipe water analogy is actually a very good one and it's quite difficult to find edge cases where it breaks down in a way that would confuse a student. There are some (for example, there's no electrical equivalent of Reynold's number or turbulence, and flow resistance varies differently with pipe diameter than wire diameter, and no good equivalent for Faraday's law) but I don't think these are likely to cause confusion. It even captures nuance like inductance, capacitance, and transmission line behaviour.
As I recall, my systems dynamics textbook even explicitly drew parallels between different domains like electricity and hydrodynamics. You're right that the counterparts aren't generally perfect especially at the edges but the analogies are often pretty good.
Intuitively it fails in making an equivalence to area which is an unrelated dimensional unit, as two lengths multiplied together equaling resistance, as well as the skin-effect related to Intensity/Current which is why insulation/isolation of wires are incredibly important.
The classical approach used charge diffusion iirc, and you can find classical examples of this in Oliver Heaviside's published works (archive.org iirc). He's the one that simplified Maxwell's 20+ equations down to the small number we use today.
> Lying to Children reverses and bastardizes this process. It starts with a single useless system which contains equal parts true and false principles (as misleading assumptions) which are tested and must be learned to competency (growing those neurons close together).
I'd love an example too, and an example of the classical system that this replaced. I'm willing to believe the worst of the school system, but I'd like to understand why.
The classical system was described, but you can find it in various historic works based on what's commonly referred to today as the Trivium and Quadrivium based curricula.
Off the top of my head, the former includes reasoning under dialectical (priori and later posteriori parts under the quadrivium).
Its a bit much to explain it in detail in a post like this but you should be able to find sound resources with what I've provided.
It largely goes back to how philosophy was taught; all the way back to Socrates/Plato/Aristotle, up through Descartes, Locke (barely, though he's more famous for social contract theory), and more modern scientists/scientific method.
The way math is taught today, you basically get to throw out almost everything you were taught at various stages, and relearn it anew on a different foundation, somehow fitting the fractured pieces back together towards learning the true foundations, which would be much easier at the start and building on top of that instead of the constant interference.
You don't really end up understanding math intuitively nor its deep connections to logic (dialectics, trivium), until you hit Abstract Algebra.
Not OP, and it was a couple decades ago, but I certainly remember professors and teachers saying things like "this isn't really how X works, but we will use the approximation for now in order to teach you this other thing". That is if you were lucky, most just taught you the wrong (or incomplete) formula.
I think there is validity to the approach but sciences would be much, much improved if taught more like history lessons. Here is how we used to think about gravity, here's the formula and it kind of worked, except... Here is planetary orbits that we used to use when we assumed they had to be circles. Here's how the data looked and here's how they accounted for it...
This would accomplish two goals - learning the wrong way for immediate use (build on sand) and building an innate understanding of how science actually progresses. Too little focus is on how we always create magic numbers and vague concepts (dark matter, for instance) to account for structural problems we have no good answer for.
Being able to "sniff the fudge" would be a super power when deciding what to write a PhD on, for instance. How much better would science be if everyone strengthened this muscle throughout their educatuon?
I included the water pipe analogy for electric theory, that is one specific example.
Also, In Algebra I've seen a flawed version of mathematical operations being taught that breaks down with negative numbers under multiplication (when the correct way is closed over multiplication). The tests were supposedly randomized (but seemed to target low-income demographics). The process is nearly identical, but the answers ultimately not correct. The teachers graded on the work to the exclusion of the correct answer. So long as you showed the correct process expected in Algebra you passed without getting the right answer. Geometry was distinct and unrelated, and by Trigonometry the class required correct process and answer. You don't find out there is a problem until Trigonometry, and the teacher either doesn't know where the person is failing comprehension, or isn't paid to reteach a class they aren't paid for but you can't go back.
I've seen and heard horror stories of students where they'd failed Trig 7+ times at the college level, and wouldn't have progressed if not for a devoted teacher helping them after-hours (basically correcting and reteaching Algebra). These kids literally would break out in a cold PTSD sweat just hearing the associated words related to math.
I did some tutoring in a non-engineering graduate masters program and some folks were just lost. Simple things like what a graph is or how to solve an equation. I really did try but it's sort of hard to teach fairly easy high school algebra (with maybe some really simple derivatives to find maxima and minima) in grad school.
It's the core problem facing the hiring practices in this field. Any truly competent developer is a generalist at heart. There is value to be had in expertise, but unless you're dealing with a decade(s) old hellscape of legacy code or are pushing the very limits of what is possible, you don't need experts. You'd almost certainly be better off with someone who has experience with the tools you don't use, providing a fresh look and cover for weaknesses your current staff has.
A regular old competent developer can quickly pick up whatever stack is used. After all, they have to; Every company is their own bespoke mess of technologies. The idea that you can just slap "15 years of React experience" on a job ad and that the unicorn you get will be day-1 maximally productive is ludicrous. There is always an onboarding time.
But employers in this field don't "get" that. For regular companies they're infested by managers imported from non-engineering fields, who treat software like it's the assembly line for baking tins or toilet paper. Startups, who already have fewer resources to train people with, are obsessed with velocity and shitting out an MVP ASAP so they can go collect the next funding round. Big Tech is better about this, but has it's own problems going on and it seems that the days of Big Tech being the big training houses is also over.
It's not even a purely collective problem. Recruitment is so expensive, but all the money spent chasing unicorns & the opportunity costs of being understaffed just get handwaved. Rather spend $500,000 on the hunt than $50,000 on training someone into the role.
And speaking of collective problems. This is a good example of how this field suffers from having no professional associations that can stop employers from sinking the field with their tragedies of the commons. (Who knows, maybe unions will get more traction now that people are being laid off & replaced with outsourced workers for no legitimate business reason.)
I can’t think of another career where management continuously does not understand the realities of how something gets built. Software best practices are on their face orthogonal to how all other parts of a business operate.
How does marketing operate? In a waterfall like model. How does finance operate? In a waterfall like model. How does product operate? Well you can see how this is going.
Then you get to software and it’s 2 week sprints, test driven development etc. and it decidedly works best not on a waterfall model, but shipping in increments.
Yet the rest of the business does not work this way, it’s the same old top down model as the rest.
This I think is why so few companies or even managers / executives “get it”
> can’t think of another career where management continuously does not understand the realities of how something gets built
All engineering. Also all government and a striking amount of finance.
Actually, this might be a hallmark of any specialist field. Specialists interface with outsiders through a management layer necessarily less competent at the specialty than they are. (Since they’re devoting time and energy to non-specialty tasks.)
While product often does operate in a waterfall model, I think this is the wrong mindset. Good product management should adopt a lot of the same principles as software development. Form a testable hypothesis, work to get it into production and begin gathering data, then based on your findings determine what the next steps are and whether to adjust the implementation, consider the problem solved or try a different approach.
> I can’t think of another career where management continuously does not understand the realities of how something gets built.
This is in part a consequence of how young our field is.
The other comment pointing out other engineering is right here. The difference is that fields like Civil Engineering are millenia old. We know that Egyptian civil engineering was advanced and shockingly modern even 4.5 millenia ago. We've basically never stopped having qualified civil engineers around who could manage large civil engineering projects & companies.
Software Development in it's modern forms has it's start still in living memory. There simply weren't people to manage the young early software development firms as they grew, so management got imported from other industries.
And to say something controversial: Other engineering has another major reason why it's usually better understood. They're held to account when they kill people.
If you're engineering a building or most other things, you must meet safety standards. Where possible you are forced to prove you meet them. E.g. Cars.
You don't get to go "Well cars don't kill people, people kill people. If someone in our cars die when they're hit by a drunk driver, that's not our problem that's the drunkard's fault." No. Your car has to hold up to a certain level of crash safety, even if it's someone else who causes the accent, your engineering work damn better hold up.
In software, we just do not do this. The very notion of "Software kills people" is controversial. Treated as a joke, "of course it can't kill people, what are you on about?". Say, you neglect on your application's security. There's an exploit, a data breach, you leak your users' GPS location. A stalker uses the data to find and kill their victim.
In our field, the popular response is to go "Well we didn't kill the victim, the stalker did. It's not our problem.". This is on some level true; 'Twas the drunk driver who caused the car crash, not the car company. But that doesn't justify the car company selling unsafe cars, why should it justify us selling unsafe software? It may be but a single drop of blood, but it's still blood on our hands as well.
As it stands, we are fortunate enough that there haven't been incidents big enough to kill so many people that governments take action to forcibly change this mindset. It would be wise that Software Development takes up this accountability on it's own accord to prevent such a disaster.
> Rather spend $500,000 on the hunt than $50,000 on training someone into the role.
Capex vs opex, that's the fundamental problem at heart. It "looks better on the numbers" to have recruiting costs than to have to set aside a senior developer plus paying the junior for a few months. That is why everyone and their dog only wants to hire seniors, because they have the skillset and experience that you can sit their ass in front of any random semi fossil project and they'll figure it out on their own.
If the stonk analysts would go and actually dive deep into the numbers to look at hiring side costs (like headhunter expenses, employee retention and the likes), you'd see a course change pretty fast... but this kind of in-depth analysis, that's only being done by a fair few short-sellers who focus on struggling companies and not big tech.
In the end, it's a "tragedy of the commons" scenario. It's fine if a few companies do that, it's fine if a lot of companies do that... but when no one wants to train juniors any more (because they immediately get poached by the big ones), suddenly society as a whole has a real and massive problem.
Our societies are driven into a concrete wall at full speed by the financialization of every tiny aspect of our lives. All that matters these days are the gods of the stonk market - screw the economy, screw the environment, screw labor laws, all that matters is appearing "numbers go up" on the next quarterly.
> but when no one wants to train juniors any more (because they immediately get poached by the big ones)
Can we stop pretending that we don't know how to solve this problem? If you hire juniors at $X/year, but they keep getting poached after 2-3 years because now they can get $X*1.5/year (or more!), then maybe you should start promoting and giving raises to them after they've gotten a couple years experience.
Seriously, this is not a hard problem to solve. If the junior has proven themselves, give them the raise they deserve instead of being all Surprised Pikachu when another company is willing to pay them what they've proven themselves worthy of.
> Our societies are driven into a concrete wall at full speed by the financialization of every tiny aspect of our lives. All that matters these days are the gods of the stonk market - screw the economy, screw the environment, screw labor laws, all that matters is appearing "numbers go up" on the next quarterly.
I have been in the various nooks and crannies of the Internet/Software dev industry my whole career (i'm 49). I can't think of any time when the stock market didn't drive software innovation. It's always been either invent something -> go public -> exit or invent something -> increase stock price of existing public corp
> It's always been either invent something -> go public -> exit or invent something -> increase stock price of existing public corp
Yes, but today more and more is invent something -> achieve dominance -> get bought up by an even larger megacorp. That drives the enshittification circle.
That's part of the problem, but I also notice the new hiring managers are incentivized to hire (or replace) employees to make their mark on the company. They then advocate for "their guys" the ones they recruited over the incumbents that are the unwilling dinosaurs in their eyes.
>For regular companies they're infested by managers imported from non-engineering fields
Someone's cousin, lets leave it at that, someones damn cousin or close friend, or anyone else with merely a pulse.
I've had interviews where the company had just been turned over from people that mattered, and you. could. tell.
One couldn't even tell me why the project I needed to do for them ::rolleyes::, their own code boilerplate(which they said would run), would have runtime issues and I needed to self debug it to even get it to a starting point.
Its like, Manager: Oh heres this non-tangential thing that they tell me you need to complete before I can consider you for the positon....
Me: Oh can I ask you anything about it?....
Manager: No
Isn't that happening already? Half the usual CS curriculum is either math (analysis, linear algebra, numerical methods) or math in anything but name (computability theory, complexity theory). There's a lot of very legitimate criticism of academia, but most of the times someone goes "academia is stupid, we should do X" it turns out X is either:
- something we've been doing since forever
- the latest trend that can be picked up just-in-time if you'll ever need it
I've worked in education in some form or another for my entire career. When I was in teacher education in college . . . some number of decades ago . . . the number one topic of conversation and topic that most of my classes were based around was how to teach critical thinking, effective reasoning, and problem solving. Methods classes were almost exclusively based on those three things.
Times have not changed. This is still the focus of teacher prep programs.
Parent comment is literally praising an experience they had in higher education, but your only takeaway is that it must be facile ridicule of academia.
In CS, it's because it came out of math departments in many cases and often didn't even really include a lot of programming because there really wasn't much to program.
Right but a looot of the criticism online is based on assumptions (either personal or inherited from other commenters) that haven’t been updated since 2006.
Well, at more elite schools at least, the general assumption is that programming is mostly something you pick up on your own. It's not CS. Some folks will disagree of course but I think that's the reality. I took an MIT Intro to Algorithms/CS MOOC course a few years back out of curiosity and there was a Python book associated with the course but you were mostly on your own with it.
When I was in college the philosophy program had the marketing slogan: “Thinking of a major? Major in thinking”.
Now as a hiring manager I’ll say I regularly find that those who’ve had humanities experience are way more capable and the hard parts of analysis and understanding. Of course I’m biased as a dual cs/philosophy major but it’s very rare I’m looking for someone who can just write a lot of code. Especially juniors as analytical thinking is way harder to teach than how to program.
> Now as a hiring manager I’ll say I regularly find that those who’ve had humanities experience are way more capable and the hard parts of analysis and understanding.
The humanities, especially the classic texts, cover human interaction and communication in a very compact form. My favorite sources are the Bible, Cicero, and Machiavelli. For example Machiavelli says if you do bad things to people do them at once, while good things you should spread out over time. This is common sense. Once you catch the flavor of his thinking it's pretty easy to work other situations out for yourself, in the same why that good engineering classes teach you how to decompose and solve technical problems.
This is also why I went into the Philosophy major - knowing how to learn and how to understand is incredibly valuable.
Unfortunately in my experience, many, many people do not see it that way. It's very common for folks to think of philosophy as "not useful / not practical".
Many people hear the word "philosophy" and mentally picture "two dudes on a couch recording a silly podcast", and not "investigative knowledge and in-depth context-sensitive learning, applied to a non-trivial problem".
It came up constantly in my early career, trying to explain to folks, "no, I actually can produce good working software and am reasonably good at it, please don't hyper-focus on the philosophy major, I promise I won't quote Scanlon to you all day."
Maybe because phylosophy focuses on weird questions (to be or not to be) and weird personas. If it was advertised as more grounded thing, the views would be different.
The way you are perceived by others dependa on your behaviour. If you wamt to be perceived differently, adjust your behaviour, don't demand others to change. They won't.
How people see it is based on the probability of any philosophy major producing good working software, not you being able to produce good working software.
yes, sometimes you need people who can grasp the tech and talk to managers. They might be intermediaries.
But don't ignore the nerdy guys who have been living deeply in a tech ecosystem all their lives. The ones who don't dabble in everything. (the wozniaks)
The #1 problem in almost all workplaces is communication related. In almost all jobs I've had in 25-30 years, finding out what needs to be done and what is broken -- is much harder than actually doing it.
We have these sprint planning meetings and the like where we throw estimates on the time some task will take but the reality is for most tasks it's maybe a couple dozen lines of actual code. The rest is all what I'd call "social engineering" and figuring out what actually needs to be done, and testing.
Meanwhile upper management is running around freaking out because they can't find enough talent with X years of Y [language/framework] experience, imagining that this is the wizard power they need.
The hardest problem at most shops is getting business domain knowledge, not technical knowledge. Or at least creating a pipeline between the people with the business knowledge and the technical knowledge that functions.
Anyways, yes I have 3/4 a PHIL major and it actually has served me well. My only regret is not finishing it. But once I started making tech industry cash it was basically impossible for me to return to school. I've met a few other people over the years like me, who dropped out in the 90s .com boom and then never went back.
Yea this is why I’m generally not that impressed by LLMs. They still force you to do the communication which is the hard part. Programming languages are inherently a solve for communicating complex steps. Programming in English isn’t actually that much of a help you just have to reinvent how to be explicit
I find Claude code unexpectedly good at analysis. With a healthy dose of skepticism. It is actually really good at reading logs and corelating events for example.
On a related note, the most accomplished people I've met didn't have degrees in the fields where they excelled and won awards. They were all philosophy majors.
Teaching people to think is perhaps the world's most under-rated skill.
My otherwise ordinary school applied the mentality that students must "Learn to learn", and that mix of skills and mindset has never stopped helping me.
Well, yes but the other 90%+ just need to get a job out of college to support their addiction to food and shelter not to be a “better citizen of the world” unless they have parents to subsidize their livelihood either through direct transfers of money or by letting them stay at home.
I told both of my (step)sons that I would only help them pay for college or trade school - their choice - if they were getting a degree in something “useful”. Not philosophy, not Ancient Chinese Art History etc.
I also told them that they would have to get loans in their own names and I would help them pay off the loans once they graduated and started working gainfully.
A professor in my very first semester called "crazy finger syndrome" the attempts to go straight to the code without decomposing the problem from a business or user perspective. It was a long time ago. It was a CS curriculum
I miss her jokes against anxious nerds that just wanted to code :(
Don't forget the rise of boot camps where some educators are not always aligned with some sort of higher ethical standards.
> "crazy finger syndrome" - the attempts to go straight to the code without decomposing the problem from a business or user perspective
Years ago I started on a new team as a senior dev, and did weeks of pair programming with a more junior dev to intro me to the codebase. His approach was maddening; I called it "spray and pray" development. He would type out lines or paragraphs of the first thing that came to mind just after sitting down and opening an editor. I'd try to talk him into actually taking even a few minutes to think about the problem first, but it never took hold. He'd be furiously typing, while I would come up with a working solution without touching a keyboard, usually with a whiteboard or notebook, but we'd have to try his first. This was c++/trading, so the type-compile-debug cycle could be 10's of minutes. I kept relaying this to my supervisor, but after a few months of this he was let go.
I make a point to solve my more difficult problems with pen and paper drawings and/or narrative text before I touch the PC. The computer is an incredibly distracting medium to work with if you are not operating under clear direction. Time spent on this forum is a perfect example.
Speaking with a relative who is a doctor recently it’s interesting how much each of our jobs are “troubleshooting”.
Coding, doctors, plumber… different information, often similar skill sets.
I worked a job doing tech support for some enterprise level networking equipment. It was the late 1990s and we were desperate for warm bodies. Hired a former truck driver who just so happened to do a lot of woodworking and other things.
Memorization and closed book tests are important for some areas. When seconds are counting the ER doctor cannot go look up how to treat a heart attack. That doctor also needs to know now only how to treat the common heart attack, but how to recognize this isn't the common heart attack but the 1 in 10,000 not a heart attack but has exactly the same symptoms as a heart attack case and give it the correct treatment.
However most of us are not in that situation. It is better for us to just look up those details as we need them because it gives us more room to handle a broader variety of situations.
Humans will never outcompete ai in that regard however. Industry will eventually optimize for humans and ai separately: ai will know a lot and think quickly, humans will provide judgement and legal accountability. We’re already on this path.
My favorite professor in my physics program would say, "You will never remember the equations I teach. But if you learn how the relationships are built and how to ask questions of those relationships, then I have done my job." He died a few years ago. I never was able to thank him for his lessons.
You cannot teach "how to to think". You have to give students thinking problems to actually train thinking. Those kinds of problems can increasingly be farmed off to AI, or at least certain subproblems in them.
I meam, yes, to an extent you can teach how to think: critical thinking and logic are topics you can teach and people who take their teaching to heart can become better thinkers. However, those topics cannot impart creativity. Critical thinking is called exactly that because it's about tools and skills for separating bad thinking from good thinking. The skill of generating good thinking probably cannot be taught; it can only be improved with problem-solving practice.
During some of the earlier web service development days, one would find people at F500 skating by in low-to-mid level jobs just cutting and pasting between spreadsheets, things would take them hours could be done in seconds, and with lower error rates, with a proper data interface.
Very anecdotally, but I hazard that most of these types of low-hanging fruit, low-value add roles are much less common since they tended to be blockers for operational improvement. Six-sigma, Lean, various flavors of Agile would often surface these low performers up and they either improved or got shown the door between 2005 - 2020.
Not that everyone is 100% all the time, every day, but what we are left with is often people that are highly competent at not just their task list but at their job.
> In the real world of work, everyone has full access to all of the available data and information.
In general, I also attend your church.
However, as I preached in that church, I had two students over the years.
* One was from an African country and told me that where he grew up, you could not "just look up data that might be relevant" because internet access was rare.
* The other was an ex US Navy officer who was stationed on a nuclear sub. She and the rest of the crew had to practice situations where they were in an emergency and cut off from the rest of the world.
Memorization of considerable amounts of data was important to both of them.
I've had a frustrating experience the past few years trying to hire junior sysadmins because of a real lack of problem solving skills once something went wrong outside of various playbooks they memorized to follow.
I don't need someone who can follow a pre-written playbook, I have ansible for that. I need someone that understands theory, regardless of specific implementations, and can problem solve effectively so they can handle unpredictable or novel issues.
To put another way, I can teach a junior the specifics of bind9 named.conf, or the specifics of our own infrastructure, but I shouldn't be expected to teach them what DNS in general is and how it works.
But the candidates we get are the opposite - they know specific tools, but lack more generalized theory and problem solving skills.
People have been saying that since the advent of formal education. Turns out standardized education is really hard to pull off and most systems focus on making the average good enough.
It’s also hard to teach people “how to think” while at the same time teaching them practical skills - there’s only so many hours in a day, and most education is setup as a way to get as many people as possible into shape for taking on jobs where “thinking” isn’t really a positive trait, as it’d lead to constant restructuring and questioning of the status quo
> My favorite professor in engineering school always gave open book tests.
My experience as a professor and a student is that this doesn't make any difference. Unless you can copy verbatim the solution to your problem from the book (which never happens), you better have a good understanding of the subject in order to solve problems in the allocated time. You're not going to acquire that knowledge during your test.
My experience as a professor and a student is that this doesn't make any difference.
Exactly the point of his test methodology.
What he asked of students on a test was to *apply* knowledge and information to *unique* problems and create a solution that did not exist in any book.
I only brought 4 things to his tests --- textbook, pencil, calculator and a capable, motivated and determined brain. And his tests revealed the limits of what you could achieve with these items.
Isn't this an argument for why you should allow open book tests rather than why you shouldn't? It certainly removes some pressure to remember some obscure detail or formula.
Isn't that just an argument for always doing open book tests, then? Seems like there's no downside, and as already mentioned, it's closer to how one works in the real world.
I had a like minded professor in university, ironically in AI. Our big tests were all 3 day take home assignments. The questions were open ended, required writing code, processing data and analyzing results.
I think the problem with this is that it requires the professor to mentally fully engage when marking assignments and many educators do not have the capacity and/or desire to do so.
Might be true, idk? For all we know that professor now gives a 2.5-day take home assignments where they are allowed to use LLMs, and then assess them in an 1 hour oral exam where they need to explain approach, results and how they ensure that their results are accurate?
I don't think the 3-day take home is the key. It's supporting educators to have the intention, agency and capacity to improvise assessment.
Same here! I always like to say that software engineering is 50% knowing the basics (How to write/read code, basic logic) and 50% having great research skills. So much of our time is spent finding documentation and understanding what it actually means as opposed to just writing code.
Each one of us has a mental toolbox that we use to solve problems. There are many more tools that we don’t have in our heads that we can look up if we know how.
The bigger your mental toolbox the more effective you will be at solving the problems. Looking up a tool and learning just enough to use it JIT is much slower than using a handy tool that you already masterfully know how to use.
This is as true for physical tools as for programming concepts like algorithms and data structures. In the worst case you won’t even know to look for a tool and will use whatever is handy, like the proverbial hammer.
> Everyone has full access to all of the available data and information
Ahh, but this is part of the problem. Yes, they have access, but there is -so much- information, it punches through our context window. So we resort to executive summaries, or convince ourselves that something that's relevant is actually not.
At least an LLM can take full view of the context in aggregate and peel out signal. There is value there, but no jobs are being replaced
I agree that an LLM is a long way from replacing most any single job held by a human in isolation. However, what I feel is missed in this discussion is that it can significantly reduce the total manpower by making humans more efficient. For instance, the job of a team of 20 can now be done by 15 or maybe even 10 depending on the class of work. I for one believe this will have a significant impact on a large number of jobs.
Not that I'm suggesting anything be "stopped". I find LLM's incredibly useful, and I'm excited about applying them to more and more of the mundane tasks that I'd rather not do in the first place, so I can spend more time solving more interesting problems.
Also, some problems don't have enough data for a solution. I had a professor that gave tests where the answer was sometimes "not solvable." Taking these tests was like sweating bullets because you were not sure if you're just too dumb to solve the problem, or there was not enough data to solve the problem. Good times!
I agree with the overall message, but I will say that there is still a great deal of value in memorisation. Memorising things gives you more internal tools to think in broader chunks, so you can solve more complicated problems.
(I do mean memorisation fairly broadly, it doesn't have to mean reciting a meaningless list of items.)
Talk is cheap. Good educators cost money, and America famously underpays (and under-appreciates) its teachers. Does he also support increasing taxes on the wealthy?
Have there been studies about abilities of different students to memorize information? I feel this is under-studied in the world of memorizing for exams
Everyone going through STEM needs to see the movie Hidden Figures for a variety of reasons, but one bit stands out as poignant: I believe it was Katherine Johnson, who is asked to calculate some rocket trajectory to determine the landing coordinates, thinks on it a bit and finally says, "Aha! Newton's method!" Then she runs down to the library to look up how to apply Newton's method. She had the conceptual tools to find a solution, but didn't have all the equations memorized. Having all the equations in short term memory only matters in a (somewhat pathological) school setting.
It is tough though, I'd like to think I learnt how to think analytically and critically. But thinking is hard, and often times I catch myself trying to outsource my thinking almost subconsciously. I'll read an article on HN and think "Let's go to the comment section and see what the opinions to choose from are", or one of the first instincts after encountering a problem is googling and now asking an LLM.
Most of us are also old enough to have had a chance to develop taste in code and writing. Many of the young generation lack the experience to distinguish good writing from LLM drivel.
wanted to chime in on the educational system. in the west, we have the 'banking system' which treat a student as a bank account and knowledge as currency, hence the dump more info into ppl to make them sm0rt attitude.
in developing areas, they actually implement more modern models commonly, as its newer and free to implement newer things.
those newer models focus more on exactly this. teach a person how to go through the process of finding solutions. rather than 'knowing a lot to enable the process of thinking'.
not saying what is better or worse, but reading this comment and article it reminds me of this.
a lot of people i see, they know tons of interesting things, but anything outside of their knowledge is a complete mystery.
all the while ppl from developing areas learn to solve issues. alot of individuals from there also, get out of their poverty and do really well for themselves.
ofcourse, this is a generalization and doesnt hold up in all cases. but i cant help think about it.
a lot of my colleagues dont know how to solve problems simply because they dont RTFM. they rely on knowledge from their education which is already outdated before they even sign up..
i try to teach them to RTFM. it seems hopeless. they look at me , downwards, because i have no papers. but if shit hits the fan, they come to me. solve the prolbem.
a wise guy i met once said (likely not his words). there are 2 type of ppl. those who think in problems, and those who think in solutions.
id related that to education, not prebaked human properties.
One of my favorite things about Feynman interviews/lectures is often his responses are about how to think. Sometimes physicists ask questions in his lectures and his answer has little to do with the physics, but how they're thinking about it. I like thinking about thinking, so Feynman is soothing.
It depends what level the education is happening at. Think of it like students being taught how to do for loops but are just copying and pasting AI output. That isn't learning. They aren't building the skills needed to debug when the AI gets something wrong with a more complicated loop, or understand the trade offs of loops vs recursion.
Finding the correct balance for a given class it hard. Generally, the lower level the education, the more it should be closed books because the more it is about being able to manually solve the smaller challenges that are already well solved so you build up the skills needed to even tackle the larger challenges. The higher the education level, the more it is about being able to apply those skills to then tackle a problem, and one of those skills is being able to pull relevant formulas and such from the larger body of known formulas.
While there’s no reasonable way to disagree with the sentiment, I don’t think I’ve ever met anyone who can “think and decompose problems” who isn’t also widely read, and knows a lot of things.
Forcing kids to sit and memorize facts isn’t suddenly going to make them a better thinker, but much of my process of being a better thinker is something akin to sitting around and memorizing facts. (With a healthy dose of interacting substantively and curiously with said facts)
My boss said we were gonna fire a bunch of people “because AI” as part of some fluff PR to pretend we were actually leaders in AI. We tried that a bit, it was a total mess and we have no clue what we’re doing, I’ve been sent out to walk back our comments.
Well they’re just trying to reduce headcount overall to get the expenses for AWS in better shape and work through some bloat. The “we’re doing layoffs because of AI” story wasn’t sticking though so looks like now they’re backtracking that story line.
He didn't actually say that. He said it's possible that within 2 years developers won't be writing much code, but he goes on to say:
"It just means that each of us has to get more in tune with what our customers need and what the actual end thing is that we're going to try to go build, because that's going to be more and more of what the work is as opposed to sitting down and actually writing code...."
If you read the full remarks they're consistent with what he says here. He says "writing code" may be a skill that's less useful, which is why it's important to hire junior devs and teach them how to learn so they learn the skills that are useful.
He is talking his book. Management thinks it adds value in the non-coding aspects of the product - such as figuring out what customers need etc. I suggest management stays in their lanes, and not make claims on how coding needs to be done, leave that to the craftsmen actually coding.
Theoretically, a large part of Amazon's worth is the skill of its workforce.
Some subset of the population likes to pretend their workforce is a cost that provides less than zero value or utility, and all the value and utility comes from shareholders.
But if this isn't true, and collective skill is worth value, then saying anyone can have that with AI at least has some headwind on your share price - which is all they care about.
Does that offset a potential tailwind from slightly higher margins?
I don't think any established company should be cheerleading that anyone can easily upset their monopoly with a couple of carefully crafted prompts.
It was always kind of strange to me, and seemed as though they were telling everyone, our moat is gone, and that is good.
If you really believed anyone could do anything with AI, then the risk of PEs collapsing would be high, which would be bad for the capital class. Now you have to correctly guess what's the next best thing constantly to keep your ROI instead of just parking it in save havens - like FAANG.
Amazon would probably say its worth is the machinery around workers that allows it to plug in arbitrary numbers of interchangeable people and have them be productive.
Bedrock/Q is a great example of how Amazon works. If we throw $XXX at the problem and YYY SDEs at the problem we should be able to build Github Copilot, GPT-3, OpenRouter and Cursor ourselves instead of trying to competitively acquire and attract talent. The fact that Codewhisperer, Q and Titan barely get spoken about on HN or Twitter tells you how successful this is.
But if you have that perspective then the equation is simple. If S3 can make 5 XXL features per year with 20 SDEs then if we adopt “Agentic AI” we should be able to build 10 XXL features with 10 SDEs.
Little care is given to organizational knowledge, experience, vision etc. that is the value (in their mind) of leadership not ICs.
What do you mean, “Amazon doesn’t really work that way”?
Parent is talking about how C-Suite doesn’t want to trumpet something that implies their entire corporate structure is extremely disadvantaged vs new entrants and your response is “Amazon wants to build everything themselves”?
Amazon isn’t some behaviorally deterministic entity, and it could (and should?) want to both preserve goodwill and build more internally vs pay multiples to acquire.
I guess it could be that people inside are not people they have to compete with, but it doesn’t seem like that’s what you're saying.
That's not necessarily inconsistent though - if you need people to guide or instruct the autonomy, then you need a pipeline of people including juniors to do that. Big companies worry about the pipeline, small companies can take that subsidy and only hire senior+, no interns, etc., if they want.
> Amazon Web Services CEO Matt Garman claims that in 2 years coding by humans won't really be a thing, and it will all be done by networks of AI's who are far smarter, cheaper, and more reliable than human coders.
Unless this guy speaks exclusively in riddles, this seems incredibly inconsistent.
There is no pipeline though. The average tenure of a junior developer even at AWS is 3 years. Everyone knows that you make less money getting promoted to an L5 (mid) than getting hired in as one. Salary compression is real. The best play is always to jump ship after 3 years. Even if you like Amazon, “boomeranging” is still the right play.
But that's fine, that's why I say for big companies - the pipeline is the entire industry, everyone potentially in the job market, not just those currently at AWS. Companies like Amazon have a large enough work force to care that there's people coming up even if they don't work there yet (or never do, but by working elsewhere free someone else to work at AWS).
They have an interest in getting those grads turned into would-be-L5s even if they leave for a different company. If they 'boomerang back' at L7 that's great. They can't if they never got a grad job.
that's interesting because that's how the consulting world works too. Start at a big firm, work for a few years, then jump to a small firm two levels above where you were. The after two years, come back to the big firm and get hired one level up from where you left. Rinse/repeat. It's the fastest promotion path in consulting.
It’s not necessarily “larger”, so much as different units. In a big company, the hiring budget is measured in headcount, but the promotion budget is measured in dollar percentage. It’s much easier to add $20k salary to get a hire done than to give that same person a $20k bump the following year.
It’s a lot cheaper to replace an employee by one who leaves at market rate than to pay all of your developers at market rate. Many are going to stick around because of inertia, their lack of ability to interview well, golden handcuffs of RSUs, they don’t feel like rebuilding the social capital at another company or the naive belief in the “mission”, “passion” etc
I went from an L5 (mid) working at AWS ProServe as a consultant (full time role) to a year later (and a shitty company in between) as a “staff architect” - like you said two levels up - at a smaller cloud consulting company.
If I had any interest in ever working for BigTech again (and I would rather get an anal probe daily with a cactus), I could relatively easily get into Google’s equivalent department as a “senior” based on my connections.
There's definitely a vibe shift underway. C-Suites are seeing that AI as a drop-in replacement for engineers is a lot farther off than initial hype suggested. They know that they'll need to attract good engineers if they want to stay competitive and that it's probably a bad idea to scare off your staff with saying that they'll be made irrelevant.
I'm not sure those are mutually exclusive? Modern coders don't touch Assembly or deal with memory directly anymore. It's entirely possible that AI leads to a world where typing code by hand is dramatically reduced too (it already has in a few domains and company sizes)
He was right tho. AI is doing all the coding. That doesn’t mean you fire junior staff. Both can be true at once- you need juniors, and pretty much all code this days is AI-generated.
He should face consequences for his cargo-cult thinking in the first place. The C-Suite isn't "getting" anything. They are simply bending like reeds in today's winds.
> “How's that going to work when ten years in the future you have no one that has learned anything,”
Pretty obvious conclusion that I think anyone who's thought seriously about this situation has already come to. However, I'm not optimistic that most companies will be able to keep themselves from doing this kind of thing, because I think it's become rather clear that it's incredibly difficult for most leadership in 2025 to prioritize long-term sustainability over short-term profitability.
That being said, internships/co-ops have been popular from companies that I'm familiar with for quite a while specifically to ensure that there are streams of potential future employees. I wonder if we'll see even more focus on internships in the future, to further skirt around the difficulties in hiring junior developers?
Might want to clarify things with your boss who says otherwise [1]? I do wish journalists would stop quoting these people unedited. No one knows what will actually happen.
I'm not sure those statements are in conflict with each other.
“My view is you absolutely want to keep hiring kids out of college and teaching them the right ways to go build software and decompose problems and think about it, just as much as you ever have.” - Matt Garman
"We will need fewer people doing some of the jobs that are being done today” - Amazon CEO Andy Jassy
If you're quoting something, the only ethical thing to do is as verbatim as possible and with a sufficient amount of context. Speeches should not be cleaned up to what you think they should have said.
Now, the question of who you go to for quotes, on the other hand .. that's how issues are really pushed around the frame.
By unedited I mean, take the message literally and quote it to support a narrative that isn’t clear or consistent. (even internally among Amazon leadership)
I very much believe that anything AWS says on the corporate level is bullshit.
From the perspective of a former employee. I knew that going in though. I was 46 at the time, AWS was my 8th job and knowing AWS’s reputation from 2nd and 3rd hand information, I didn’t even entertain an opportunity that would have forced me to relocate.
I interviewed for a “field by design” role that was “permanently remote” [sic].
But even those positions had an RTO mandate after I already left.
There's what AWS leadership says and then there's what actually sticks.
There's an endless series of one pagers with this idea or that idea, but from what I witnessed first hand, the ones that stuck were the ones that made money.
Jassy was a decent guy when I was there, but that was a decade ago. A CEO is a PR machine more than anything else, and the AI hype train has been so strong that if you do anything other than saying AI is the truth, the light and the way, you lose market share to competitors.
AI, much like automation in general, does allow fewer people to do more, but in my experience, customer desires expand to fill a vacuum and if fewer people can do more, they'll want more to the point that they'll keep on hiring more and more people.
Some exceptions occur for people getting Tenure without post doc or people doing some other things like taking undergraduate in one or two years. But no one expect that we for whole skip the first two and then get any senior researchers.
The same idea applies anywhere, the rule is that if you don't have juniors then you don't get seniors so better prepare your bot to do everything.
> Garman is also not keen on another idea about AI – measuring its value by what percentage of code it contributes at an organization.
You really want to believe, maybe even need to believe, that anyone who comes up with this idea in their head has never written a single line of code in their life.
It is on its face absurd. And yet I don't doubt for a second that Garman et al. have to fend off legions of hacks who froth at the mouth over this kind of thing.
> "Measuring software productivity by lines of code is like measuring progress on an airplane by how much it weighs." -- Bill Gates
Do we reward the employee who has added the most weight? Do we celebrate when the AI has added a lot of weight?
At first, it seems like, no, we shouldn't, but actually, it depends. If a person or AI is adding a lot of weight, but it is really important weight, like the engines or the main structure of the plane, then yeah, even though it adds a lot of weight, it's still doing genuinely impressive work. A heavy airplane is more impressive than a light weight one (usually).
I just can’t resist myself when airplanes come up in discussion.
I completely understand your analogy and you are right. However just to nitpick, it is actually super important to have a weight on the airplane at the right place. You have to make sure that your aeroplane does not become tail heavy or it is not recoverable from a stall. Also a heavier aeroplane, within its gross weight, is actually safer as the safe manoeuverable speed increases with weight.
If someone adds more code to the wrong places for the sake of adding more code, the software may not be recoverable for future changes or from bugs. You also often need to add code in the right places for robustness.
right, thats why its a great analogy - because you also need to have at least some code in a successful piece of software. But simply measuring by the amount of code leads to weird and perverse incentives - code added without thought is not good, and too much code can itself be a problem. Of course, the literal balancing aspect isn't as important.
Just to nitpick your nitpick, that’s only true up to a point, and the range of safe weights isn’t all that big really - max payload on most planes is a fraction of the empty weight. And planes can be overweight, reducing weight is a good thing and perhaps needed far more often than adding weight is needed. The point of the analogy was that over a certain weight, the plane doesn’t fly at all. If progress on a plane is safety, stability, or speed, we can measure those things directly. If weight distribution is important to those, that’s great we can measure weight and distribution in service of stability, but weight isn’t the primary thing we use.
Like with airplane weight, you absolutely need some code to get something done, and sometimes more is better. But is more better as a rule? Absolutely not.
This is a pretty narrow take on aviation safety. A heavier airplane has a higher stall speed, more energy for the brakes to dissipate, longer takeoff/landing distances, a worse climb rate… I’ll happily sacrifice maneuvering speed for better takeoff/landing/climb performance.
> the safe manoeuverable speed increases with weight
The reason this is true is because at a higher weight, you'll stall at max deflection before you can put enough stress on the airframe to be a problem. That is to say, at a given speed a heavier airplane will fall out of the air [hyperbole, it will merely stall - significantly reduced lift] before it can rip the wings/elevator off [hyperbole - damage the airframe]. That makes it questionable whether heavier is safer - just changes the failure mode.
> That is to say, at a given speed a heavier airplane will fall out of the air [hyperbole, it will merely stall - significantly reduced lift] before it can rip the wings/elevator off [hyperbole - damage the airframe]
Turbulence, especially generated by thunderstorms, or close to it.
Progress on airplanes is often tracked by # of engineering drawings released, which means that 1000s of little clips, brackets, fittings, etc. can sometimes misrepresent the amount of engineering work that has taken place compared to preparing a giant monolithic bulkhead or spar for release. I have actually proposed measuring progress by part weight instead of count to my PMs for this reason
It’s an analogy that gets the job done and is targeted at non-tech managers.
It’s not perfect. Dead code has no “weight” unless you’re in a heavily storage-constrained environment. But 10,000 unnecessary rivets has an effect on the airplane everywhere, all the time.
Assuming it is truly dead and not executable (which someone would have to verify is & remains the case), dead code exerts a pressure on every human engineer who has to read (around) it, determine that it is still dead, etc. It also creates risk that it will be inadvertently activated and create e.g. security exposure.
Yes, we all love pedantry around here (that’s probably 99% of the reason I wrote the original comment!)
But if your position is that the percentage of time in the software lifecycle that dead code has a negative effect on a system is anywhere close to the percentage of time in an aircraft lifecycle that extra non-functional rivets (or other unnecessary weight objects) has a negative effect on the aircraft, you’re just wrong.
it's still directionally accurate though. Dead code has a weight that must be paid. Sometimes the best commits are the ones where you delete a ton of lines.
In this analogy, I'd say dead code corresponds to airplane parts that aren't actually installed on the aircraft. When people talk about the folly of measuring productivity in lines of code, they aren't referring to the uselessness of dead code, they're referring to the harms that come from live code that's way bigger than it needs to be.
This reminds me of a piece on folklore.org by Andy Hertzfeld[0], regarding Bill Atkinson. A "KPI" was introduced at Apple in which engineers were required to report how many lines of code they had written over the week. Bill (allegedly) claimed "-2000" (a completely, astonishingly negative report), and supposedly the managers reconsidered the validity of the "KPI" and stopped using it.
I don't know how true this is in fact, but I do know how true this is in my work - you cannot apply some arbitrary "make the number bigger" goal to everything and expect it to improve anything. It feels a bit weird seeing "write more lines of code" becoming a key metric again. It never worked, and is damn-near provably never going to work. The value of source code is not in any way tied to its quantity, but value still proves hard to quantify, 40 years later.
Given the way that a lot of AI coding actually works, it’s like asking what percent of code was written by hitting tab to autocomplete (intellisense) or what percent of a document benefited from spellcheck.
While most of us know the next word guessing is how it works in reality…
That sentiment ignores the magic of how well this works. There are mind blowing moments using AI coding, to pretend that it’s “just auto correct and tab complete” is just as deceiving as “you can vibe code complete programs”.
"...just raised a $20M Series B and are looking to expand the team and products offered. We are fully bought-in to generative AI — over 40% of our codebase is built and maintained by AI, and we expect this number to continue to grow as the tech evolves and the space matures."
"What does your availability over the next couple of weeks look like to chat about this opportunity?"
"Yeah, quite busy over the next couple of weeks actually… the next couple of decades, really - awful how quickly time fills by itself these days, right? I'd have contributed towards lowering that 40% number which seems contrary to your goals anyway. But here's my card, should you need help with debugging something tricky some time in the near future and nobody manages to figure it out internally. I may be able to make room for you if you can afford it. I might be VERY busy though."
I want to have the model re-write patent applications, and if any portion of your patent filing was replicated by it your patent is denied as obvious and derivative.
Something I wonder about the percent of code - I remember like 5-10 years ago there was a series of articles about Google generating a lot of their code programmatically, I wonder if they just adapted their code gen to AI.
I bet Google has a lot of tools to say convert a library from one language to another or generate a library based on an API spec. The 30% of code these LLMs are supposedly writing is probably in this camp, not net novel new features.
As always, the truth is somewhere in the middle. AI is not going to replace everyone tomorrow, but I also don't think we can ignore productivity improvements from AI. It's not going to replace engineers completely now or in the near future, but AI will probably reduce the number of engineers needed to solve a problem.
ChatGPT is better than any junior developer I’ve ever worked with. Junior devs have always been a net negative for the first year or so.
From a person who is responsible for delivering projects, I’ve never thought “it sure would be nice if I had a few junior devs”. Why when I can poach an underpaid mid level developer for 20% more?
I've never had a junior dev be a "net negative." Maybe you're just not supervising or mentoring them at all? The first thing I tell all new hires under me is that their job is to solve more problems than they create, and so far it's worked out.
Truly depends on the organization and systems. I’m at a small firm with too few Senior staff, lots of fire-fighting going on among us, etc. We have loads of low-hanging fruit for our Juniors so we tend to have very quick results after an initial onboarding.
I just “wrote” 2000 lines of code for a project between Node for the AWS CDK and Python using the AWS SDK (Boto3). Between both, ChatGPT needed to “know” the correct API for 12 services, SQL and HTML (for a static report). The only thing it got wrong with a one shot approach was a specific Bedrock message payload for a specific LLM model. That was even just a matter of saying “verify the payload on the web using the official docs”.
Yes it was just as well structured as I - someone who has been coding as a hobby or professionally for four decades - would have done.
The difference is probably that I only do green field POC implementations as a solely developer/cloud architect on a project if I am doing hands on keyboard work.
The other part of my job is leading larger projects where I purposefully don’t commit to pulling stories off the board since I’m always in meetings with customers, project managers, sales or helping other engineers.
I might even then do a separate POC as a research project/enablement. But it won’t be modifying existing code that I didn’t design.
That's great for you. I ask Sonnet 4 to make a migration and a form in Laravel Filament, and it regularly shits itself. I'm curious what those 12 services were, they must've had unchanging, well documented APIs.
That’s the advantage of working with AWS services, everything is well documented with plenty of official and unofficial code showing how to do most things.
Even for a service I know is new, I can just tell it to “look up the official documentation”
Using ChatGPT 5 Fast
AWS CDK apps (separate ones) using Node
- EC2 (create an instance)
- Aurora MySQL Serverless v2
- Create a VPC with no internet access - the EC2 instance was used as a jump box using Session Manager
- VPC Endpoints for Aurora control plane, SNS, S3, DDB, Bedrock, SQS, Session Manager
- Lambda including using the Docker lambda builder
- DDB
- it also created the proper narrowly scoped IAM permissions for tfe lambdas (I told it the services the Lambdas cared about)
The various Lambdas in Python using Boto3
- Bedrock including the Converse and Invoke APIs for the Nova and Anthropic families
- knowing how to process SQS Messages coming in as events
- MySQL flavored SQL for Upserts
- DDB reads
In another project the services were similar with the addition of Amazon Transcribe.
I’m a great mentor when given the time. Two former interns for whom I was their official mentor during my time at AWS got return offers and are thriving two years after I left. I threw one in front of a customer to lead the project within three months after they came back after graduating. They were able to come through technically and had the soft skills. I told them my training approach is to “throw them at the bus. But never under the bus.”
I’m also a great teacher. That’s my $DayJob and has been for the past decade first bringing in new to the company processes and technologies, leading initiatives, teaching other developers, working with sales, CxOs (smaller companies), directors, explaining large “organizational transformation” proposals etc. working at startups and then doing the same in cloud consulting first working at AWS (ProServe full time role) and now working as a staff architect full time at a third party consulting company.
But when I have been responsible for delivery, I only hire people who have experience “dealing with ambiguity” and show that I can give them a decently complicated problem and they can take the ball and run with it and make decent decisions and do research. I don’t even do coding interviews - when I interview it’s strictly behavioral and talking through their past projects, decision making processes, how they overcame challenges etc.
In terms of AWS LPs, it’s “Taking Ownership” (yeah quoting Amazon LPs made me throw up a little).
The most impressive folks Ive worked with are almost always straight out of school. It's before they've developed confidence about their skills and realized they can be more successful by starting their own business. People who get promoted three times in just 5 years sort of good.
Did their project manager and/or team lead think when they were hired “they are really going to be a great asset to my team and are going to help me complete my sprint/quarterly goals”?
When I ask for additional headcount, I’m looking at the next quarter since that’s what my manager is judging me based on.
My evaluations are based on quarterly goals and quarterly deliverables. No one at a corporation cares about anything above how it affects them.
Bringing junior developers up to speed just for them to jump ship within three years or less doesn’t benefit anyone at the corporate level. Sure they jump ship because of salary compression and inversion, where internet raises don’t correspond to market rates. Even first level managers don’t have a say so or budget to affect that.
This is true for even BigTech companies. A former intern I mentored who got a return offer a year before I left AWS just got promoted to an L5 and their comp package was 20% less than new hires coming in at an l5.
Everyone will be long gone from the company if not completely retired by the time that happens.
> Bringing junior developers up to speed just for them to jump ship within three years or less doesn’t benefit anyone at the corporate level.
What? Of course it does. If that's happening everywhere, that means other companies' juniors are also jumping ship to come work for you while yours jump ship to work elsewhere. The only companies that don't see a benefit from mentoring new talent are those with substandard compensation.
That’s true, but why should I take on the work of being at the beginning of the pipeline instead of hiring a mid level developer. My incentives are to meet my quarterly goals and show “impact”.
To a first approximation, no company pays internal employees at market rates in an increasing comp environment after a couple of years especially during the first few years of an employee’s career where their marker rate rapidly increases once they get real world experience.
On the other hand, the startup I worked for pre-AWS with 60 people couldn’t, wouldn’t and shouldn’t have paid me the amount I made when I got hired at AWS.
I'm a technical co-founder rapidly building a software product. I've been coding since 2006. We have every incentive to have AI just build our product. But it can't. I keep trying to get it to...but it can't. Oh, it tries, but the code it writes is often overly complex and overly-verbose. I started out being amazed at the way it could solve problems, but that's because I gave it small, bounded, well-defined problems. But as expectations with agentic coding rose, I gave it more abstract problems and it quickly hit the ceiling. As was said, the engineering task is identifying the problem and decomposing it. I'd love to hear from someone who's used agentic coding with more success. So far I've tried Co-pilot, Windsurf, and Alex sidebar for Xcode projects. The most success I have is via a direct question with details to Gemini in the browser, usually a variant of "write a function to do X"
> As was said, the engineering task is identifying the problem and decomposing it.
In my experience if you do this and break the problem down into small pieces, the AI can implement the pieces for you.
It can save a lot of time typing and googling for docs.
That said, once the result exceeds a certain level of complexity, you can't really ask it to implement changes to existing code anymore, since it stops understanding it.
At which point you now have to do it yourself, but you know the codebase less well than if you'd hand written it.
So, my upshot is so far that it works great for small projects and for prototyping, but the gain after a certain level of complexity is probably quite small.
But then, I've also find quite some value in using it as a code search engine and to answer questions about the code, so maybe if nothing else that would be where the benefit comes from.
> At which point you now have to do it yourself, but you know the codebase less well than if you'd hand written it.
Appreciate you saying this because it is my biggest gripe in these conversations. Even if it makes me faster I now have to put time into reading the code multiple times because I have to internalize it.
Since the code I merge into production "is still my responsibility" as the HN comments go, then I need to really read and think more deeply about what AI wrote as opposed to reading a teammate's PR code. In my case that is slower than the 20% speedup I get by applying AI to problems.
I'm sure I can get even more speed if I improve prompts, when I use the AI, agentic vs non-agentic, etc. but I just don't think the ceiling is high enough yet. Plus I am someone who seems more prone to AI making me lazier than others so I just need to schedule when I use it and make that time as minimal as possible.
>teach “how do you think and how do you decompose problems”
That's rich coming from AWS!
I think he meant "how do you think about adding unnecessary complexity to problems such that it can enable the maximum amount of meetings, design docs and promo packages for years to come"!
Junior staff will be necessary but you'll have to defend them from the bean-counters.
You need people who can validate LLM-generated code. It takes people with testing and architecture expertise to do so. You only get those things by having humans get expertise through experience.
Yesterday, I was asked to scrape data from a website. My friend used ChatGPT to scrape data but didn't succeded even spent 3h+. I looked website code and understand with my web knowledge and do some research with LLM. Then I described how to scrape data to LLM it took 30 minutes overall. The LLM cant create best way but you can create with using LLM. Everything is same, at the end of the day you need someone who can really think.
LLM's can do anything, but the decision tree for what you can do in life is almost infinite. LLM's still need a coherent designer to make progress towards a goal.
it is not that easy, there is lazy loading in the page that is triggered by scroll of specific sections. You need to find clever way, no way to scrape with bs4, so tough with even selenium.
It is too late it is already happening. The evolution of tech field is people being more experienced and not AI. But AI will be there for questions and easy one liners. Properly formalized documentation, even TLDRs.
The cost of not hiring and training juniors is trying to retain your seniors while continuously resetting expectations with them about how they are the only human accountable for more and more stuff.
Finally someone from a top position said this. After all the trash the CEOs have been spewing and sensationalizing every AI improvement, for a change, a person in a non-engineering role speaks the truth.
Point is nobody has figured out how much AI can replace humans. People. There is so much of hype out there as every tech celebrity sharing their opinions without responsibility of owning them. We have to wait & see. We could change courses when we know the reality. Until then, do what we know well.
If AI is so great and had PhD level skills (Musk) then logic says you should be replacing all of your _senior_ developers. That is not the conclusion they reached which implies that the coding ability is not that hot.
Q.E.D.
LLMs are actually -the worst- at doing very specific repetitive things. It'd be much more appropriate for one to replace the CEO (the generalist) rather than junior staff.
> “My view is you absolutely want to keep hiring kids out of college and teaching them the right ways to go build software and decompose problems and think about it, just as much as you ever have.”
Instead you should replace senior staff who make way more.
I mean I used Copilot / JetBrains etc. to work on my code base but for large scale changes it did so much damage that it took me days to fix it and actually slowed me down. These systems are just like juniors in their capabilities, actually worse because junior developers are still people and able to think and interact with you coherently over days or weeks or months, these models aren’t even at that level I think.
Claude code is better than a junior programmer by a lot and these guys think it only gets better from there and they have people with decades in the industry to burn through before they have to worry about retraining a new crop.
Most people don't notice but there has been a inflation in headcounts over the years now. This happened around the time microservices architecture trend took over.
All of sudden to ensure better support and separation of concerns people needed a team with a manager for each service. If this hadn't been the case, the industry as a whole can likely work with 40% - 50% less people eventually. Thats because at any given point in time even with a large monolithic codebase only 10 - 20% of the code base is in active evolution, what that means in microservices world is equivalent amount teams are sitting idle.
When I started out huge C++ and Java code bases were pretty much the norm, and it was also one of the reasons why things were hard and barrier to entry high. In this microservices world, things are small enough that any small group of even low productivity employees can make things work. That is quite literally true, because smaller things that work well don't even need all that many changes on a everyday basis.
To me its these kind of places that are in real trouble. There is not enough work to justify keeping dozens to even hundreds of teams, their managements and their hierarchies all working for quite literally doing nothing.
Its almost an everyday song that I hear, that big companies are full of hundreds or thousands of employees doing nothing.
I think sometimes the definition of work gets narrowed to a point so infinitesimal that everyone but the speaker is just a lazy nobody.
There was an excellent article on here about working at enterprise scale. My experience has been similar. You get to do work that feels really real, almost like school assignments with instant feedback and obvious rewards when you're at a small company. When I worked at big companies it all felt like bullshit until I screwed it up and a senator was interested in "Learning more" (for example).
The last few 9s are awful hard to chase down and a lot of the steps of handling edge case failures or features are extremely manual.
There are two freight trains currently smashing into each other:
1.) Elon fired 80% of twitter and 3 years later it still hasn't collapsed or fallen into technical calamity. Every tech board/CEO took note of that.
2.) Every kid and their sister going to college who wants a middle class life with generous working conditions is targeting tech. Every teenage nerd saw those over employed guys making $600k from their couch during the pandemic.
On the other hand while yes it's still running, twitter is mostly not releasing new features, and has completely devolved into the worst place on the internet. Not to mention most accounts now actually are bots like Elon claimed they were 3 years ago.
I dont know; they may well have been on the ads and moderation side. And they did add the hangouts/voice calls stuff, but that may have been an acquisition, I'm not sure.
I hope the tech boards and CEOs don’t miss the not very subtle point that twitter has very quickly doubled in size in 2 years and is still growing after the big layoff and they had to scramble to fix some notable mistakes they made when firing that many people. 80% is already a hugely misleading marketing number.
Also need to add, that a large part of the 80% that got kicked, was moderator staff. So it makes sense that after they removed too many developers, they ended up rehiring them.
Take in account, Twitter their front end, the stuff that people interact with was only like 15% of the actual code base. The rest was analytics for the data (selling data, marketing analytic for advertisers etc).
But as they are not reintroducing moderators, the company is "still down by 63.6% from the numbers before the mass layoffs".
So technically, Twitter is probably back or even bigger on the IT staff then before Musk came.
> In this microservices world, things are small enough that any small group of even low productivity employees can make things work. That is quite literally true, because smaller things that work well don't even need all that many changes on a everyday basis.
You're committing the classic fallacy around microservices here. The services themselves are simpler. The whole software is not.
When you take a classic monolith and split it up into microservices that are individually simple, the complexity does not go away, it simply moves into the higher abstractions. The complexity now lives in how the microservices interact.
In reality, the barrier to entry on monoliths wasn't that high either. You could get "low productivity employees" (I'd recommend you just call them "novices" or "juniors") to do the work, it'd just be best served with tomato sauce rather than deployed to production.
The same applies to microservices. You can have inexperienced devs build out individual microservices, but to stitch them together well is hard, arguably harder than ye-olde-monolith now that Java and more recent languages have good module systems.
> If this hadn't been the case, the industry as a whole can likely work with 40% - 50% less people eventually. Thats because at any given point in time even with a large monolithic codebase only 10 - 20% of the code base is in active evolution, what that means in microservices world is equivalent amount teams are sitting idle.
I think it depends on the industry. In safety critical systems, you need to be testing, making documentation, architectural artifacts, meeting with customers, etc
There's not that much idle time. Unless you mean idle time actually writing code and that's not always a full time job.
I think most people misunderstand the relationship between business logic, architecture and headcount.
Big businesses don’t inherently require the complexity of architecture they have. There is always a path-dependent evolution and vestigial complexity proportional to how large and fast they grew.
The real purpose of large scale architecture is to scale teams much moreso than business logic. But why does headcount grow? Is it because domains require it? Sure that’s what ambitious middle managers will say, but the real reason is you have money to invest in growth (whether from revenue or from a VC). For any complex architecture there is usually a dramatically simpler one that could still move the essential bits around, it just might not support the same number of engineers delineated into different teams with narrower responsibilities.
The general headcount growth and architecture trajectory is therefore governed by business success. When we’re growing we hire and we create complex architecture to chase growth in as many directions as possible. Eventually when growth slows we have a system that is so complex it requires a lot of people just to understand and maintain—even if the headcount is longer justified those with power in the human structure will bend over backwards to justify themselves. This is where the playbook changes and a private equity (or Elon) mentality is applied to just ruthlessly cut and force the rest of the people how to keep the lights on.
I consider advances in AI and productivity orthogonal to all this. It will affect how people do their jobs, what is possible, and the economics of that activity, but the fundamental dynamics of scale and architectural complexity will remain. They’ll still hire more people to grow and look for ways to apply them.
This is blatantly incorrect. Before microservices became the norm you still had a lot of teams and hiring, but the teams would be working with the same code base and deployment pipeline. Every company that became successful and needed to scale invented their own bespoke way to do this; microservices just made it a pattern that could be repeatedly applied.
Well yeah... computers are really powerful. you don't need docker swarm or any other newfangled thing. Just perl and apache and mysql and you can ship to tens of millions of users before you hit scaling limits.
It would be sad if you are correct. Your company might not be able to justify keeping dozens and hundreds of teams employed, but what happens when other companies can't justify paying dozens and hundreds of teams who are the customers buying your product? Those who gleefully downsize might well deserve the market erosion they cause.
I think that the starting point is that productivity/developer has been declining for a while, especially at large companies. And this leads to the "bloated" headcount.
The question is why. You mention microservices. I'm not convinced.
Many think it is "horizontals". Possible, these taxes add up it is true.
Perhaps it is cultural? Perhaps it has to do with the workforce in some manner. I don't know and AFAIK it has not been rigorously studied.
A lot of companies that have stopped hiring junior employees are going to be really hurting in a couple of years, once all of their seniors have left and they have no replacements trained and ready to go.
junior engineers aren't hired to get tons of work done; they're hired to learn, grow, and eventually become senior engineers. ai can't replace that, but only help it happen faster (in theory anyway).
Time will show you are right and almost all the other dumbasses here at HN are wrong. Which is hardly surprising since they are incapable of applying themselves around their coming replacement.
They are 100% engineers and 100% engineers have no ability to adapt to other professions. Coding is dead, if you think otherwise then I hope you are right! But I doubt it because so far I have only heard arguments to the counter that are obviously wrong and retarded.
Unfortunately, this is the kind of view that is at once completely correct and anathema to private equity because they can squeeze a next quarter return by firing a chunk of the labor force.
Are we trying to guilt trip corporations to do socially responsible thing regarding young workers skill acquisition?
Haven't we learned that it almost always ends up in hollow PR and marketing theater?
Basically the solution to this is extending education so that people entering workforce are already at senior level. Of course this can't be financed by the students, because their careers get shortened by longer education. So we need higher taxes on the entities that reap the new spoils. Namely those corporations that now can pass on hiring junior employees.
> “Often times fewer lines of code is way better than more lines of code,” he observed. “So I'm never really sure why that's the exciting metric that people like to brag about.”
I remember someone that had a .sig that I loved (Can't remember where. If he's here, kudos!):
> I hate code, and want as little of it in my programs as possible.
No one's getting replaced, but you may not hire that new person that otherwise would have been needed. Five years ago, you would have hired a junior to crank out UI components, or well specc'd CRUD endpoints for some big new feature initiative. Now you probably won't.
I’m really tired of this trope. I’ve spent my whole career on “boring CRUD” and the number of relational db backed apps I’ve seen written by devs who’ve never heard of isolation levels is concerning (including myself for a time).
Coincidentally, as soon as these apps see any scale issues pop up.
On the other hand, that extra money can be used to expand the business in other ways, plus most kids coming out of college these days are going to be experts in getting jobs done with AI (although they will need a lot of training in writing actual secure and maintainable code).
Even the highest ranking engineers should be experts. I don’t understand why there’s this focus on juniors as the people who know AI best.
Using AI isn’t rocket science. Like you’re talking about using AI as if typing a prompt in English is some kind of hard to learn skill. Do you know English? Check. Can you give instructions? Check. Can you clarify instructions? Check.
> I don’t understand why there’s this focus on juniors as the people who know AI best.
Because junior engineers have no problem with wholeheartedly embracing AI - they don't have enough experience to know what doesn't work yet.
In my personal experience, engineers who have experience are much more hesitant to embrace AI and learn everything about it, because they've seen that there are no magic bullets out there. Or they're just set in their ways.
To management that's AI obsessed, they want those juniors over anyone that would say "Maybe AI isn't everything it's cracked up to be." And it really, really helps that junior engineers are the cheapest to hire.
Sure. First line tech support as well. In many situations customers will get vastly superior service if AI agent answers the call.
At least in my personal case, struggling with renewal at Virgin Broadband, multiple humans wasted probably an hour of everyone's time overall on the phone bouncing me around departments, unable to comprehend my request, trying to upsell and pitch irrelevant services, applying contextually inappropriate talking scripts while never approaching what I was asking them in the first place. Giving up on those brainless meat bags and engaging with their chat bot, I was able to resolve what I needed in 10 minutes.
> Sure. First line tech support as well. In many situations customers will get vastly superior service if AI agent answers the call.
In IT, if at a minimum, AI would triage the problem intelligently (and not sound like a bot while doing it), that would save my more expensive engineers a lot more time.
This is mostly because CS folks are given such sales and retention targets; and while I’ve never encountered a helpful support bot even in the age of LLMs, I presume in your case the company management was just happy to have a support bot talking to people without said metrics.
In India most of the banks now have apps that do nearly all the banking you can do by visiting a branch personally. To that extent this future is already here.
When I had to close my loan and had to visit a branch nearly a few times, the manager tells me, significant portion of his people's time now goes into actual banking- which according to him was selling products(fixed deposits, insurances, credit cards) and not customer support(which the bank thinks is not its job and has to because there is no other alternative to it currently).
Again, you assume those people have choice, you definitely should search more how people on these jobs are pressured to reach quotas and are abused in many ways. A simple search on Reddit you can see plenty of reports about it:
It's refreshing to finally see CEOs and other business leaders coming around to what experienced, skeptical engineers have been saying for this entire hype cycle.
I assumed it would happen at some point, but I am relieved that the change in sentiment has started before the bubble pops - maybe this will lesson the economic impact.
Yeah, the whole AI thing has very unpleasant similarities to the dot com bubble that burst to the massive detriment of the careers of the people that were working back then.
The parallels in how industry members talk about it is similar as well. No one denies that the internet boom was important and impactful, but it's also undeniable that companies wasted unfathomable amounts of cash for no return at the cost of worker well being.
I completely agree.
On a side note.. ya’ll must be prompt wizards if you can actually use the LLM code.
I use it for debugging sometimes to get an idea, or a quick sketch up of an UI.
As for actual code.. the code it writes is a huge mess of spaghetti code, overly verbose, with serious performance and security risks, and complete misunderstanding of pretty much every design pattern I give it..
I read AI coding negativity on Hacker News and Reddit with more and more astonishment every day. It's like we live in different worlds. I expect the breadth of tooling is partly responsible. What it means to you to "use the LLM code" could be very different from what it means to me. What LLM are we talking about? What context does it have? What IDE are you using?
Personally, I wrote 200K lines of my B2B SaaS before agentic coding came around. With Sonnet 4 in Agent mode, I'd say I now write maybe 20% of the ongoing code from day to day, perhaps less. Interactive Sonnet in VS Code and GitHub Copilot Agents (autonomous agents running on GitHub's servers) do the other 80%. The more I document in Markdown, the higher that percentage becomes. I then carefully review and test.
It’s not just you, I think some engineers benefit a lot from AI and some don’t. It’s probably a combination of factors including: AI skepticism, mental rigidity, how popular the tech stack is, and type of engineering. Some problems are going to be very straightforward.
I also think it’s that people don’t know how to use the tool very well. In my experience I don’t guide it to do any kind of software pattern or ideology. I think that just confuses the tool. I give it very little detail and have it do tasks that are evident from the code base.
Sometimes I ask it to do rather large tasks and occasionally the output is like 80% of the way there and I can fix it up until it’s useful.
> I also think it’s that people don’t know how to use the tool very well.
I think this is very important. You have to look at what it suggests critically, and take what makes sense. The original comment was absolutely correct that AI-generated code is way too verbose and disconnected from the realities of the application and large-scale software design, but there can be kernels of good ideas in its output.
Yah. Latest thing I wrote was
* Code using sympy to generate math problems testing different skills for students, with difficulty values affecting what kinds of things are selected, and various transforms to problems possible (e.g. having to solve for z+4 of 4a+b instead of x) to test different subskills
(On this part, the LLM did pretty well. The code was correct after a couple of quick iterations, and the base classes and end-use interfaces are correct. There's a few things in the middle that are unnecessarily "superstitious" and check for conditions that can't happen, and so I need to work with the LLM to clean it up.
* Code to use IRT to estimate the probability that students have each skill and to request problems with appropriate combinations of skills and difficulties for each student.
(This was somewhat garbage. Good database & backend, but the interface to use it was not nice and it kind of contaminated things).
* Code to recognize QR codes in the corners of worksheet, find answer boxes, and feed the image to ChatGPT to determine whether the scribble in the box is the answer in the correct form.
(This was 100%, first time. I adjusted the prompt it chose to better clarify my intent in borderline cases).
The output was, overall, pretty similar to what I'd get from a junior engineer under my supervision-- a bit wacky in places that aren't quite worth fixing, a little bit of technical debt, a couple of things more clever that I didn't expect myself, etc. But I did all of this in three hours and $12 expended.
The total time supervising it was probably similar to the amount of time spent supervising the junior engineer... but the LLM turns things around quick enough that I don't need to context switch.
I think a lot of it is tool familiarity. I can do a lot with Cursor but frankly I find out about "big" new stuff every day like agents.md. If I wasn't paying attention or also able to use Cursor at home then I'd probably learn more inefficiently. Learning how to use rule globs versus project instructions was a big learning moment. As I did more LLM work on our internal tools that was also a big lesson in prompting and compaction.
Certain parts of HN and Reddit I think are very invested in nay-saying because it threatens their livelihoods or sense of self. A lot of these folks have identities that are very tied up in being craftful coders rather than business problem solvers.
> B2B SaaS
Perhaps that's part of it.
People here work on all kinds of industries. Some of us are implementing JIT compilers, mission-critical embedded systems or distributed databases. In code bases like this you can't just wing it without breaking a million things, so LLM agents tend to perform really poorly.
> People here work on all kinds of industries.
Yes, it would be nice to have a lot more context (pun intended) when people post how many LoC they introduced.
B2B SaaS? Then can I assume that a browser is involved and that a big part of that 200k LoC is the verbose styling DSL we all use? On the other hand, Nginx, a production-grade web server, is 250k LoC (251,232 to be exact [1]). These two things are not comparable.
The point being that, as I'm sure we all agree, LoC is not a helpful metric for comparison without more context, and different projects have vastly different amounts of information/feature density per LoC.
[1] https://openhub.net/p/nginx
On the other hand, fault-intolerant codebases are also often highly defined and almost always have rigorous automated tests already, which are two contexts where coding agents specifically excel in.
Perhaps the issue is you were used to writing 200k lines of code. Most engineers would be agast at that. Lines of code is a debit not a credit
I am now making an emotional reaction based on zero knowledge of the B2B codebase's environment, but to be honest I think it is relevant to the discussion on why people are "worlds apart".
200k lines of code is a failure state. At this point you have lost control and can only make changes to the codebase through immense effort, and not at a tolerable pace.
Agentic code writers are good at giving you this size of mess and at helping to shovel stuff around to make changes that are hard for humans due to the unusable state of the codebase.
If overgrown barely manageble codebases are all a person's ever known and they think it's normal that changes are hard and time-consuming and needing reams of code, I understand that they believe AI agents are useful as code writers. I think they do not have the foundation to tell mediocre from good code.
I am extremely aware of the judgemental hubris of this comment. I'd not normally huff my own farts in public this obnoxiously, but I honestly feel it is useful for the "AI hater vs AI sucker" discussion to be honest about this type of emotion.
It really depends on what your use case is. E.g. of you're dealing with a lot of legacy integrations, dealing with all the edge cases can require a lot of code that you can't refactor away through cleverness.
Each integration is hopefully only a few thousand lines of code, but if you have 50 integrations you can easily break 100k loc just dealing with those. They just need to be encapsulated well so that the integration cruft is isolated from the core business logic, and they become relatively simple to reason about
In that case, LLMs are full on debt-machines.
My AI experience has varied wildly depending on the problem I'm working on. For web apps in Python, they're fantastic. For hacking on old engineering calculation code written in C/C++, it's an unmitigated disaster and an active hindrance.
And also ask: "How much money do you spend on LLMs?"
In the long run, that is going to be what drives their quality. At some point the conversation is going to evolve from whether or not AI-assisted coding works to what the price point is to get the quality you need, and whether or not that price matches its value.
> It's like we live in different worlds.
There is the huge variance in prompt specificity as well as the subtle differences inherent to the models. People often don't give examples when they talk about their experiences with AI so it's hard to get a read on what a good prompt looks like for a given model or even what a good workflow is for getting useful code out of it.
Some gave. Some even recorded it, and showed it, because they thought that they are good with it. But they weren’t good at all.
They were slower than coding by hand, if you wanted to keep quality. Some were almost as quick as copy-pasting from the code just above the generated one, but their quality was worse. They even kept some bugs in the code during their reviews.
So the different world is probably what the acceptable level of quality means. I know a lot of coders who don’t give a shit whether it makes sense what they’re doing. What their bad solution will cause in the long run. They ignore everything else, just the “done” state next to their tasks in Jira. They will never solve complex bugs, they simply don’t care enough. At a lot of places, they are the majority. For them, LLM can be an improvement.
Claude Code the other day made a test for me, which mocked everything out from the live code. Everything was green, everything was good. On paper. A lot of people simply wouldn’t care to even review properly. That thing can generate a few thousands of lines of semi usable code per hour. It’s not built to review it properly. Serena MCP for example specifically built to not review what it does. It’s stated by their creators.
> Personally, I wrote 200K lines of my B2B SaaS
That would probably be 1000 line of Common Lisp.
that no one could read
I think that is the 200 lines of the perl version.
you put linefeeds in your perl?
I deal with a few code bases at work and the quality differs a lot between projects and frameworks.
We have 1-2 small python services based on Flask and Pydantic, very structured and a well-written development and extension guide. The newer Copilot models perform very well with this, and improving the dev guidelines keep making it better. Very nice.
We also have a central configuration of applications in the infrastructure and what systems they need. A lot of similarly shaped JSON files, now with a well-documented JSON schema (which is nice to have anyway). Again, very high quality. Someone recently joked we should throw these service requests at a model and let it create PRs to review.
But currently I'm working in Vector and it's Vector remap language... it's enough of a mess that I'm faster working without any copilot "assistance". I think the main issue is that there is very little VRL code out in the open, and the remaps depend on a lot of unseen context, which one would have to work on giving to the LLM. Had similar experiences with OPA and a few more of these DSLs.
B2B SaaS in most cases are sophisticated masks over some structured data, perhaps with great ux, automation and convenience, so I can see LLMs be more successful there, even so because there is more training data and many processes are streamlined. Not all domains are equal, go try develop a serious game, not the yet another simple and broken arcade, with llms and you'll have a different take
I think its down to language and domain more than tools.
No model ive tried can write, usefully debug or even explain cmake. (It invents new syntax if it gets stuck, i often have to prompt multiple AI to know if even the first response in the context was made-up)
My luck with embedded c has been atrocious for existing codebase (burning millions of tolkens), but passable for small scripts. (Arduino projects)
My experience with python is much better. Suggesting relevant libraries and functions, debugging odd errors, or even making small script on its own. Even the original github copilot which i got access to early was excellent on python.
Alot of people that seem to have fully embraced agentic vibe-coding seem to be in the web or node.js domain. Which I've not done myself since pre-AI.
I've tried most (free or trial) major models or schemes in hope that i find any of them useful, but not found much use yet.
GitHub copilot, Microsoft copilot, Gemini, loveable, gpt, cursor with Claude models, you name it.
What tech stack do you use?
Betting in advance that it's JavaScript or Python, probably with very mainstream libraries or frameworks.
FWIW. Claude Code does great job for me on complex domain Rust projects, but I just use it one relatively small feature/code chunk at the time, where oftentimes it can pick up existing patterns etc. (I try to point it at similar existing code/feature if I have it). I do not let it write anything creative where it has to come up with own design (either high-level architectural, or low level facilities). Basically I draw the lines manually, and let it color the space between, using existing reference pictures. Works very, very well for me.
Is this meant to detract from their situation? These tech stacks are mainstream because so many use them... it's only natural that AI would be the best at writing code in contexts where it has the most available training data.
> These tech stacks are mainstream because so many use them
That's a tautology. No, those tech stacks are mainstream because it is easy to get something that looks OK up and running quickly. That's it. That's what makes a framework go mainstream: can you download it and get something pretty on the screen quickly? Long-term maintenance and clarity is absolutely not a strong selection force for what goes mainstream, and in fact can be an opposing force, since achieving long-term clarity comes with tradeoffs that hinder the feeling of "going fast and breaking things" within the first hour of hearing about the framework. A framework being popular means it has optimized for inexperienced developers feeling fast early, which is literally a slightly negative signal for its quality.
No, it's a clarification. There is massive difference between domains, and the parent post did not specify.
If the AI can only decently do JS and Python, then it can fully explain the observed disparity in opinion of its usefulness.
You are exactly right in my case - JavaScript and Python dealing with the AWS CDK and SDK. Where there is plenty of documentation and code samples.
Even when it occasionally gets it wrong, it’s just a matter of telling ChatGPT - “verify your code using the official documentation”.
But honestly, even before LLMs when deciding on which technology, service, or frameworks to use I would always go with the most popular ones because they are the easiest to hire for, easiest to find documentation and answers for and when I myself was looking for a job, easiest to be the perfect match for the most jobs.
Yeah, but most devs are working on brownfield projects where they did not choose any part of the tech stack.
They can choose jobs. Starting with my 3rd job in 2008, I always chose my employer based on how it would help me get my n+1 job and that was based on tech stack I would be using.
Once a saw a misalignment between market demands and current tech stack my employer was using, I changed jobs. I’m on job #10 now.
When did you get your last 3 jobs?
Mid 2020 - at AWS ProServe the internal consulting arm of AWS - full time job
Late 2023 - full time at a third party AWS consulting company. It took around two weeks after I started looking to get an offer
Late 2024 - “Staff consultant” third party consulting company. An internal recruiter reached out to me.
Before 2020 I was just a run of the mill C#/JS enterprise developer. I didn’t open the AWS console for the first time until mid 2018.
[flagged]
It is quite a putdown to tell someone else that if you wrote their program it would be 10x shorter.
That's not in keeping with either the spirit of this site or its rules: https://news.ycombinator.com/newsguidelines.html.
Fair: it was rude. Moderation is hard and I respect what you do. But it's also a sentiment several other comments expressed. It's the conversation we're having. Can we have any discussions of code quality without making assumptions about each others' code quality? I mean, yeah, I could probably have done better.
> "That would probably be 1000 line of Common Lisp." https://news.ycombinator.com/item?id=44974495
> "Perhaps the issue is you were used to writing 200k lines of code. Most engineers would be agast at that." https://news.ycombinator.com/item?id=44976074
> "200k lines of code is a failure state ... I'd not normally huff my own farts in public this obnoxiously, but I honestly feel it is useful for the "AI hater vs AI sucker" discussion to be honest about this type of emotion." https://news.ycombinator.com/item?id=44976328
[flagged]
I understand the provocation, but please don't respond to a bad comment by breaking the site guidelines yourself. That only makes things worse.
Your GP comment was great, and probably the thing to do with a supercilious reply is just not bother responding (which is easier said than done of course). You can usually trust other users to assess the thread fairly (e.g.https://news.ycombinator.com/item?id=44975623).
https://news.ycombinator.com/newsguidelines.html
> What makes you think I'm not "a developer who strongly values brevity and clarity"
Some pieces of evidence that make me think that:
1. The base rate of developers who write massively overly verbose code is about 99%, and there's not a ton of signal to deviate from that base rate other than the fact that you post on HN (probably a mild positive signal).
2. An LLM writes 80% of your code now, and my prior on LLM code output is that it's on par with a forgetful junior dev who writes very verbose code.
3. 200K lines of code is a lot. It just is. Again, without more signal, it's hard to deviate from the base rate of what 200K-line codebases look like in the wild. 99.5% of them are spaghettified messes with tons of copy-pasting and redundancy and code-by-numbers scaffolded code (and now, LLM output).
This is the state of software today. Keep in mind the bad programmers who make verbose spaghettified messes are completely convinced they're code-ninja geniuses; perhaps even more so than those who write clean and elegant code. You're allowed to write me off as an internet rando who doesn't know you, of course. To me, you're not you, you're every programmer who writes a 200k LOC B2B SaaS application and uses an LLM for 80% of their code, and the vast, vast majority of those people are -- well, not people who share my values. Not people who can code cleanly, concisely, and elegantly. You're a unicorn; cool beans.
Before you used LLMs, how often were you copy/pasting blocks of code (more than 1 line)? How often were you using "scaffolds" to create baseline codefiles that you then modified? How often were you copy/pasting code from Stack Overflow and other sources?
At least to me what you said sounded like 200k is just with LLMs but before agents. But it's a very reasonable amount of code for 9 years of work.
This is such a bizarre comment. You have no idea what code base they are talking about, their skill level, or anything.
200kloc is what, 4 reams of paper, double sided? So, 10% of that famous Margaret Hamilton picture (which is roughly "two spaceships worth of flight code".) I'm not sure the intuition that gives you is good but at least it slots the raw amount in as "big but not crazy big" (the "9 years work" rather than "weekend project" measurement elsethread also helps with that.)
What tooling are you using?
I use aider and your description doesn't match my experience, even with a relatively bad-at-coding model (gpt-5). It does actually work and it does generate "good" code - it even matches the style of the existing code.
Prompting is very important, and in an existing code base the success rate is immensely higher if you can hint at a specific implementation - i.e. something a senior who is familiar with the codebase somewhat can do, but a junior may struggle with.
It's important to be clear eyed about where we are here. I think overall I am still faster doing things manually than iterating with aider on an existing code base, but the margin is not very much, and it's only going to get better.
Even though it can do some work a junior could do, it can't ever replace a junior human... because a junior human also goes to meetings, drives discussions, and eventually becomes a senior! But management may not care about that fact.
I agree. AI is a wonderful tool for making fuzzy queries on vast amounts of information. More and more I'm finding that Kagi's Assistant is my first stop before an actual search. It may help inform me about vocabulary I'm lacking which I can then go successfully comb more pages with until I find what I need.
But I have not yet been able to consistently get value out of vibe coding. It's great for one-off tasks. I use it to create matplotlib charts just by telling it what I want and showing it the schema of the data I have. It nails that about 90% of the time. I have it spit out close-ended shell scripts, like recently I had it write me a small CLI tool to organize my Raw photos into a directory structure I want by reading the EXIF data and sorting the images accordingly. It's great for this stuff.
But anything bigger it seems to do useless crap. Creates data models that already exist in the project. Makes unrelated changes. Hallucinates API functions that don't exist. It's just not worth it to me to have to check its work. By the time I've done that, I could have written it myself, and writing the code is usually the most pleasurable part of the job to me.
I think the way I'm finding LLMs to be useful is that they are a brilliant interface to query with, but I have not yet seen any use cases I like where the output is saved, directly incorporated into work, or presented to another human that did not do the prompting.
Have you tried Opus? It's what got me past using LLMs only marginally. Standard disclaimers apply in that you need to know what it's good for and guide it well, but there's no doubt at this point it's a huge productivity boost, even if you have high standards - you just have to tell it what those standards are sometimes.
Opus was also the threshold for me where I started getting real value out of (correctly applied) LLMs for coding.
The one thing I've found AI is good at is parsing through the hundreds of ad ridden, barely usable websites for answers to my questions. I use the Duck Duck Go AI a lot to answer questions. I trust it about as far as I can throw the datacenter it resides in, but it's useful for quickly verifiable things. Especially stuff like syntax and command line options for various programs.
> The one thing I've found AI is good at is parsing through the hundreds of ad ridden, barely usable websites for answers to my questions.
One thing I can guarantee you is that this won't last. No sane MBA will ignore that revenue stream.
Image hosting services, all over again.
You are entirely correct. The enshittification will continue. All we can do is enjoy these things while they are still usable.
Nope, this only applies to a small percent of content, where a relatively small number of people needs access to it and the incentive to create derivative work based on it is low, or where there's a huge amount of content that's frequently changing (think airfares). But yes, they will protect it more.
For content that doesn't change frequently and is used by a lot of people it will be hard to control access to it or derivative works based on it.
I don't think you're considering the enshittification route here. I'm sure it will be: Ask ChatGPT a question -> "While I'm thinking, here's something from our sponsor which is tailored to your question" -> lame answer which requires you to ask another question. And on and on. While you're asking these questions, a profile of you is built and sold on the market.
It's even worse. "Native advertising":
Which car should I buy?
"The Toyota DZx4 is the best EV on the market according to multiple analysts. It had the following benefits: ...
If the DZx4 is out of your budget, the Nissan Avenger is a great budget option: ..."
Each spot the result of an automated live auction.
Now imagine that for everything and also some suggestions along the way ("If you need financing, Cash Direct offers same day loans...").
Advertising with LLMs will be incredibly insidious and lucrative. And most likely, unblockable.
It's one of those you get what you put in kind of deals.
If you spend a lot of time thinking about what you want, describing the inner workings, edge cases, architecture and library choices, and put that into a thoughtful markdown, then maybe after a couple of iterations you will get half decent code. It certainly makes a difference between that and a short "implement X" prompt.
But it makes one think - at that point (writing a good prompt that is basically a spec), you've basically solved the problem already. So LLM in this case is little more than a glorified electric typewriter. It types faster than you, but you did most of the thinking.
Right, and then after you do all the thinking and the specs, you have to read and understand and own every single line it generated. And speaking for myself, I am no where near as good at thinking through code I am reviewing as thinking through the code I am writing.
Other people will put up PRs full of code they don't understand. I'm not saying everyone who is reporting success with LLMs are doing that, but I hear it a lot. I call those people clowns, and I'd fire anyone who did that.
If it passes the unit tests I make it write and works for my sample manual cases I absolutely will not spend time reading the implementation details unless and until something comes up. Sometimes garbage makes its way into git but working code is better than no code and the mess can be cleaned up later. If you have correctness at the interface and function level you can get a lot done quickly. Technical debt is going to come out somewhere no matter what you do.
If AI is writing the code and the unit tests, how do you really know its working? Who watches the watchman
The trick is to not give a fuck. This works great in a lot of apps, which are useless to begin with. It may also be a reasonable strategy in an early-stage startup yet to achieve product-market fit, but your plan has to be to scrap it and rewrite it and we all know how that usually turns out.
This is an excellent point. Sure in an ideal world we should care very much about every line of code committed, but in the real world pushing garbage might be a valid compromise given things like crunch, sales pitches due tomorrow etc.
No, that's a much stronger statement. I'm not talking about ideals. I'm talking about running a business that is mature, growing and going to be around in five years. You could literally kill such a business running it on a pile of AI slop that becomes unmaintainable.
The bigger the task, the more messy it'll get. GPT5 can write a single UI component for me no problem. A new endpoint? If it's simple, no problem. The risk increases as the complexity of the task does.
I break complex task down into simple tasks when using ChatGPT just like I did before ChatGPT with modular design.
I agree with the article but also believe LLM coding can boost my productivity and ability to write code over long stretches. Sure getting it to write a whole feature, high opportunity of risk. But getting it to build out a simple api with examples above and below it, piece of cake, takes a few seconds and would have taken me a few minutes.
It's just that being the dumbest thing we ever heard still doesn't stop some people from doing it anyway. And that goes for many kinds of LLM application.
Likewise with Powershell. It's good to give you an approach or some ideas, but copy/paste fails about 80% of the time.
Granted, I may be a inexpert prompter, but at the same time, I'm asking for basic things, as a test, and it just fails miserably most of the time.
I haven't had that experience, but I tend to keep my prompts very focused with a tightly limited scope. Put a different way, if I had a junior or mid level developer, and I wanted them to create a single-purpose class of 100-200 lines at most, that's how I write my prompts.
I think it has a lot to do with skill level. Lower skilled developers seem to feel it gives them a lot of benefit. Higher skilled developers just get frustrated looking at all the errors in produces.
Which model?
Sounds like you are using it entirely wrong then...
Just yesterday I uploaded a few files of my code (each about 3000+ lines) into a gpt5 project and asked in assistance in changing a lot of database calls into a caching system, and it proceeded to create a full 500 line file with all the caching objects and functions I needed. Then we went section through section of the main 3000+ line file to change parts of the database queries into the cached version. [I didn't even really need to do this, it basically detected everything I would need changing at once and gave me most of it, but I wanted to do it in smaller chunks so I was sure what was going on]
Could I have done this without AI? Sure.. but this was basically like having a second pair of eyes and validating what I'm doing. And saving me a bunch of time so I'm not writing everything from scratch. I have the base template of what I need then I can improve it from there.
All the code it wrote was perfectly clean.. and this is not a one off, I've been using it daily for the last year for everything. It almost completely replaces my need to have a junior developer helping me.
How large is that code-base overall? Would you be able to let the LLM look at the entirety of it without it crapping out?
It definitely sounds nice to go and change a few queries, but did it also consider the potential impacts in other parts of the source or in adjacent running systems? The query itself here might not be the best example, but you get what I mean.
I know, I don't understand what problems people are having with getting usable code. Maybe the models don't work well with certain languages? Works great with C++. I've gotten thousands of lines of clean compiling on the first try and obviously correct code from ChatGPT, Gemini, and Claude.
I've been assuming the people who are having issues are junior devs, who don't know the vocabulary well enough yet to steer these things in the right direction. I wouldn't say I'm a prompt wizard, but I do understand context and the surface area of the things I'm asking the llm to do.
From my experience the further you get from the sort of stuff that easily accessible on Stack Overflow the worse it gets. I've had few problems having an AI write out some minor python scripts, but yield severely poorer results with Unreal C++ code and badly hallucinate nonsense if asked in general anything about Unreal architecture and API.
Does the Unreal API change a bit over versions? I've noticed when asking to do a simple telnet server in Rust it was hallucinating like crazy but when I went to the documentation it was clear the api was changing a lot from version to version. I don't think they do well with API churn. That's my hypothesis anyway.
There would be significant changes from 4 to 5, but sadly I haven’t had any improvement if clarifying version.
You mean like it turned on Hibernate or it wrote some custom rolled in app cache layer?
I usually find these kinds of caching solutions to be extremely complicated (well the cache invalidating part) and I'm a bit curious what approach it took.
You mention it only updated a single file so I guess it's not using any updates to the session handling so either sticky sessions are not assumed or something else is going on. So then how do you invalidate the app level cache for a user across all machine instances? I have a lot of trauma from the old web days of people figuring this out so I'm really curious to hear about how this AI one shot it in a single file.
This is C# so basically just automatically detected that I had 4 object types I was working with that were being updated to the database that I want to keep in a concurrent dictionary type of cache. So it created the dictionaries for each object with the appropriate keys, created functions for each object type if I touch an object to get that one updated etc.
It created the function to load in the data, then the finalize where it writes to the DB what was touched and clears the cache.
Again- I'm not saying this is anything particularly fancy, but it did the general concept of what I wanted. Also this is all iterative; when it creates something I talk to it like a person to say "hey I want to actually load in all the data, even though we will only be writing what changed" and all that kind of stuff.
Also the bigger help wasn't really the creation of the cache, it was helping to make the changes and detect what needed to be modified.
End of the day even if I want to go a slightly different route of how it did the caching; it creates all the framework so I can simplify if needed.
A lot of times for me using this LLM approach is to get all the boilerplate out of the way.. sometimes just starting the process by yourself of something is daunting. I find this to be a great way to begin.
At least one CEO seems to get it. Anyone touting this idea of skipping junior talent in favor of AI is dooming their company in the long run. When your senior talent leaves to start their own companies, where will that leave you?
I’m not even sure AI is good for any engineer, let alone junior engineers. Software engineering at any level is a journey of discovery and learning. Any time I use it I can hear my algebra teacher telling me not to use a calculator or I won’t learn anything.
But overall I’m starting to feel like AI is simply the natural culmination of US economic policy for the last 45 years: short term gains for the top 1% at the expense of a healthy business and the economy in the long term for the rest of us. Jack Welch would be so proud.
> When your senior talent leaves to start their own companies, where will that leave you?
The CEO didn't express any concerns about "talent leaving". He is saying "keep the juniors" but he's implying "fire the seniors". This is in line with long standing industry trends and it's confirmed by the flowing quote from the OP:
>> [the junior replacement] notion led to the “dumbest thing I've ever heard” quote, followed by a justification that junior staff are “probably the least expensive employees you have” and also the most engaged with AI tools.
He is pushing for more of the same, viewing competence and skill as threats and liability to be "fixed". He's warning the industry to stay the course and keep the dumbing-down game moving as fast possible.
I think AI has overall helped me learn.
There are lots of personal projects that I have wanted to build for years but have pushed off because the “getting started cost” is too high, I get frustrated and annoyed and don’t get far before giving up. Being able to get the tedious crap out of the way lowers the barrier to entry and I can actually do the real project, and get it past some finish line.
Am I learning as much as I would had I powered through it without AI assistance? Probably not, but I am definitely learning more than I would if I had simply not finished (or even started) the project at all.
What was your previous approach? From what I've seen, a lot of people are very reluctant about picking a book or read through a documentation before they try stuff. And then they got exposed to "cryptic" error message and then throw the towel.
The biggest blocker for me would be that I would go through a "Getting Started" guide and that would go well until it doesn't. Either there would be an edge case that the guide didn't take into account or the guide would be out of date. Sometimes I would get an arcane error message that was a little difficult to parse.
There were also cases where the interesting part of what I'm working on (e.g. something with distributed computing) required a fair amount of stuff that I don't find interesting to get started (e.g. spinning up a Kafka and Zookeeper cluster), where I might have to spend hours screwing around with config files and make a bunch of mistakes before I get something more or less working.
If I was sufficiently interested, I would power through the issue, by either more thoroughly reading through documentation or searching through StackOverflow or going onto a project IRC, so it's not like I would never finish a project, but having a lower barrier of entry by being able to directly paste an error message or generate a working starter config helps a lot with getting past the initial hump, especially to get to the parts that I find more interesting.
> At least one CEO seems to get it.
> (…)
> I’m not even sure AI is good for any engineer
In that case I’m not sure you really agree with this CEO, who is all-in on the idea of LLMs for coding, going so far as to proudly say 80% of engineers at AWS use it and that that number will only rise. Listen to the interview, you don’t even need ten minutes.
You also risk senior talent who stay but doesn't want to change or adopt, at least with any urgency. AI will accelerate that journey of discovery and learning, so juniors are going to learn super fast.
That’s still to be determined. Blindly accepting code suggestions thrown at you without understanding them is not the same thing as learning.
In the last few months we have worked with startups who have vibe coded themselves into an abyss. Either because they never made the correct hires in the first place or they let technical talent go. [1]
The thinking was that they could iterate faster, ship better code, and have an always on 10x engineer in the form of Claude code.
I've observed perfectly rational founders become addicted to the dopamine hit as they see Claude code output what looks like weeks or years of software engineering work.
It's overgenerous to allow anyone to believe AI can actually "think" or "reason" through complex problems. Perhaps we should be measuring time saved typing rather than cognition.
[1] vibebusters.com
He wants educators to instead teach “how do you think and how do you decompose problems”
Ahmen! I attend this same church.
My favorite professor in engineering school always gave open book tests.
In the real world of work, everyone has full access to all the available data and information.
Very few jobs involve paying someone simply to look up data in a book or on the internet. What they will pay for is someone who can analyze, understand, reason and apply data and information in unique ways needed to solve problems.
Doing this is called "engineering". And this is what this professor taught.
In undergrad I took an abstract algebra class. It was very difficult and one of the things the teacher did was have us memorize proofs. In fact, all of his tests were the same format: reproduce a well-known proof from memory, and then complete a novel proof. At first I was aghast at this rote memorization - I maybe even found it offensive. But an amazing thing happened - I realized that it was impossible to memorize a proof without understanding it! Moreover, producing the novel proofs required the same kinds of "components" and now because they were "installed" in my brain I could use them more intuitively. (Looking back I'd say it enabled an efficient search of a tree of sequences of steps).
Memorization is not a panacea. I never found memorizing l33t code problems to be edifying. I think it's because those kinds of tight, self-referential, clever programs are far removed from the activity of writing applications. Most working programmers do not run into a novel algorithm problem but once or twice a career. Application programming has more the flavor of a human-mediated graph-traversal, where the human has access to a node's local state and they improvise movement and mutation using only that local state plus some rapidly decaying stack. That is, there is no well-defined sequence for any given real-world problem, only heuristics.
Memorizing is a super power / skill. I work in a ridiculously complex environment and have to learn and know so much. Memorizing and spaced repetition are like little islands my brain can start building bridges between. I used to think memorizing was anti-first principles, but it is just good. Our brains can memorize so much if we make them. And then we can connect and pattern matching using higher order thinking.
There's this [1] which is one of my favorite articles on that topic; definitely worth a read.
[1] https://www.pearlleff.com/in-praise-of-memorization
Recognizing the patterns and applying patterned solutions is where I see success in my niche of healthcare interoperability. So much of my time is spent watching people do things,process and how they use data. It's amazing how much people remember to do their job, but me coming in and be able to bridge the doctor and the lab to share data easier is like Im an alchemist. It's really not a problem I've been able to see ai solve without suggesting solutions that are too simple or too costly and in that goldilocks zone everyone will be happy with
In education, I have heard it called “fluency”.
What's even better about memorization is that you have an objective method to test your own understanding. It is so easy to believe you understand something when you don't! But, at least with math, I think if you can reproduce the proof from memory you can be very confident that you aren't deluding yourself.
Hmmm... It's the other way around for me. I find it hard to memorise things I don't actually understand.
I remember being given a proof of why RSA encryption is secure. All the other students just regurgitated it. It made superficial sense I guess.
However, I could not understand the proof and felt quite stupid. Eventually I went to my professor for help. He admitted the proof he had given was incomplete (and showed me why it still worked). He also said he hadn't expected anyone to notice it wasn't a complete proof.
> Hmmm... It's the other way around for me. I find it hard to memorise things I don't actually understand.
I think you two are agreeing. GP said that they found they couldn't memorize something until they actually understood it
You're correct, I read it the wrong way!
Hope now you'll remember it :P
>> I realized that it was impossible to memorize a proof without understanding it!
> I find it hard to memorise things I don't actually understand.
Isn't it the parent's point?
Yep, I misread the parent post.
During my elementary school years, there was a teacher who told me that I didn't need to memorize it as long as I understand them. I taught he was the coolest guy ever.
Only when I got late twenties, I realized how wrong he was. Memorization and understanding go hand in hand, but if one of them has to come first than it's memorization. He probably said that because that was what kids (who were forced to do rote memorization) wanted to hear.
You could argue this is just moving the memorization to meta-facts, but I found all throughout school that if you understand some slightly higher level key thing, memorization at the level you're supposed to be working in becomes at best a slight shortcut for some things. You can derive it all on the fly.
Sort of like how most of the trigonometric identities that kids are made to memorize fall out immediately from e^iθ = cosθ+isinθ (could be taken as the definitions of cos,sin), e^ae^b=e^(a+b) (a fact they knew before learning trig), and a little bit of basic algebraic fiddling.
Or like how inverse Fourier transforms are just the obvious extension of the idea behind writing a 2-d vector as a sum of its x and y projections. If you get the 2d thing, accept that it works the exact same in n-d (including n infinite), accept integrals are just generalized sums, and functions are vectors, and I guess remember that e^iwt are the basis you want, you can reason through what the formula must be immediately.
> you can reason through what the formula must be immediately.
At least up to various factors of 2π ;-)
Probably. I hated memorization when I was a student too, because it was boring. But as soon as I did some teaching, my attitude changed to, "Just memorize it, it'll make your life so much easier." It's rough watching kids try to multiply when they don't have their times tables memorized, or translate a language when they haven't memorized the vocabulary words in the lesson so they have to look up each one.
There's things that you need to know (2*2 = 4) and there are things that you need to understand (multiplication rules). Both can happen with practice, but they're not that related.
Memorization is more like a shortcut. You don't need to go through the problem solving process to know the result. But with understanding, you master the heuristic factors needed to know when to take the shortcut and when to go through the problem solving route.
The Dreyfus Skill Model [0] is a good explanation. Novice typically have to memorize, then as they master the subject, their decision making becomes more heuristic based.
LLMs don't do well with heuristics, and by the times you've nailed down all the problems data, you could have been done. What they excels at is memorization, but all the formulaic stuff have been extracted into frameworks and libraries for the most popular languages.
[0]: https://en.wikipedia.org/wiki/Dreyfus_model_of_skill_acquisi...
> It's rough watching kids try to multiply when they don't have their times tables memorized
As someone who never learned my multiplication tables – it’s fine. I have a few cached lookups and my brain is fast at factoring.
8*6? Oh that’s just 4*2*6= 4*12 = 48. Easy :)
Smart people can get away with tricks like that.
It's a lot easier to memorize things when it's your job i find.
maybe they pay hits a reward centre in my brain.
I never memorized multiplication tables and was always one of those "good in math" kids. An attempt to memorize that stuff ended with me confusing results and being unable to guess when I did something wrong. Knowing "tricks" and understanding how multiplication works makes life easier.
> "Just memorize it, it'll make your life so much easier."
That is because you evaluate cost of memorization to 0, because someone else is paying it. And you evaluate the cost of making mistakes due to constantly forgetting and being unable to correct to 0, because simply the kid gets blamed for not having perfect memory.
> or translate a language when they haven't memorized the vocabulary words in the lesson so they have to look up each one
Teaching language by having people translate a lot is an outdated pedagogy - it simply did not produced people capable to understand and produce the language. If the kids are translating sentences word by word, there was something going on wrongly before.
I think we memorize the understanding. For me it also works better understanding how something works than memoryzing results. I remember in high school, in maths trigonometrics, there were a list of 20 something formulas derived from a single one. Everyboby was memorizing the whole list of formulas; i just had to memorize a simple formula and the underdtanding of how to derive the others from the fundamental one on the fly.
Probably the best example of this I can think of (for me at least) from mathematics is calculating combinations. I have it burned into my memory that (n choose r) = (n permute r) / (r permute r), and (n permute r) = n! / (n - r)!
Can I derive these from first principles? Sure, but after not seeing it for years, it might take me 10+ minutes to think through everything and correct any mistakes I make in the derivation.
But if I start with the formula? Takes me 5 seconds to sanity check the combination formula, and maybe 20 to sanity check the permutation formula. Just reading it to myself in English slowly is enough because the justification kind of just falls right out of the formula and definition.
So, yeah, they go hand in hand. You want to understand it but you sure as heck want to memorize the important stuff instead of relying on your ability to prove everything from ZFC...
As with most things, it depends. If you truly do understand something, then you can derive a required result from first principles. _Given sufficient time_. Often in an exam situation you are time-constrained, and having memorized a shortcut cut be beneficial. Not to mention retaining is much easier when you understand the topic, so memorization becomes easier.
Ish? I never ever memorized the multiplication tables. To this day, I don't think I know them fully. I still did quite well in math by knowing how to quiz the various equations. Not just know them, but how to ask questions about moving terms and such.
You don't need to memorize to understand. You can rederive it every time.
You need to memorize it to use it subconsciously while solving more complex problems. Other ways you won't fit more complex solutions into your working memory,vso whole classes of problems will be too hard for you.
It is waaaay easier to remember when you understand. The professor had it exactly right - if you learn to understand, you frequently end up remembering. But, memorization does not lead to understanding at all.
My controversial education hot take: Pointless rote memorization is bad and frustrating, but early education could use more directed memorization.
As you discovered: A properly structured memorization of carefully selected real world material forces you to come up with tricks and techniques to remember things. With structured information (proofs in your case) you start learning that the most efficient way to memorize is to understand, which then reduces the memorization problem into one of categorizing the proof and understanding the logical steps to get from one step to another. In doing so, you are forced to learn and understand the material.
Another controversial take (for HN, anyway) is that this is what happens when programmers study LeetCode. There’s a meme that the way to interview prep is to “memorize LeetCode”. You can tell who hasn’t done much LeetCode interviewing if they think memorizing a lot of problems is a viable way to pass interviews. People who attempt this discover that there are far too many questions to memorize and the best jobs have already written their own questions that aren’t out of LeetCode. Even if you do get a direct LeetCode problem in an interview, a good interview will expect you to explain your logic, describe how you arrived at the solution, and might introduce a change if they suspect you’re regurgitating memorized answers.
Instead, the strategy that actually works is to learn the categories of LeetCode style questions, understand the much smaller number of algorithms, and learn how to apply them to new problems. It’s far easier to memorize the dozen or so patterns used in LeetCode problems (binary search, two pointers, greedy, backtracking, and so on) and then learn how to apply those. By practicing you’re not memorizing the specific problems, you’re teaching yourself how to apply algorithms.
Side note: I’m not advocating for or against LeetCode, I’m trying to explain a viable strategy for today’s interview format.
Exactly. I agree with the leetcode part. A lot of problems in the world are composite of simpler smaller problems. Leetcode should teach you the basic patterns and how to combine them to solve real world problems. How will you ever solve a real world problem without knowing a few algorithms beforehand. For example, my brother was talking about how a Roomba would map a room. He was imagining 0 to represent free space and 1 as inaccessible points. This quickly reminded me of Number of Islands problem from leetcode. Yeah, there might be a lot of changes required to that problem but one could simple represent it as two problems.
1. Represent different objects in the room as some form of machine understandable form in a matrix 2. Find the number of Islands or find the Islands themselves.
Memorization of, like, multiplication tables gives us a poor view of the more interesting type of memorization. Remembering types of problems we’ve seen. Remembering landmarks and paths, vs just remembering what’s in every cell of a big grid.
I still don’t like leetcode, though.
> Memorization of, like, multiplication tables gives us a poor view of the more interesting type of memorization.
Memorizing multiplication tables is the first place many children encounter this strategy: The teacher shows you that you could try to memorize all of the combinations, or you could start learning some of the patterns and techniques. When multiplying by 5 the answer will end in 0 or 5. When multiplying by 2 the answer will be an even number, and so on.
I think there may have been a miscommunication somewhere on the chain of Mathematicians-Teachers-Students if that was the plan, when I was in elementary school.
Anecdotally (I only worked with math students as a tutor for a couple years), that math requires a lot of the boring type of memorization seems to be a really widespread misunderstanding.
My abstract algebra class had it exactly backwards. It started with a lot of needless formalism culminating in galois theory. This was boring to most students as they had no clue why the formalism was invented in the first place.
Instead, I wished it showed how the sausage was actually made in the original writings of galois [1]. This would have been far more interesting to students, as it showed the struggles that went into making the product - not to mention the colorful personality of the founder.
The history of how concepts were invented for the problems faced is far more motivating to students to build a mental model than canned capsules of knowledge.
[1] https://www.ams.org/notices/201207/rtx120700912p.pdf
> This was boring to most students as they had no clue why the formalism was invented in the first place.
> The history of how concepts were invented for the problems faced is far more motivating to students to build a mental model than canned capsules of knowledge.
That's something I really like about 3blue1brown, and he says it straight up [0]:
> My goal is for you to come away feeling like you could have invented calculus yourself. That is, cover all those core ideas, but in a way that makes clear where they actually come from, and what they really mean, using an all-around visual approach.
[0]: https://www.youtube.com/watch?v=WUvTyaaNkzM
Interesting I had the same problem and suffered in grades back in school simply because I couldn't memorize much without understanding. However, I seemed to be the only one because every single other student, including those with top grades, were happy to memorize and regurgitate. I wonder how they're doing now.
I would wager some folks can memorize without understanding? I do think memorization is underrated, though.
There is also something to the practice of reproducing something. I always took this as a form of "machine learning" for us. Just as you get better at juggling by actually juggling, you get better at thinking about math by thinking about math.
Rote memorization is essentially that, yes.
Depends on the subject - I can remember multiple subjects where the teacher would give you a formula to memorise without explaining why or where it came from. You had to take it as an axiom. The teachers also didn't say - hey, if you want to know why did we arrive to this, have a read here, no, it was just given.
Ofc you could also say that's for the student to find out, but I've had other things on my mind
Fortunately that was not my experience in abstract algebra. The tests and homework were novel proofs that we hadn't seen in class. It was one of my favorite classes / subjects. Someone did tell me in college that they did the memorization thing in German Universities.
Code-wise, I spent a lot of time in college reading other people's code. But no memorization. I remember David Betz advsys, Tim Budd's "Little Smalltalk", and Matt Dillon's "DME Editor" and C compiler.
>Memorization is not a panacea.
It is What you memorize that is important, you can't have a good discussion about a topic if you don't have the facts and logic of the topic in memory. On the other hand using memory to paper over bad design instead of simplifying or properly modularizing it, leads to that 'the worst code I have seen is code I wrote six months ago' feeling.
Is it the memorisation that had the desired effect or the having to come up with the novel proofs? Many schools seem to do the memorising part, but not the creating part.
Your comment about memorizing as part of understanding makes a lot of sense to me, especially as one possible technique to get get unstuck in grasping a concept.
If it doesn’t work for you on l33t code problems, what techniques are you finding more effective in that case?
I was part of an ACM programming team in college. We would review classes of problems based on the type of solution necessary, and learn those techniques for solving them. We were permitted a notebook, and ours was full of the general outline of each of these classes and techniques. Along with specific examples of the more common algorithms we might encounter.
As a concrete example, there is a class of problems that are well served by dynamic programming. So we would review specific examples like Dijkstra's algorithm for shortest path. Or Wagner–Fischer algorithm for Levenshtein-style string editing. But we would also learn, often via these concrete examples, of how to classify and structure a problem into a dynamic programming solution.
I have no idea if this is what is meant by "l33t code solutions", but I thought it would be a helpful response anyway. But the bottom line is that these are not common in industry, because hard computer science is not necessary for typical business problems. The same way you don't require material sciences advancements to build a typical house. Instead it flows the other way, where advancements in materials sciences will trickle down to changing what the typical house build looks like.
>If it doesn’t work for you on l33t code problems, what techniques are you finding more effective in that case?
Memorization of l33t code DOES work well as prep for l33t code tests. I just don't think l33t code has much to do with application programming. I've long felt that "computer science" is physics for computers, low on the abstraction ladder, and there are missing labels for the higher complexity subjects built on it. Imagine if all physical sciences were called "physics" and so in order to get a job as a biologist you should expect to be asked questions about the Schroedinger equation and the standard model. We desperately need "application engineering" to be a distinct subject taught at the university level.
You mean like to Software Engineering?
That's a real major that's been around for a couple of decades which focuses on software development (testing, version control, design patterns) with less focus on the more theoretical parts of computer science? There are even specialties within the Software Engineering major that focus specifically on databases or embedded systems.
What I understand from the GP is that memorizing l33t code won't help you learn anything useful. Not that understanding the solutions won't help you memorize them.
I find it's helpful to have context to frame what I'm memorizing to help me understand the value.
Indeed, not just math. Biology requires immense amounts of memorization. Nature is littered with exceptions.
> But an amazing thing happened - I realized that it was impossible to memorize a proof without understanding it!
This may be true of mathematical proofs, but it surely must not be true in general. Memorizing long strings of digits of pi probably isn’t much easier if you understand geometry. Memorizing famous speeches probably isn’t much easier if you understand the historical context.
> Memorizing famous speeches probably isn’t much easier if you understand the historical context.
Not commenting on the merits of critical thinking vs memorization either way, but I think it would be meaningfully easier to memorize famous speeches if you understand the historical context.
Right, isn't building up a (imaginary) context how people memorize pi?
It's funny, because I had the exact opposite experience with abstract algebra.
The professor explained things, we did proofs in class, we had problem sets, and then he gave us open-book semi-open-professor take-home exams that took us most of a week to do.
Proof classes were mostly fine. Boring, sometimes ridiculously shit[0], but mostly fine. Being told we have a week for this exam that will kick our ass was significantly better for synthesizing things we'd learned. I used the proofs we had. I used sections of the textbook we hadn't covered. I traded some points on the exam for hints. And it was significantly more engaging than any other class' exams.
[0] Coming up with novel things to prove that don't require some unrelated leap of intuition that only one student gets is really hard to do. Damn you Dr. B, needing to figure out that you have to define a third equation h(x) as (f(x) - g(x))/(f(x) + g(x)) as the first step of a proof isn't reasonable in a 60 minute exam.
memorization + application = comprehension. Rinse and repeat.
Whether leet code or anything else.
Mathematics pedagogy today is in a pretty sorrowful state due to bad actors and willful blindness at all levels that require public trust.
A dominant majority in public schools starting late 1970s seems to follow the "Lying to Children" approach which is often mistakenly recognized as by-rote teaching but are based in Paulo Freire's works that are in turn based on Mao's torture discoveries from the 1950s.
This approach contrary to classical approaches leverages torturous process which seems to be purposefully built to fracture and weed out the intelligent individual from useful fields, imposing sufficient thresholds of stress to impose PTSD or psychosis, selecting for and filtering in favor of those who can flexibly/willfully blind/corrupt themselves.
Such sequences include Algebra->Geometry->Trigonometry where gimmicks in undisclosed changes to grading cause circular trauma loops with the abandonment of Math-dependent careers thereafter, similar structures are also found in Uni, for Economics, Business, and Physics which utilize similar fail-scenarios burning bridges where you can't go back when the failure lagged from the first sequence, and you passed the second unrelated sequence. No help occurs, inducing confusion and frustration to PTSD levels, before the teacher offers the Alice in Wonderland Technique, "If you aren't able to do these things, perhaps you shouldn't go into a field that uses it". (ref Kubark Report, Declassified CIA Manual)
Have you been able to discern whether these "patterns" as you've called them aren't just the practical reversion to the classical approach (Trivium/Quadrivium)? Also known as the first-principles approach after all the filtering has been done.
To compare: Classical approaches start with nothing but a useful real system and observations which don't entrench false assumptions as truth, which are then reduced to components and relationships to form a model. The model is then checked for accuracy against current data to separate truth from false in those relationships/assertions in an iterative process with the end goal being to predict future events in similar systems accurately. The approach uses both a priori and a posteriori components to reasoning.
Lying to Children reverses and bastardizes this process. It starts with a single useless system which contains equal parts true and false principles (as misleading assumptions) which are tested and must be learned to competency (growing those neurons close together). Upon the next iteration one must unlearn the false parts while relearning the true parts (but we can't really unlearn, we can only strengthen or weaken) which in turn creates inconsistent mental states imposing stress (torture). This is repeated in an ongoing basis often circular in nature (structuring), and leveraging psychological blindspots (clustering), with several purposefully structured failings (elements) to gatekeep math through torturous process which is the basis for science and other risky subject matter. As the student progresses towards mastery (gnosis), the systems become increasingly more useful. One must repeatedly struggle in their sessions to learn, with the basis being if you aren't struggling you aren't learning. This mostly uses a faux a priori reasoning without properties of metaphysical objectivity (tied to objective measure, at least not until the very end).
If you don't recognize this, an example would be the electrical water pipe pressure analogy. Diffusion of charge in-like materials, with Intensity (Current) towards the outermost layer was the first-principled approach pre-1978 (I=V/R). The Water Analogy fails when the naive student tries to relate the behavior to pressure equations that ends up being contradictory at points in the system in a number of places introducing stumbling blocks that must be unlearned.
Torture being the purposefully directed imposition of psychological stress beyond a individuals capacity to cope towards physiological stages of heightened suggestability and mental breakdown (where rational thought is reduced or non-existent in the intelligent).
It is often recognized by its characteristic subgroups of Elements (cognitive dissonance, a lack of agency to remove oneself and coercion/compulsion with real or perceived loss or the threat thereof), Structuring (circular patterns of strictness followed by leniency in a loop, fractionation), and Clustering (psychological blindspots).
Wait, the electrical pipe water analogy is actually a very good one and it's quite difficult to find edge cases where it breaks down in a way that would confuse a student. There are some (for example, there's no electrical equivalent of Reynold's number or turbulence, and flow resistance varies differently with pipe diameter than wire diameter, and no good equivalent for Faraday's law) but I don't think these are likely to cause confusion. It even captures nuance like inductance, capacitance, and transmission line behaviour.
As I recall, my systems dynamics textbook even explicitly drew parallels between different domains like electricity and hydrodynamics. You're right that the counterparts aren't generally perfect especially at the edges but the analogies are often pretty good.
Intuitively it fails in making an equivalence to area which is an unrelated dimensional unit, as two lengths multiplied together equaling resistance, as well as the skin-effect related to Intensity/Current which is why insulation/isolation of wires are incredibly important.
The classical approach used charge diffusion iirc, and you can find classical examples of this in Oliver Heaviside's published works (archive.org iirc). He's the one that simplified Maxwell's 20+ equations down to the small number we use today.
> Lying to Children reverses and bastardizes this process. It starts with a single useless system which contains equal parts true and false principles (as misleading assumptions) which are tested and must be learned to competency (growing those neurons close together).
Can you provide some concrete examples of it?
I'd love an example too, and an example of the classical system that this replaced. I'm willing to believe the worst of the school system, but I'd like to understand why.
The classical system was described, but you can find it in various historic works based on what's commonly referred to today as the Trivium and Quadrivium based curricula.
Off the top of my head, the former includes reasoning under dialectical (priori and later posteriori parts under the quadrivium).
Its a bit much to explain it in detail in a post like this but you should be able to find sound resources with what I've provided.
It largely goes back to how philosophy was taught; all the way back to Socrates/Plato/Aristotle, up through Descartes, Locke (barely, though he's more famous for social contract theory), and more modern scientists/scientific method.
The way math is taught today, you basically get to throw out almost everything you were taught at various stages, and relearn it anew on a different foundation, somehow fitting the fractured pieces back together towards learning the true foundations, which would be much easier at the start and building on top of that instead of the constant interference.
You don't really end up understanding math intuitively nor its deep connections to logic (dialectics, trivium), until you hit Abstract Algebra.
You want to teach abstract algebra to middle schoolers?
Not OP, and it was a couple decades ago, but I certainly remember professors and teachers saying things like "this isn't really how X works, but we will use the approximation for now in order to teach you this other thing". That is if you were lucky, most just taught you the wrong (or incomplete) formula.
I think there is validity to the approach but sciences would be much, much improved if taught more like history lessons. Here is how we used to think about gravity, here's the formula and it kind of worked, except... Here is planetary orbits that we used to use when we assumed they had to be circles. Here's how the data looked and here's how they accounted for it...
This would accomplish two goals - learning the wrong way for immediate use (build on sand) and building an innate understanding of how science actually progresses. Too little focus is on how we always create magic numbers and vague concepts (dark matter, for instance) to account for structural problems we have no good answer for.
Being able to "sniff the fudge" would be a super power when deciding what to write a PhD on, for instance. How much better would science be if everyone strengthened this muscle throughout their educatuon?
I included the water pipe analogy for electric theory, that is one specific example.
Also, In Algebra I've seen a flawed version of mathematical operations being taught that breaks down with negative numbers under multiplication (when the correct way is closed over multiplication). The tests were supposedly randomized (but seemed to target low-income demographics). The process is nearly identical, but the answers ultimately not correct. The teachers graded on the work to the exclusion of the correct answer. So long as you showed the correct process expected in Algebra you passed without getting the right answer. Geometry was distinct and unrelated, and by Trigonometry the class required correct process and answer. You don't find out there is a problem until Trigonometry, and the teacher either doesn't know where the person is failing comprehension, or isn't paid to reteach a class they aren't paid for but you can't go back.
I've seen and heard horror stories of students where they'd failed Trig 7+ times at the college level, and wouldn't have progressed if not for a devoted teacher helping them after-hours (basically correcting and reteaching Algebra). These kids literally would break out in a cold PTSD sweat just hearing the associated words related to math.
I did some tutoring in a non-engineering graduate masters program and some folks were just lost. Simple things like what a graph is or how to solve an equation. I really did try but it's sort of hard to teach fairly easy high school algebra (with maybe some really simple derivatives to find maxima and minima) in grad school.
It's the core problem facing the hiring practices in this field. Any truly competent developer is a generalist at heart. There is value to be had in expertise, but unless you're dealing with a decade(s) old hellscape of legacy code or are pushing the very limits of what is possible, you don't need experts. You'd almost certainly be better off with someone who has experience with the tools you don't use, providing a fresh look and cover for weaknesses your current staff has.
A regular old competent developer can quickly pick up whatever stack is used. After all, they have to; Every company is their own bespoke mess of technologies. The idea that you can just slap "15 years of React experience" on a job ad and that the unicorn you get will be day-1 maximally productive is ludicrous. There is always an onboarding time.
But employers in this field don't "get" that. For regular companies they're infested by managers imported from non-engineering fields, who treat software like it's the assembly line for baking tins or toilet paper. Startups, who already have fewer resources to train people with, are obsessed with velocity and shitting out an MVP ASAP so they can go collect the next funding round. Big Tech is better about this, but has it's own problems going on and it seems that the days of Big Tech being the big training houses is also over.
It's not even a purely collective problem. Recruitment is so expensive, but all the money spent chasing unicorns & the opportunity costs of being understaffed just get handwaved. Rather spend $500,000 on the hunt than $50,000 on training someone into the role.
And speaking of collective problems. This is a good example of how this field suffers from having no professional associations that can stop employers from sinking the field with their tragedies of the commons. (Who knows, maybe unions will get more traction now that people are being laid off & replaced with outsourced workers for no legitimate business reason.)
I can’t think of another career where management continuously does not understand the realities of how something gets built. Software best practices are on their face orthogonal to how all other parts of a business operate.
How does marketing operate? In a waterfall like model. How does finance operate? In a waterfall like model. How does product operate? Well you can see how this is going.
Then you get to software and it’s 2 week sprints, test driven development etc. and it decidedly works best not on a waterfall model, but shipping in increments.
Yet the rest of the business does not work this way, it’s the same old top down model as the rest.
This I think is why so few companies or even managers / executives “get it”
> can’t think of another career where management continuously does not understand the realities of how something gets built
All engineering. Also all government and a striking amount of finance.
Actually, this might be a hallmark of any specialist field. Specialists interface with outsiders through a management layer necessarily less competent at the specialty than they are. (Since they’re devoting time and energy to non-specialty tasks.)
While product often does operate in a waterfall model, I think this is the wrong mindset. Good product management should adopt a lot of the same principles as software development. Form a testable hypothesis, work to get it into production and begin gathering data, then based on your findings determine what the next steps are and whether to adjust the implementation, consider the problem solved or try a different approach.
> I can’t think of another career where management continuously does not understand the realities of how something gets built.
This is in part a consequence of how young our field is.
The other comment pointing out other engineering is right here. The difference is that fields like Civil Engineering are millenia old. We know that Egyptian civil engineering was advanced and shockingly modern even 4.5 millenia ago. We've basically never stopped having qualified civil engineers around who could manage large civil engineering projects & companies.
Software Development in it's modern forms has it's start still in living memory. There simply weren't people to manage the young early software development firms as they grew, so management got imported from other industries.
And to say something controversial: Other engineering has another major reason why it's usually better understood. They're held to account when they kill people.
If you're engineering a building or most other things, you must meet safety standards. Where possible you are forced to prove you meet them. E.g. Cars.
You don't get to go "Well cars don't kill people, people kill people. If someone in our cars die when they're hit by a drunk driver, that's not our problem that's the drunkard's fault." No. Your car has to hold up to a certain level of crash safety, even if it's someone else who causes the accent, your engineering work damn better hold up.
In software, we just do not do this. The very notion of "Software kills people" is controversial. Treated as a joke, "of course it can't kill people, what are you on about?". Say, you neglect on your application's security. There's an exploit, a data breach, you leak your users' GPS location. A stalker uses the data to find and kill their victim.
In our field, the popular response is to go "Well we didn't kill the victim, the stalker did. It's not our problem.". This is on some level true; 'Twas the drunk driver who caused the car crash, not the car company. But that doesn't justify the car company selling unsafe cars, why should it justify us selling unsafe software? It may be but a single drop of blood, but it's still blood on our hands as well.
As it stands, we are fortunate enough that there haven't been incidents big enough to kill so many people that governments take action to forcibly change this mindset. It would be wise that Software Development takes up this accountability on it's own accord to prevent such a disaster.
> Rather spend $500,000 on the hunt than $50,000 on training someone into the role.
Capex vs opex, that's the fundamental problem at heart. It "looks better on the numbers" to have recruiting costs than to have to set aside a senior developer plus paying the junior for a few months. That is why everyone and their dog only wants to hire seniors, because they have the skillset and experience that you can sit their ass in front of any random semi fossil project and they'll figure it out on their own.
If the stonk analysts would go and actually dive deep into the numbers to look at hiring side costs (like headhunter expenses, employee retention and the likes), you'd see a course change pretty fast... but this kind of in-depth analysis, that's only being done by a fair few short-sellers who focus on struggling companies and not big tech.
In the end, it's a "tragedy of the commons" scenario. It's fine if a few companies do that, it's fine if a lot of companies do that... but when no one wants to train juniors any more (because they immediately get poached by the big ones), suddenly society as a whole has a real and massive problem.
Our societies are driven into a concrete wall at full speed by the financialization of every tiny aspect of our lives. All that matters these days are the gods of the stonk market - screw the economy, screw the environment, screw labor laws, all that matters is appearing "numbers go up" on the next quarterly.
> but when no one wants to train juniors any more (because they immediately get poached by the big ones)
Can we stop pretending that we don't know how to solve this problem? If you hire juniors at $X/year, but they keep getting poached after 2-3 years because now they can get $X*1.5/year (or more!), then maybe you should start promoting and giving raises to them after they've gotten a couple years experience.
Seriously, this is not a hard problem to solve. If the junior has proven themselves, give them the raise they deserve instead of being all Surprised Pikachu when another company is willing to pay them what they've proven themselves worthy of.
> Our societies are driven into a concrete wall at full speed by the financialization of every tiny aspect of our lives. All that matters these days are the gods of the stonk market - screw the economy, screw the environment, screw labor laws, all that matters is appearing "numbers go up" on the next quarterly.
I have been in the various nooks and crannies of the Internet/Software dev industry my whole career (i'm 49). I can't think of any time when the stock market didn't drive software innovation. It's always been either invent something -> go public -> exit or invent something -> increase stock price of existing public corp
> It's always been either invent something -> go public -> exit or invent something -> increase stock price of existing public corp
Yes, but today more and more is invent something -> achieve dominance -> get bought up by an even larger megacorp. That drives the enshittification circle.
> Capex vs opex
That's part of the problem, but I also notice the new hiring managers are incentivized to hire (or replace) employees to make their mark on the company. They then advocate for "their guys" the ones they recruited over the incumbents that are the unwilling dinosaurs in their eyes.
>For regular companies they're infested by managers imported from non-engineering fields
Someone's cousin, lets leave it at that, someones damn cousin or close friend, or anyone else with merely a pulse. I've had interviews where the company had just been turned over from people that mattered, and you. could. tell.
One couldn't even tell me why the project I needed to do for them ::rolleyes::, their own code boilerplate(which they said would run), would have runtime issues and I needed to self debug it to even get it to a starting point.
Its like, Manager: Oh heres this non-tangential thing that they tell me you need to complete before I can consider you for the positon.... Me: Oh can I ask you anything about it?.... Manager: No
That we talk about "building" software doesn't help.
Could not agree more. Whenever I hear monikers like "Java developer" or "python developer" as a job description I roll my eyes slightly.
Isn't that happening already? Half the usual CS curriculum is either math (analysis, linear algebra, numerical methods) or math in anything but name (computability theory, complexity theory). There's a lot of very legitimate criticism of academia, but most of the times someone goes "academia is stupid, we should do X" it turns out X is either:
- something we've been doing since forever
- the latest trend that can be picked up just-in-time if you'll ever need it
I've worked in education in some form or another for my entire career. When I was in teacher education in college . . . some number of decades ago . . . the number one topic of conversation and topic that most of my classes were based around was how to teach critical thinking, effective reasoning, and problem solving. Methods classes were almost exclusively based on those three things.
Times have not changed. This is still the focus of teacher prep programs.
Parent comment is literally praising an experience they had in higher education, but your only takeaway is that it must be facile ridicule of academia.
Was directed at TFA, not parent comment.
In CS, it's because it came out of math departments in many cases and often didn't even really include a lot of programming because there really wasn't much to program.
Right but a looot of the criticism online is based on assumptions (either personal or inherited from other commenters) that haven’t been updated since 2006.
Well, at more elite schools at least, the general assumption is that programming is mostly something you pick up on your own. It's not CS. Some folks will disagree of course but I think that's the reality. I took an MIT Intro to Algorithms/CS MOOC course a few years back out of curiosity and there was a Python book associated with the course but you were mostly on your own with it.
When I was in college the philosophy program had the marketing slogan: “Thinking of a major? Major in thinking”.
Now as a hiring manager I’ll say I regularly find that those who’ve had humanities experience are way more capable and the hard parts of analysis and understanding. Of course I’m biased as a dual cs/philosophy major but it’s very rare I’m looking for someone who can just write a lot of code. Especially juniors as analytical thinking is way harder to teach than how to program.
> Now as a hiring manager I’ll say I regularly find that those who’ve had humanities experience are way more capable and the hard parts of analysis and understanding.
The humanities, especially the classic texts, cover human interaction and communication in a very compact form. My favorite sources are the Bible, Cicero, and Machiavelli. For example Machiavelli says if you do bad things to people do them at once, while good things you should spread out over time. This is common sense. Once you catch the flavor of his thinking it's pretty easy to work other situations out for yourself, in the same why that good engineering classes teach you how to decompose and solve technical problems.
This is also why I went into the Philosophy major - knowing how to learn and how to understand is incredibly valuable.
Unfortunately in my experience, many, many people do not see it that way. It's very common for folks to think of philosophy as "not useful / not practical".
Many people hear the word "philosophy" and mentally picture "two dudes on a couch recording a silly podcast", and not "investigative knowledge and in-depth context-sensitive learning, applied to a non-trivial problem".
It came up constantly in my early career, trying to explain to folks, "no, I actually can produce good working software and am reasonably good at it, please don't hyper-focus on the philosophy major, I promise I won't quote Scanlon to you all day."
Maybe because phylosophy focuses on weird questions (to be or not to be) and weird personas. If it was advertised as more grounded thing, the views would be different.
The way you are perceived by others dependa on your behaviour. If you wamt to be perceived differently, adjust your behaviour, don't demand others to change. They won't.
How people see it is based on the probability of any philosophy major producing good working software, not you being able to produce good working software.
I worked for someone who I believe was undergrad philosophy and then got a masters in CS.
I would say you have some bias.
yes, sometimes you need people who can grasp the tech and talk to managers. They might be intermediaries.
But don't ignore the nerdy guys who have been living deeply in a tech ecosystem all their lives. The ones who don't dabble in everything. (the wozniaks)
The #1 problem in almost all workplaces is communication related. In almost all jobs I've had in 25-30 years, finding out what needs to be done and what is broken -- is much harder than actually doing it.
We have these sprint planning meetings and the like where we throw estimates on the time some task will take but the reality is for most tasks it's maybe a couple dozen lines of actual code. The rest is all what I'd call "social engineering" and figuring out what actually needs to be done, and testing.
Meanwhile upper management is running around freaking out because they can't find enough talent with X years of Y [language/framework] experience, imagining that this is the wizard power they need.
The hardest problem at most shops is getting business domain knowledge, not technical knowledge. Or at least creating a pipeline between the people with the business knowledge and the technical knowledge that functions.
Anyways, yes I have 3/4 a PHIL major and it actually has served me well. My only regret is not finishing it. But once I started making tech industry cash it was basically impossible for me to return to school. I've met a few other people over the years like me, who dropped out in the 90s .com boom and then never went back.
Yea this is why I’m generally not that impressed by LLMs. They still force you to do the communication which is the hard part. Programming languages are inherently a solve for communicating complex steps. Programming in English isn’t actually that much of a help you just have to reinvent how to be explicit
I find Claude code unexpectedly good at analysis. With a healthy dose of skepticism. It is actually really good at reading logs and corelating events for example.
On a related note, the most accomplished people I've met didn't have degrees in the fields where they excelled and won awards. They were all philosophy majors.
Teaching people to think is perhaps the world's most under-rated skill.
My otherwise ordinary school applied the mentality that students must "Learn to learn", and that mix of skills and mindset has never stopped helping me.
I think good historians are on the same foot as philosophers in the arena of "thinking really fucking hard and making an airtight analysis".
Well, yes but the other 90%+ just need to get a job out of college to support their addiction to food and shelter not to be a “better citizen of the world” unless they have parents to subsidize their livelihood either through direct transfers of money or by letting them stay at home.
I told both of my (step)sons that I would only help them pay for college or trade school - their choice - if they were getting a degree in something “useful”. Not philosophy, not Ancient Chinese Art History etc.
I also told them that they would have to get loans in their own names and I would help them pay off the loans once they graduated and started working gainfully.
A professor in my very first semester called "crazy finger syndrome" the attempts to go straight to the code without decomposing the problem from a business or user perspective. It was a long time ago. It was a CS curriculum
I miss her jokes against anxious nerds that just wanted to code :(
Don't forget the rise of boot camps where some educators are not always aligned with some sort of higher ethical standards.
> "crazy finger syndrome" - the attempts to go straight to the code without decomposing the problem from a business or user perspective
Years ago I started on a new team as a senior dev, and did weeks of pair programming with a more junior dev to intro me to the codebase. His approach was maddening; I called it "spray and pray" development. He would type out lines or paragraphs of the first thing that came to mind just after sitting down and opening an editor. I'd try to talk him into actually taking even a few minutes to think about the problem first, but it never took hold. He'd be furiously typing, while I would come up with a working solution without touching a keyboard, usually with a whiteboard or notebook, but we'd have to try his first. This was c++/trading, so the type-compile-debug cycle could be 10's of minutes. I kept relaying this to my supervisor, but after a few months of this he was let go.
I make a point to solve my more difficult problems with pen and paper drawings and/or narrative text before I touch the PC. The computer is an incredibly distracting medium to work with if you are not operating under clear direction. Time spent on this forum is a perfect example.
Speaking with a relative who is a doctor recently it’s interesting how much each of our jobs are “troubleshooting”.
Coding, doctors, plumber… different information, often similar skill sets.
I worked a job doing tech support for some enterprise level networking equipment. It was the late 1990s and we were desperate for warm bodies. Hired a former truck driver who just so happened to do a lot of woodworking and other things.
Great hire.
Memorization and closed book tests are important for some areas. When seconds are counting the ER doctor cannot go look up how to treat a heart attack. That doctor also needs to know now only how to treat the common heart attack, but how to recognize this isn't the common heart attack but the 1 in 10,000 not a heart attack but has exactly the same symptoms as a heart attack case and give it the correct treatment.
However most of us are not in that situation. It is better for us to just look up those details as we need them because it gives us more room to handle a broader variety of situations.
Humans will never outcompete ai in that regard however. Industry will eventually optimize for humans and ai separately: ai will know a lot and think quickly, humans will provide judgement and legal accountability. We’re already on this path.
My favorite professor in my physics program would say, "You will never remember the equations I teach. But if you learn how the relationships are built and how to ask questions of those relationships, then I have done my job." He died a few years ago. I never was able to thank him for his lessons.
You just did.
Being resourceful is an extremely valuable skill in the real world, and basically shut out of the education world.
Unlike my teachers, none of my bosses ever put me in an empty room with only a pencil and a sheet of paper to solve given problems.
You cannot teach "how to to think". You have to give students thinking problems to actually train thinking. Those kinds of problems can increasingly be farmed off to AI, or at least certain subproblems in them.
I meam, yes, to an extent you can teach how to think: critical thinking and logic are topics you can teach and people who take their teaching to heart can become better thinkers. However, those topics cannot impart creativity. Critical thinking is called exactly that because it's about tools and skills for separating bad thinking from good thinking. The skill of generating good thinking probably cannot be taught; it can only be improved with problem-solving practice.
During some of the earlier web service development days, one would find people at F500 skating by in low-to-mid level jobs just cutting and pasting between spreadsheets, things would take them hours could be done in seconds, and with lower error rates, with a proper data interface.
Very anecdotally, but I hazard that most of these types of low-hanging fruit, low-value add roles are much less common since they tended to be blockers for operational improvement. Six-sigma, Lean, various flavors of Agile would often surface these low performers up and they either improved or got shown the door between 2005 - 2020.
Not that everyone is 100% all the time, every day, but what we are left with is often people that are highly competent at not just their task list but at their job.
> In the real world of work, everyone has full access to all of the available data and information.
In general, I also attend your church.
However, as I preached in that church, I had two students over the years.
* One was from an African country and told me that where he grew up, you could not "just look up data that might be relevant" because internet access was rare.
* The other was an ex US Navy officer who was stationed on a nuclear sub. She and the rest of the crew had to practice situations where they were in an emergency and cut off from the rest of the world.
Memorization of considerable amounts of data was important to both of them.
Agreed coming from the ops world also.
I've had a frustrating experience the past few years trying to hire junior sysadmins because of a real lack of problem solving skills once something went wrong outside of various playbooks they memorized to follow.
I don't need someone who can follow a pre-written playbook, I have ansible for that. I need someone that understands theory, regardless of specific implementations, and can problem solve effectively so they can handle unpredictable or novel issues.
To put another way, I can teach a junior the specifics of bind9 named.conf, or the specifics of our own infrastructure, but I shouldn't be expected to teach them what DNS in general is and how it works.
But the candidates we get are the opposite - they know specific tools, but lack more generalized theory and problem solving skills.
People have been saying that since the advent of formal education. Turns out standardized education is really hard to pull off and most systems focus on making the average good enough.
It’s also hard to teach people “how to think” while at the same time teaching them practical skills - there’s only so many hours in a day, and most education is setup as a way to get as many people as possible into shape for taking on jobs where “thinking” isn’t really a positive trait, as it’d lead to constant restructuring and questioning of the status quo
> My favorite professor in engineering school always gave open book tests.
My experience as a professor and a student is that this doesn't make any difference. Unless you can copy verbatim the solution to your problem from the book (which never happens), you better have a good understanding of the subject in order to solve problems in the allocated time. You're not going to acquire that knowledge during your test.
My experience as a professor and a student is that this doesn't make any difference.
Exactly the point of his test methodology.
What he asked of students on a test was to *apply* knowledge and information to *unique* problems and create a solution that did not exist in any book.
I only brought 4 things to his tests --- textbook, pencil, calculator and a capable, motivated and determined brain. And his tests revealed the limits of what you could achieve with these items.
Isn't this an argument for why you should allow open book tests rather than why you shouldn't? It certainly removes some pressure to remember some obscure detail or formula.
Isn't that just an argument for always doing open book tests, then? Seems like there's no downside, and as already mentioned, it's closer to how one works in the real world.
I had a like minded professor in university, ironically in AI. Our big tests were all 3 day take home assignments. The questions were open ended, required writing code, processing data and analyzing results.
I think the problem with this is that it requires the professor to mentally fully engage when marking assignments and many educators do not have the capacity and/or desire to do so.
Sadly, I doubt 3-day take-home assignments have much future as a means of assessment in the age of LLMs.
Might be true, idk? For all we know that professor now gives a 2.5-day take home assignments where they are allowed to use LLMs, and then assess them in an 1 hour oral exam where they need to explain approach, results and how they ensure that their results are accurate?
I don't think the 3-day take home is the key. It's supporting educators to have the intention, agency and capacity to improvise assessment.
Same here! I always like to say that software engineering is 50% knowing the basics (How to write/read code, basic logic) and 50% having great research skills. So much of our time is spent finding documentation and understanding what it actually means as opposed to just writing code.
Agree, hopefully this insight / attitude will become more and more prevalent.
For anyone looking for resources, may we recommend:
* The Art of Doing Science and Engineering by Richard Hamming (lectures are available on YouTube as well)
* Measurement by Paul Lockhart (for teaching mindset)
Each one of us has a mental toolbox that we use to solve problems. There are many more tools that we don’t have in our heads that we can look up if we know how.
The bigger your mental toolbox the more effective you will be at solving the problems. Looking up a tool and learning just enough to use it JIT is much slower than using a handy tool that you already masterfully know how to use.
This is as true for physical tools as for programming concepts like algorithms and data structures. In the worst case you won’t even know to look for a tool and will use whatever is handy, like the proverbial hammer.
> Everyone has full access to all of the available data and information
Ahh, but this is part of the problem. Yes, they have access, but there is -so much- information, it punches through our context window. So we resort to executive summaries, or convince ourselves that something that's relevant is actually not.
At least an LLM can take full view of the context in aggregate and peel out signal. There is value there, but no jobs are being replaced
>but no jobs are being replaced
I agree that an LLM is a long way from replacing most any single job held by a human in isolation. However, what I feel is missed in this discussion is that it can significantly reduce the total manpower by making humans more efficient. For instance, the job of a team of 20 can now be done by 15 or maybe even 10 depending on the class of work. I for one believe this will have a significant impact on a large number of jobs.
Not that I'm suggesting anything be "stopped". I find LLM's incredibly useful, and I'm excited about applying them to more and more of the mundane tasks that I'd rather not do in the first place, so I can spend more time solving more interesting problems.
Also, some problems don't have enough data for a solution. I had a professor that gave tests where the answer was sometimes "not solvable." Taking these tests was like sweating bullets because you were not sure if you're just too dumb to solve the problem, or there was not enough data to solve the problem. Good times!
I agree with the overall message, but I will say that there is still a great deal of value in memorisation. Memorising things gives you more internal tools to think in broader chunks, so you can solve more complicated problems.
(I do mean memorisation fairly broadly, it doesn't have to mean reciting a meaningless list of items.)
Talk is cheap. Good educators cost money, and America famously underpays (and under-appreciates) its teachers. Does he also support increasing taxes on the wealthy?
Have there been studies about abilities of different students to memorize information? I feel this is under-studied in the world of memorizing for exams
Everyone going through STEM needs to see the movie Hidden Figures for a variety of reasons, but one bit stands out as poignant: I believe it was Katherine Johnson, who is asked to calculate some rocket trajectory to determine the landing coordinates, thinks on it a bit and finally says, "Aha! Newton's method!" Then she runs down to the library to look up how to apply Newton's method. She had the conceptual tools to find a solution, but didn't have all the equations memorized. Having all the equations in short term memory only matters in a (somewhat pathological) school setting.
It is tough though, I'd like to think I learnt how to think analytically and critically. But thinking is hard, and often times I catch myself trying to outsource my thinking almost subconsciously. I'll read an article on HN and think "Let's go to the comment section and see what the opinions to choose from are", or one of the first instincts after encountering a problem is googling and now asking an LLM.
Most of us are also old enough to have had a chance to develop taste in code and writing. Many of the young generation lack the experience to distinguish good writing from LLM drivel.
Yeah. Memorization and trivial knowledge is an optimization mechanism.
wanted to chime in on the educational system. in the west, we have the 'banking system' which treat a student as a bank account and knowledge as currency, hence the dump more info into ppl to make them sm0rt attitude.
in developing areas, they actually implement more modern models commonly, as its newer and free to implement newer things.
those newer models focus more on exactly this. teach a person how to go through the process of finding solutions. rather than 'knowing a lot to enable the process of thinking'.
not saying what is better or worse, but reading this comment and article it reminds me of this.
a lot of people i see, they know tons of interesting things, but anything outside of their knowledge is a complete mystery.
all the while ppl from developing areas learn to solve issues. alot of individuals from there also, get out of their poverty and do really well for themselves.
ofcourse, this is a generalization and doesnt hold up in all cases. but i cant help think about it.
a lot of my colleagues dont know how to solve problems simply because they dont RTFM. they rely on knowledge from their education which is already outdated before they even sign up.. i try to teach them to RTFM. it seems hopeless. they look at me , downwards, because i have no papers. but if shit hits the fan, they come to me. solve the prolbem.
a wise guy i met once said (likely not his words). there are 2 type of ppl. those who think in problems, and those who think in solutions.
id related that to education, not prebaked human properties.
One of my favorite things about Feynman interviews/lectures is often his responses are about how to think. Sometimes physicists ask questions in his lectures and his answer has little to do with the physics, but how they're thinking about it. I like thinking about thinking, so Feynman is soothing.
It depends what level the education is happening at. Think of it like students being taught how to do for loops but are just copying and pasting AI output. That isn't learning. They aren't building the skills needed to debug when the AI gets something wrong with a more complicated loop, or understand the trade offs of loops vs recursion.
Finding the correct balance for a given class it hard. Generally, the lower level the education, the more it should be closed books because the more it is about being able to manually solve the smaller challenges that are already well solved so you build up the skills needed to even tackle the larger challenges. The higher the education level, the more it is about being able to apply those skills to then tackle a problem, and one of those skills is being able to pull relevant formulas and such from the larger body of known formulas.
Even more broadly, it's "critical thinking," which definitely seems to be on the decline (though I'm sure old people have said this for millennia)
While there’s no reasonable way to disagree with the sentiment, I don’t think I’ve ever met anyone who can “think and decompose problems” who isn’t also widely read, and knows a lot of things.
Forcing kids to sit and memorize facts isn’t suddenly going to make them a better thinker, but much of my process of being a better thinker is something akin to sitting around and memorizing facts. (With a healthy dose of interacting substantively and curiously with said facts)
So to summarize:
My boss said we were gonna fire a bunch of people “because AI” as part of some fluff PR to pretend we were actually leaders in AI. We tried that a bit, it was a total mess and we have no clue what we’re doing, I’ve been sent out to walk back our comments.
Boss->VP: "We need to fire people because AI"
VP->Public: "We'll replace all our engineers with AI in two years"
Boss->VP: "I mean we need to fire VPs because AI"
VP->Public: "Replacing people with AI is stupid"
They are still not hiring junior engineers
Well they’re just trying to reduce headcount overall to get the expenses for AWS in better shape and work through some bloat. The “we’re doing layoffs because of AI” story wasn’t sticking though so looks like now they’re backtracking that story line.
Looks like the AWS CEO has changed religion. A year back, he was aboard the ai-train - saying AI will do all coding in 2 years [1]
Finally, the c-suite is getting it.
[1] https://news.ycombinator.com/item?id=41462545
He didn't actually say that. He said it's possible that within 2 years developers won't be writing much code, but he goes on to say:
"It just means that each of us has to get more in tune with what our customers need and what the actual end thing is that we're going to try to go build, because that's going to be more and more of what the work is as opposed to sitting down and actually writing code...."
https://www.businessinsider.com/aws-ceo-developers-stop-codi...
If you read the full remarks they're consistent with what he says here. He says "writing code" may be a skill that's less useful, which is why it's important to hire junior devs and teach them how to learn so they learn the skills that are useful.
He is talking his book. Management thinks it adds value in the non-coding aspects of the product - such as figuring out what customers need etc. I suggest management stays in their lanes, and not make claims on how coding needs to be done, leave that to the craftsmen actually coding.
This place is just reddit now.
> Please don't post comments saying that HN is turning into Reddit. It's a semi-noob illusion, as old as the hills.
https://news.ycombinator.com/newsguidelines.html
Theoretically, a large part of Amazon's worth is the skill of its workforce.
Some subset of the population likes to pretend their workforce is a cost that provides less than zero value or utility, and all the value and utility comes from shareholders.
But if this isn't true, and collective skill is worth value, then saying anyone can have that with AI at least has some headwind on your share price - which is all they care about.
Does that offset a potential tailwind from slightly higher margins?
I don't think any established company should be cheerleading that anyone can easily upset their monopoly with a couple of carefully crafted prompts.
It was always kind of strange to me, and seemed as though they were telling everyone, our moat is gone, and that is good.
If you really believed anyone could do anything with AI, then the risk of PEs collapsing would be high, which would be bad for the capital class. Now you have to correctly guess what's the next best thing constantly to keep your ROI instead of just parking it in save havens - like FAANG.
Amazon would probably say its worth is the machinery around workers that allows it to plug in arbitrary numbers of interchangeable people and have them be productive.
Amazon doesn't really work this way.
Bedrock/Q is a great example of how Amazon works. If we throw $XXX at the problem and YYY SDEs at the problem we should be able to build Github Copilot, GPT-3, OpenRouter and Cursor ourselves instead of trying to competitively acquire and attract talent. The fact that Codewhisperer, Q and Titan barely get spoken about on HN or Twitter tells you how successful this is.
But if you have that perspective then the equation is simple. If S3 can make 5 XXL features per year with 20 SDEs then if we adopt “Agentic AI” we should be able to build 10 XXL features with 10 SDEs.
Little care is given to organizational knowledge, experience, vision etc. that is the value (in their mind) of leadership not ICs.
What do you mean, “Amazon doesn’t really work that way”?
Parent is talking about how C-Suite doesn’t want to trumpet something that implies their entire corporate structure is extremely disadvantaged vs new entrants and your response is “Amazon wants to build everything themselves”?
Amazon isn’t some behaviorally deterministic entity, and it could (and should?) want to both preserve goodwill and build more internally vs pay multiples to acquire.
I guess it could be that people inside are not people they have to compete with, but it doesn’t seem like that’s what you're saying.
It’s a requirement for the C-suite to always be aware of which way the wind is currently blowing.
It’s one thing to be aware of which way the wind is blowing, and quite another to let yourself be blown by the wind.
Not to say that‘s what the AWS CEO is doing—maybe it is, maybe it isn’t, I haven’t checked—I’m just commenting on the general idea.
An LLM would be more efficient at this task, truth be told.
it could be if a new model was released every day where the training set included everything that happened yesterday.
That's not necessarily inconsistent though - if you need people to guide or instruct the autonomy, then you need a pipeline of people including juniors to do that. Big companies worry about the pipeline, small companies can take that subsidy and only hire senior+, no interns, etc., if they want.
> That's not necessarily inconsistent though
Pasting the quote for reference:
> Amazon Web Services CEO Matt Garman claims that in 2 years coding by humans won't really be a thing, and it will all be done by networks of AI's who are far smarter, cheaper, and more reliable than human coders.
Unless this guy speaks exclusively in riddles, this seems incredibly inconsistent.
There is no pipeline though. The average tenure of a junior developer even at AWS is 3 years. Everyone knows that you make less money getting promoted to an L5 (mid) than getting hired in as one. Salary compression is real. The best play is always to jump ship after 3 years. Even if you like Amazon, “boomeranging” is still the right play.
But that's fine, that's why I say for big companies - the pipeline is the entire industry, everyone potentially in the job market, not just those currently at AWS. Companies like Amazon have a large enough work force to care that there's people coming up even if they don't work there yet (or never do, but by working elsewhere free someone else to work at AWS).
They have an interest in getting those grads turned into would-be-L5s even if they leave for a different company. If they 'boomerang back' at L7 that's great. They can't if they never got a grad job.
that's interesting because that's how the consulting world works too. Start at a big firm, work for a few years, then jump to a small firm two levels above where you were. The after two years, come back to the big firm and get hired one level up from where you left. Rinse/repeat. It's the fastest promotion path in consulting.
Why is the hiring budget so much larger than the promotion budget?
It’s not necessarily “larger”, so much as different units. In a big company, the hiring budget is measured in headcount, but the promotion budget is measured in dollar percentage. It’s much easier to add $20k salary to get a hire done than to give that same person a $20k bump the following year.
It’s a lot cheaper to replace an employee by one who leaves at market rate than to pay all of your developers at market rate. Many are going to stick around because of inertia, their lack of ability to interview well, golden handcuffs of RSUs, they don’t feel like rebuilding the social capital at another company or the naive belief in the “mission”, “passion” etc
I went from an L5 (mid) working at AWS ProServe as a consultant (full time role) to a year later (and a shitty company in between) as a “staff architect” - like you said two levels up - at a smaller cloud consulting company.
If I had any interest in ever working for BigTech again (and I would rather get an anal probe daily with a cactus), I could relatively easily get into Google’s equivalent department as a “senior” based on my connections.
To be fair, the two statements are not inconsistent. He can continue to hire junior devs, but their job might involve much less actual coding.
There's definitely a vibe shift underway. C-Suites are seeing that AI as a drop-in replacement for engineers is a lot farther off than initial hype suggested. They know that they'll need to attract good engineers if they want to stay competitive and that it's probably a bad idea to scare off your staff with saying that they'll be made irrelevant.
I'm not sure those are mutually exclusive? Modern coders don't touch Assembly or deal with memory directly anymore. It's entirely possible that AI leads to a world where typing code by hand is dramatically reduced too (it already has in a few domains and company sizes)
He was right tho. AI is doing all the coding. That doesn’t mean you fire junior staff. Both can be true at once- you need juniors, and pretty much all code this days is AI-generated.
>Finally, the c-suite is getting it.
It can only mean one thing: the music is about to stop.
He should face consequences for his cargo-cult thinking in the first place. The C-Suite isn't "getting" anything. They are simply bending like reeds in today's winds.
Man how would we live without these folks and their sensational acumen. Millions well spent!
"Oh, how terrible" - he exclaims while counting dollar bills.
> “How's that going to work when ten years in the future you have no one that has learned anything,”
Pretty obvious conclusion that I think anyone who's thought seriously about this situation has already come to. However, I'm not optimistic that most companies will be able to keep themselves from doing this kind of thing, because I think it's become rather clear that it's incredibly difficult for most leadership in 2025 to prioritize long-term sustainability over short-term profitability.
That being said, internships/co-ops have been popular from companies that I'm familiar with for quite a while specifically to ensure that there are streams of potential future employees. I wonder if we'll see even more focus on internships in the future, to further skirt around the difficulties in hiring junior developers?
Might want to clarify things with your boss who says otherwise [1]? I do wish journalists would stop quoting these people unedited. No one knows what will actually happen.
[1]: https://www.shrm.org/topics-tools/news/technology/ai-will-sh...
I'm not sure those statements are in conflict with each other.
“My view is you absolutely want to keep hiring kids out of college and teaching them the right ways to go build software and decompose problems and think about it, just as much as you ever have.” - Matt Garman
"We will need fewer people doing some of the jobs that are being done today” - Amazon CEO Andy Jassy
Maybe they differ in degree but not in sentiment.
> quoting these people unedited
If you're quoting something, the only ethical thing to do is as verbatim as possible and with a sufficient amount of context. Speeches should not be cleaned up to what you think they should have said.
Now, the question of who you go to for quotes, on the other hand .. that's how issues are really pushed around the frame.
By unedited I mean, take the message literally and quote it to support a narrative that isn’t clear or consistent. (even internally among Amazon leadership)
Those statements are entirely consistent to me.
I very much believe that anything AWS says on the corporate level is bullshit.
From the perspective of a former employee. I knew that going in though. I was 46 at the time, AWS was my 8th job and knowing AWS’s reputation from 2nd and 3rd hand information, I didn’t even entertain an opportunity that would have forced me to relocate.
I interviewed for a “field by design” role that was “permanently remote” [sic].
But even those positions had an RTO mandate after I already left.
There's what AWS leadership says and then there's what actually sticks.
There's an endless series of one pagers with this idea or that idea, but from what I witnessed first hand, the ones that stuck were the ones that made money.
Jassy was a decent guy when I was there, but that was a decade ago. A CEO is a PR machine more than anything else, and the AI hype train has been so strong that if you do anything other than saying AI is the truth, the light and the way, you lose market share to competitors.
AI, much like automation in general, does allow fewer people to do more, but in my experience, customer desires expand to fill a vacuum and if fewer people can do more, they'll want more to the point that they'll keep on hiring more and more people.
In academia the research pipeline is this
Undergraduate -> Graduate Student -> Post-doc -> Tenure/Senior
Some exceptions occur for people getting Tenure without post doc or people doing some other things like taking undergraduate in one or two years. But no one expect that we for whole skip the first two and then get any senior researchers.
The same idea applies anywhere, the rule is that if you don't have juniors then you don't get seniors so better prepare your bot to do everything.
> Garman is also not keen on another idea about AI – measuring its value by what percentage of code it contributes at an organization.
You really want to believe, maybe even need to believe, that anyone who comes up with this idea in their head has never written a single line of code in their life.
It is on its face absurd. And yet I don't doubt for a second that Garman et al. have to fend off legions of hacks who froth at the mouth over this kind of thing.
Time to apply the best analogy I've ever heard.
> "Measuring software productivity by lines of code is like measuring progress on an airplane by how much it weighs." -- Bill Gates
Do we reward the employee who has added the most weight? Do we celebrate when the AI has added a lot of weight?
At first, it seems like, no, we shouldn't, but actually, it depends. If a person or AI is adding a lot of weight, but it is really important weight, like the engines or the main structure of the plane, then yeah, even though it adds a lot of weight, it's still doing genuinely impressive work. A heavy airplane is more impressive than a light weight one (usually).
I just can’t resist myself when airplanes come up in discussion.
I completely understand your analogy and you are right. However just to nitpick, it is actually super important to have a weight on the airplane at the right place. You have to make sure that your aeroplane does not become tail heavy or it is not recoverable from a stall. Also a heavier aeroplane, within its gross weight, is actually safer as the safe manoeuverable speed increases with weight.
I think this makes the analogy even more apt.
If someone adds more code to the wrong places for the sake of adding more code, the software may not be recoverable for future changes or from bugs. You also often need to add code in the right places for robustness.
right, thats why its a great analogy - because you also need to have at least some code in a successful piece of software. But simply measuring by the amount of code leads to weird and perverse incentives - code added without thought is not good, and too much code can itself be a problem. Of course, the literal balancing aspect isn't as important.
> a heavier aeroplane … is actually safer
Just to nitpick your nitpick, that’s only true up to a point, and the range of safe weights isn’t all that big really - max payload on most planes is a fraction of the empty weight. And planes can be overweight, reducing weight is a good thing and perhaps needed far more often than adding weight is needed. The point of the analogy was that over a certain weight, the plane doesn’t fly at all. If progress on a plane is safety, stability, or speed, we can measure those things directly. If weight distribution is important to those, that’s great we can measure weight and distribution in service of stability, but weight isn’t the primary thing we use.
Like with airplane weight, you absolutely need some code to get something done, and sometimes more is better. But is more better as a rule? Absolutely not.
This is a pretty narrow take on aviation safety. A heavier airplane has a higher stall speed, more energy for the brakes to dissipate, longer takeoff/landing distances, a worse climb rate… I’ll happily sacrifice maneuvering speed for better takeoff/landing/climb performance.
> the safe manoeuverable speed increases with weight
The reason this is true is because at a higher weight, you'll stall at max deflection before you can put enough stress on the airframe to be a problem. That is to say, at a given speed a heavier airplane will fall out of the air [hyperbole, it will merely stall - significantly reduced lift] before it can rip the wings/elevator off [hyperbole - damage the airframe]. That makes it questionable whether heavier is safer - just changes the failure mode.
> That is to say, at a given speed a heavier airplane will fall out of the air [hyperbole, it will merely stall - significantly reduced lift] before it can rip the wings/elevator off [hyperbole - damage the airframe]
Turbulence, especially generated by thunderstorms, or close to it.
Maneuvering speed is Va which is about max deflection on a single control surface, I think you're thinking of Vno if you're referring to turbulence
Progress on airplanes is often tracked by # of engineering drawings released, which means that 1000s of little clips, brackets, fittings, etc. can sometimes misrepresent the amount of engineering work that has taken place compared to preparing a giant monolithic bulkhead or spar for release. I have actually proposed measuring progress by part weight instead of count to my PMs for this reason
> the best analogy I've ever heard.
It’s an analogy that gets the job done and is targeted at non-tech managers.
It’s not perfect. Dead code has no “weight” unless you’re in a heavily storage-constrained environment. But 10,000 unnecessary rivets has an effect on the airplane everywhere, all the time.
> Dead code has no “weight”
Assuming it is truly dead and not executable (which someone would have to verify is & remains the case), dead code exerts a pressure on every human engineer who has to read (around) it, determine that it is still dead, etc. It also creates risk that it will be inadvertently activated and create e.g. security exposure.
Yes, we all love pedantry around here (that’s probably 99% of the reason I wrote the original comment!)
But if your position is that the percentage of time in the software lifecycle that dead code has a negative effect on a system is anywhere close to the percentage of time in an aircraft lifecycle that extra non-functional rivets (or other unnecessary weight objects) has a negative effect on the aircraft, you’re just wrong.
it's still directionally accurate though. Dead code has a weight that must be paid. Sometimes the best commits are the ones where you delete a ton of lines.
When you are thinking of development and refactoring, dead code absolutely has weight.
In this analogy, I'd say dead code corresponds to airplane parts that aren't actually installed on the aircraft. When people talk about the folly of measuring productivity in lines of code, they aren't referring to the uselessness of dead code, they're referring to the harms that come from live code that's way bigger than it needs to be.
This reminds me of a piece on folklore.org by Andy Hertzfeld[0], regarding Bill Atkinson. A "KPI" was introduced at Apple in which engineers were required to report how many lines of code they had written over the week. Bill (allegedly) claimed "-2000" (a completely, astonishingly negative report), and supposedly the managers reconsidered the validity of the "KPI" and stopped using it.
I don't know how true this is in fact, but I do know how true this is in my work - you cannot apply some arbitrary "make the number bigger" goal to everything and expect it to improve anything. It feels a bit weird seeing "write more lines of code" becoming a key metric again. It never worked, and is damn-near provably never going to work. The value of source code is not in any way tied to its quantity, but value still proves hard to quantify, 40 years later.
0. https://www.folklore.org/Negative_2000_Lines_Of_Code.html
Goodhart's law: when a measure becomes a target, it ceases to be a good measure.
Given the way that a lot of AI coding actually works, it’s like asking what percent of code was written by hitting tab to autocomplete (intellisense) or what percent of a document benefited from spellcheck.
While most of us know the next word guessing is how it works in reality…
That sentiment ignores the magic of how well this works. There are mind blowing moments using AI coding, to pretend that it’s “just auto correct and tab complete” is just as deceiving as “you can vibe code complete programs”.
When I see these stats, I think of all the ways "percentage of code" could be defined.
I ask an AI 4 times to write a method for me. After it keeps failing, I just write it myself. AI wrote 80% of the code!
All that said, I'm very keen on companies telling me how much of their codebase was written by AI.
I just won't use that information in quite the excitable, optimistic way they offer it.
"...just raised a $20M Series B and are looking to expand the team and products offered. We are fully bought-in to generative AI — over 40% of our codebase is built and maintained by AI, and we expect this number to continue to grow as the tech evolves and the space matures."
"What does your availability over the next couple of weeks look like to chat about this opportunity?"
"Yeah, quite busy over the next couple of weeks actually… the next couple of decades, really - awful how quickly time fills by itself these days, right? I'd have contributed towards lowering that 40% number which seems contrary to your goals anyway. But here's my card, should you need help with debugging something tricky some time in the near future and nobody manages to figure it out internally. I may be able to make room for you if you can afford it. I might be VERY busy though."
I want to have the model re-write patent applications, and if any portion of your patent filing was replicated by it your patent is denied as obvious and derivative.
It is like measuring company output based on stuff done through codegen...
Something I wonder about the percent of code - I remember like 5-10 years ago there was a series of articles about Google generating a lot of their code programmatically, I wonder if they just adapted their code gen to AI.
I bet Google has a lot of tools to say convert a library from one language to another or generate a library based on an API spec. The 30% of code these LLMs are supposedly writing is probably in this camp, not net novel new features.
Is that why gmail loads so slowly these days
As always, the truth is somewhere in the middle. AI is not going to replace everyone tomorrow, but I also don't think we can ignore productivity improvements from AI. It's not going to replace engineers completely now or in the near future, but AI will probably reduce the number of engineers needed to solve a problem.
I would bet that anyone who's worked with these models extensively would agree.
I'll never forget the sama AGI posts before o3 launched and the subsequent doomer posting from techies. Feels so stupid in hindsight.
The AGI doomerism was a marketing strategy. Now everyone gets what AI is and we're just watching a new iteration on search, ai's read all the docs.
It was always stupid, but no one is immune to hype, just different types of hype.
Especially with the amount of money that was put into just astroturfing the technology as more than it is.
ChatGPT is better than any junior developer I’ve ever worked with. Junior devs have always been a net negative for the first year or so.
From a person who is responsible for delivering projects, I’ve never thought “it sure would be nice if I had a few junior devs”. Why when I can poach an underpaid mid level developer for 20% more?
I've never had a junior dev be a "net negative." Maybe you're just not supervising or mentoring them at all? The first thing I tell all new hires under me is that their job is to solve more problems than they create, and so far it's worked out.
I’ve had interns be a net negative, I’ve had Juniors be a net negative, I’ve had Seniors be a net negative and even managers!
Turns out some people suck, but most of them don’t suck.
But by definition, junior developers with no experience are going to need more handholding and tale time away from experience developers.
Truly depends on the organization and systems. I’m at a small firm with too few Senior staff, lots of fire-fighting going on among us, etc. We have loads of low-hanging fruit for our Juniors so we tend to have very quick results after an initial onboarding.
> junior developers with no experience are going to need more handholding
Unlike AI, which gives me fake methods, broken code, and wrong advice with full confidence.
I just “wrote” 2000 lines of code for a project between Node for the AWS CDK and Python using the AWS SDK (Boto3). Between both, ChatGPT needed to “know” the correct API for 12 services, SQL and HTML (for a static report). The only thing it got wrong with a one shot approach was a specific Bedrock message payload for a specific LLM model. That was even just a matter of saying “verify the payload on the web using the official docs”.
Yes it was just as well structured as I - someone who has been coding as a hobby or professionally for four decades - would have done.
> I just “wrote” 2000 lines of code for a project between Node...
I think I wrote -200 lines of code on my last PR. I may be doing something bad for that number to be negative.
The difference is probably that I only do green field POC implementations as a solely developer/cloud architect on a project if I am doing hands on keyboard work.
The other part of my job is leading larger projects where I purposefully don’t commit to pulling stories off the board since I’m always in meetings with customers, project managers, sales or helping other engineers.
I might even then do a separate POC as a research project/enablement. But it won’t be modifying existing code that I didn’t design.
That's great for you. I ask Sonnet 4 to make a migration and a form in Laravel Filament, and it regularly shits itself. I'm curious what those 12 services were, they must've had unchanging, well documented APIs.
That’s the advantage of working with AWS services, everything is well documented with plenty of official and unofficial code showing how to do most things.
Even for a service I know is new, I can just tell it to “look up the official documentation”
Using ChatGPT 5 Fast
AWS CDK apps (separate ones) using Node
- EC2 (create an instance)
- Aurora MySQL Serverless v2
- Create a VPC with no internet access - the EC2 instance was used as a jump box using Session Manager
- VPC Endpoints for Aurora control plane, SNS, S3, DDB, Bedrock, SQS, Session Manager
- Lambda including using the Docker lambda builder
- DDB
- it also created the proper narrowly scoped IAM permissions for tfe lambdas (I told it the services the Lambdas cared about)
The various Lambdas in Python using Boto3
- Bedrock including the Converse and Invoke APIs for the Nova and Anthropic families
- knowing how to process SQS Messages coming in as events
- MySQL flavored SQL for Upserts
- DDB reads
In another project the services were similar with the addition of Amazon Transcribe.
I think you’re just telling on what kind of mentor you are with your comment.
I’m a great mentor when given the time. Two former interns for whom I was their official mentor during my time at AWS got return offers and are thriving two years after I left. I threw one in front of a customer to lead the project within three months after they came back after graduating. They were able to come through technically and had the soft skills. I told them my training approach is to “throw them at the bus. But never under the bus.”
I’m also a great teacher. That’s my $DayJob and has been for the past decade first bringing in new to the company processes and technologies, leading initiatives, teaching other developers, working with sales, CxOs (smaller companies), directors, explaining large “organizational transformation” proposals etc. working at startups and then doing the same in cloud consulting first working at AWS (ProServe full time role) and now working as a staff architect full time at a third party consulting company.
But when I have been responsible for delivery, I only hire people who have experience “dealing with ambiguity” and show that I can give them a decently complicated problem and they can take the ball and run with it and make decent decisions and do research. I don’t even do coding interviews - when I interview it’s strictly behavioral and talking through their past projects, decision making processes, how they overcame challenges etc.
In terms of AWS LPs, it’s “Taking Ownership” (yeah quoting Amazon LPs made me throw up a little).
The most impressive folks Ive worked with are almost always straight out of school. It's before they've developed confidence about their skills and realized they can be more successful by starting their own business. People who get promoted three times in just 5 years sort of good.
Did their project manager and/or team lead think when they were hired “they are really going to be a great asset to my team and are going to help me complete my sprint/quarterly goals”?
When I ask for additional headcount, I’m looking at the next quarter since that’s what my manager is judging me based on.
I've never worked with a junior developer that was incapable of learning or following instructions unless I formatted them in a specific way.
I definitely have
That's due to the person, not their development experience level
What happens when you retire and there are no juniors to replace you?
That sounds like an incentive issue.
My evaluations are based on quarterly goals and quarterly deliverables. No one at a corporation cares about anything above how it affects them.
Bringing junior developers up to speed just for them to jump ship within three years or less doesn’t benefit anyone at the corporate level. Sure they jump ship because of salary compression and inversion, where internet raises don’t correspond to market rates. Even first level managers don’t have a say so or budget to affect that.
This is true for even BigTech companies. A former intern I mentored who got a return offer a year before I left AWS just got promoted to an L5 and their comp package was 20% less than new hires coming in at an l5.
Everyone will be long gone from the company if not completely retired by the time that happens.
> Bringing junior developers up to speed just for them to jump ship within three years or less doesn’t benefit anyone at the corporate level.
What? Of course it does. If that's happening everywhere, that means other companies' juniors are also jumping ship to come work for you while yours jump ship to work elsewhere. The only companies that don't see a benefit from mentoring new talent are those with substandard compensation.
That’s true, but why should I take on the work of being at the beginning of the pipeline instead of hiring a mid level developer. My incentives are to meet my quarterly goals and show “impact”.
To a first approximation, no company pays internal employees at market rates in an increasing comp environment after a couple of years especially during the first few years of an employee’s career where their marker rate rapidly increases once they get real world experience.
On the other hand, the startup I worked for pre-AWS with 60 people couldn’t, wouldn’t and shouldn’t have paid me the amount I made when I got hired at AWS.
I'm a technical co-founder rapidly building a software product. I've been coding since 2006. We have every incentive to have AI just build our product. But it can't. I keep trying to get it to...but it can't. Oh, it tries, but the code it writes is often overly complex and overly-verbose. I started out being amazed at the way it could solve problems, but that's because I gave it small, bounded, well-defined problems. But as expectations with agentic coding rose, I gave it more abstract problems and it quickly hit the ceiling. As was said, the engineering task is identifying the problem and decomposing it. I'd love to hear from someone who's used agentic coding with more success. So far I've tried Co-pilot, Windsurf, and Alex sidebar for Xcode projects. The most success I have is via a direct question with details to Gemini in the browser, usually a variant of "write a function to do X"
> As was said, the engineering task is identifying the problem and decomposing it.
In my experience if you do this and break the problem down into small pieces, the AI can implement the pieces for you.
It can save a lot of time typing and googling for docs.
That said, once the result exceeds a certain level of complexity, you can't really ask it to implement changes to existing code anymore, since it stops understanding it.
At which point you now have to do it yourself, but you know the codebase less well than if you'd hand written it.
So, my upshot is so far that it works great for small projects and for prototyping, but the gain after a certain level of complexity is probably quite small.
But then, I've also find quite some value in using it as a code search engine and to answer questions about the code, so maybe if nothing else that would be where the benefit comes from.
> At which point you now have to do it yourself, but you know the codebase less well than if you'd hand written it.
Appreciate you saying this because it is my biggest gripe in these conversations. Even if it makes me faster I now have to put time into reading the code multiple times because I have to internalize it.
Since the code I merge into production "is still my responsibility" as the HN comments go, then I need to really read and think more deeply about what AI wrote as opposed to reading a teammate's PR code. In my case that is slower than the 20% speedup I get by applying AI to problems.
I'm sure I can get even more speed if I improve prompts, when I use the AI, agentic vs non-agentic, etc. but I just don't think the ceiling is high enough yet. Plus I am someone who seems more prone to AI making me lazier than others so I just need to schedule when I use it and make that time as minimal as possible.
Current generation of AI agents are great at writing a block of codes. Similar to writing a great paragraph. Know your tools.
AWS is a very infrastructure intensive project with extremely tight SLAs, and no UI, makes a lot of sense.
>teach “how do you think and how do you decompose problems”
That's rich coming from AWS!
I think he meant "how do you think about adding unnecessary complexity to problems such that it can enable the maximum amount of meetings, design docs and promo packages for years to come"!
"Learning how to learn" is by far the most important lesson anyone can obtain. That's not just for AI/software/tech, but for anything.
There is a dedicated website on this topic: https://learnhowtolearn.org
Junior staff will be necessary but you'll have to defend them from the bean-counters.
You need people who can validate LLM-generated code. It takes people with testing and architecture expertise to do so. You only get those things by having humans get expertise through experience.
I was going to say something, then I realized my cynicism is already at maximum.
Well yeah, they're just doing this with H1B's and OPT. The other kind of "AI".
Yesterday, I was asked to scrape data from a website. My friend used ChatGPT to scrape data but didn't succeded even spent 3h+. I looked website code and understand with my web knowledge and do some research with LLM. Then I described how to scrape data to LLM it took 30 minutes overall. The LLM cant create best way but you can create with using LLM. Everything is same, at the end of the day you need someone who can really think.
LLM's can do anything, but the decision tree for what you can do in life is almost infinite. LLM's still need a coherent designer to make progress towards a goal.
LLMs can do small things well, but you must use small parts to form big picture.
Or you could’ve used xpath and bs4 and have been done in an hour or two and have more understandable code.
it is not that easy, there is lazy loading in the page that is triggered by scroll of specific sections. You need to find clever way, no way to scrape with bs4, so tough with even selenium.
wget -m ?
It is too late it is already happening. The evolution of tech field is people being more experienced and not AI. But AI will be there for questions and easy one liners. Properly formalized documentation, even TLDRs.
The cost of not hiring and training juniors is trying to retain your seniors while continuously resetting expectations with them about how they are the only human accountable for more and more stuff.
AWS CEO says what he has to say to push his own agenda and obviously to align himself with the most currently popular view.
Finally someone from a top position said this. After all the trash the CEOs have been spewing and sensationalizing every AI improvement, for a change, a person in a non-engineering role speaks the truth.
Point is nobody has figured out how much AI can replace humans. People. There is so much of hype out there as every tech celebrity sharing their opinions without responsibility of owning them. We have to wait & see. We could change courses when we know the reality. Until then, do what we know well.
My respect for people that take this approach is very high. This is the right way to approach integration of technology.
Can SOME people's jobs be replaced by AI. Maybe on paper. But there are tons of tradeoffs to START with that approach and assume fidelity of outcome.
Remark is at 12:02 in the video.
https://www.youtube.com/watch?v=nfocTxMzOP4&t=722s
Did a double take at Berman bring described as an AI investor. He does invest but a more appropriate description would be "AI YouTuber".
I don't mean that as a negative, he's doing great work explaining AI to (dev) masses!
If AI is so great and had PhD level skills (Musk) then logic says you should be replacing all of your _senior_ developers. That is not the conclusion they reached which implies that the coding ability is not that hot. Q.E.D.
Agreed.
LLMs are actually -the worst- at doing very specific repetitive things. It'd be much more appropriate for one to replace the CEO (the generalist) rather than junior staff.
I can't wait for that damn bubble to explode, really...
This is becoming unbreathable for hackers.
It's already exploding.
The hype train is going to keep on moving for a while yet though.
That's right it should be used to replace senior stuff right away
Why do so many people act like they’re mutually exclusive? Junior staff can use AI too.
Sometimes, the C suite is there because they're actually good at their jobs.
> “My view is you absolutely want to keep hiring kids out of college and teaching them the right ways to go build software and decompose problems and think about it, just as much as you ever have.”
Instead you should replace senior staff who make way more.
I mean I used Copilot / JetBrains etc. to work on my code base but for large scale changes it did so much damage that it took me days to fix it and actually slowed me down. These systems are just like juniors in their capabilities, actually worse because junior developers are still people and able to think and interact with you coherently over days or weeks or months, these models aren’t even at that level I think.
No it isn't, he's lying. Sorry guys.
Claude code is better than a junior programmer by a lot and these guys think it only gets better from there and they have people with decades in the industry to burn through before they have to worry about retraining a new crop.
Most people don't notice but there has been a inflation in headcounts over the years now. This happened around the time microservices architecture trend took over.
All of sudden to ensure better support and separation of concerns people needed a team with a manager for each service. If this hadn't been the case, the industry as a whole can likely work with 40% - 50% less people eventually. Thats because at any given point in time even with a large monolithic codebase only 10 - 20% of the code base is in active evolution, what that means in microservices world is equivalent amount teams are sitting idle.
When I started out huge C++ and Java code bases were pretty much the norm, and it was also one of the reasons why things were hard and barrier to entry high. In this microservices world, things are small enough that any small group of even low productivity employees can make things work. That is quite literally true, because smaller things that work well don't even need all that many changes on a everyday basis.
To me its these kind of places that are in real trouble. There is not enough work to justify keeping dozens to even hundreds of teams, their managements and their hierarchies all working for quite literally doing nothing.
Its almost an everyday song that I hear, that big companies are full of hundreds or thousands of employees doing nothing.
I think sometimes the definition of work gets narrowed to a point so infinitesimal that everyone but the speaker is just a lazy nobody.
There was an excellent article on here about working at enterprise scale. My experience has been similar. You get to do work that feels really real, almost like school assignments with instant feedback and obvious rewards when you're at a small company. When I worked at big companies it all felt like bullshit until I screwed it up and a senator was interested in "Learning more" (for example).
The last few 9s are awful hard to chase down and a lot of the steps of handling edge case failures or features are extremely manual.
There are two freight trains currently smashing into each other:
1.) Elon fired 80% of twitter and 3 years later it still hasn't collapsed or fallen into technical calamity. Every tech board/CEO took note of that.
2.) Every kid and their sister going to college who wants a middle class life with generous working conditions is targeting tech. Every teenage nerd saw those over employed guys making $600k from their couch during the pandemic.
On the other hand while yes it's still running, twitter is mostly not releasing new features, and has completely devolved into the worst place on the internet. Not to mention most accounts now actually are bots like Elon claimed they were 3 years ago.
> twitter is mostly not releasing new features
Were they even releasing new features anyways?
I can't think of any new features that Twitter implemented even the 5 years preceding Musk's buyout.
I dont know; they may well have been on the ads and moderation side. And they did add the hangouts/voice calls stuff, but that may have been an acquisition, I'm not sure.
I hope the tech boards and CEOs don’t miss the not very subtle point that twitter has very quickly doubled in size in 2 years and is still growing after the big layoff and they had to scramble to fix some notable mistakes they made when firing that many people. 80% is already a hugely misleading marketing number.
Edit: huh, what’s with the downvote, is this wrong? Did I overstate it? Here’s the data: https://www.demandsage.com/twitter-employees/
Also need to add, that a large part of the 80% that got kicked, was moderator staff. So it makes sense that after they removed too many developers, they ended up rehiring them.
Take in account, Twitter their front end, the stuff that people interact with was only like 15% of the actual code base. The rest was analytics for the data (selling data, marketing analytic for advertisers etc).
But as they are not reintroducing moderators, the company is "still down by 63.6% from the numbers before the mass layoffs".
So technically, Twitter is probably back or even bigger on the IT staff then before Musk came.
Looking at your numbers, twitter went from 7,490 employees before Musk's layoffs to 2,840 in 2024. That's still a reduction by 63%.
> In this microservices world, things are small enough that any small group of even low productivity employees can make things work. That is quite literally true, because smaller things that work well don't even need all that many changes on a everyday basis.
You're committing the classic fallacy around microservices here. The services themselves are simpler. The whole software is not.
When you take a classic monolith and split it up into microservices that are individually simple, the complexity does not go away, it simply moves into the higher abstractions. The complexity now lives in how the microservices interact.
In reality, the barrier to entry on monoliths wasn't that high either. You could get "low productivity employees" (I'd recommend you just call them "novices" or "juniors") to do the work, it'd just be best served with tomato sauce rather than deployed to production.
The same applies to microservices. You can have inexperienced devs build out individual microservices, but to stitch them together well is hard, arguably harder than ye-olde-monolith now that Java and more recent languages have good module systems.
> If this hadn't been the case, the industry as a whole can likely work with 40% - 50% less people eventually. Thats because at any given point in time even with a large monolithic codebase only 10 - 20% of the code base is in active evolution, what that means in microservices world is equivalent amount teams are sitting idle.
I think it depends on the industry. In safety critical systems, you need to be testing, making documentation, architectural artifacts, meeting with customers, etc
There's not that much idle time. Unless you mean idle time actually writing code and that's not always a full time job.
I think most people misunderstand the relationship between business logic, architecture and headcount.
Big businesses don’t inherently require the complexity of architecture they have. There is always a path-dependent evolution and vestigial complexity proportional to how large and fast they grew.
The real purpose of large scale architecture is to scale teams much moreso than business logic. But why does headcount grow? Is it because domains require it? Sure that’s what ambitious middle managers will say, but the real reason is you have money to invest in growth (whether from revenue or from a VC). For any complex architecture there is usually a dramatically simpler one that could still move the essential bits around, it just might not support the same number of engineers delineated into different teams with narrower responsibilities.
The general headcount growth and architecture trajectory is therefore governed by business success. When we’re growing we hire and we create complex architecture to chase growth in as many directions as possible. Eventually when growth slows we have a system that is so complex it requires a lot of people just to understand and maintain—even if the headcount is longer justified those with power in the human structure will bend over backwards to justify themselves. This is where the playbook changes and a private equity (or Elon) mentality is applied to just ruthlessly cut and force the rest of the people how to keep the lights on.
I consider advances in AI and productivity orthogonal to all this. It will affect how people do their jobs, what is possible, and the economics of that activity, but the fundamental dynamics of scale and architectural complexity will remain. They’ll still hire more people to grow and look for ways to apply them.
This is blatantly incorrect. Before microservices became the norm you still had a lot of teams and hiring, but the teams would be working with the same code base and deployment pipeline. Every company that became successful and needed to scale invented their own bespoke way to do this; microservices just made it a pattern that could be repeatedly applied.
Amazon has been at the forefront of micro services at scale since 2002.
https://nordicapis.com/the-bezos-api-mandate-amazons-manifes...
Well yeah... computers are really powerful. you don't need docker swarm or any other newfangled thing. Just perl and apache and mysql and you can ship to tens of millions of users before you hit scaling limits.
It would be sad if you are correct. Your company might not be able to justify keeping dozens and hundreds of teams employed, but what happens when other companies can't justify paying dozens and hundreds of teams who are the customers buying your product? Those who gleefully downsize might well deserve the market erosion they cause.
I think that the starting point is that productivity/developer has been declining for a while, especially at large companies. And this leads to the "bloated" headcount.
The question is why. You mention microservices. I'm not convinced.
Many think it is "horizontals". Possible, these taxes add up it is true.
Perhaps it is cultural? Perhaps it has to do with the workforce in some manner. I don't know and AFAIK it has not been rigorously studied.
A lot of companies that have stopped hiring junior employees are going to be really hurting in a couple of years, once all of their seniors have left and they have no replacements trained and ready to go.
Maybe source of "AI replacing junior staff" is the statement AWS CEO made during a private meeting with client.
Not often article makes me want to buy shares. Like... never. This article is spot on.
junior engineers aren't hired to get tons of work done; they're hired to learn, grow, and eventually become senior engineers. ai can't replace that, but only help it happen faster (in theory anyway).
But HN's favorite idiot says that people who have this take are "nuts":
> but the code is shitty, like that of a junior developer
> Does an intern cost $20/month? Because that’s what Cursor.ai costs.
> Also: let’s stop kidding ourselves about how good our human first cuts really are.
https://fly.io/blog/youre-all-nuts/
So which one is it?
Is HN's favorite idiot "dumb" or is this CEO "nuts"?
Please HN, I NEED YOU TELL ME IF HOW STUPID I AM!
Time will show you are right and almost all the other dumbasses here at HN are wrong. Which is hardly surprising since they are incapable of applying themselves around their coming replacement.
They are 100% engineers and 100% engineers have no ability to adapt to other professions. Coding is dead, if you think otherwise then I hope you are right! But I doubt it because so far I have only heard arguments to the counter that are obviously wrong and retarded.
That’s a shit article by Thomas, people should stop quoting it.
This is just to walk back previous statements by Andy Jessy. Political theater
Unfortunately, this is the kind of view that is at once completely correct and anathema to private equity because they can squeeze a next quarter return by firing a chunk of the labor force.
Are we trying to guilt trip corporations to do socially responsible thing regarding young workers skill acquisition?
Haven't we learned that it almost always ends up in hollow PR and marketing theater?
Basically the solution to this is extending education so that people entering workforce are already at senior level. Of course this can't be financed by the students, because their careers get shortened by longer education. So we need higher taxes on the entities that reap the new spoils. Namely those corporations that now can pass on hiring junior employees.
Finally some fucking sanity from a megacorp CEO. Been a long time.
"AGI" always has been a narrative scam after late 2022.
Reinforcement is the only way to fly
Bravo.. Finally a voice of reason.
As someone who works in AI, any CEO who says that AI is going to replace junior workers has no f*cking clue what they are talking about.
> “Often times fewer lines of code is way better than more lines of code,” he observed. “So I'm never really sure why that's the exciting metric that people like to brag about.”
I remember someone that had a .sig that I loved (Can't remember where. If he's here, kudos!):
> I hate code, and want as little of it in my programs as possible.
The bubble has, if not burst, at least gotten to the stage where it’s bulging out uncomfortably and losing cohesion.
Simple, just replace the CEO with an LLM and it will be singing a different tune :-P
Juniors are cheaper than Ai tokens and easier to fire and hire.
finally some common sense from the detached c-suitors
Everything is boiled down to lambda calculus.
I heard from several sources that AWS has a mandate to put GenAI in everything and force everyone to use it so... yeah.
They're deliberately popping the bubble. If they'd actually thought this and cared, they'd have said it 2 years ago, before the layoffs.
Stop getting played.
The amount of developer cope here is over 9000
"We have always been at war with Eastasia"
that was not the kicker....it was the ending comment...
Better learn how to learn as we are not training(or is that paying) you to learn...
No one's getting replaced, but you may not hire that new person that otherwise would have been needed. Five years ago, you would have hired a junior to crank out UI components, or well specc'd CRUD endpoints for some big new feature initiative. Now you probably won't.
> well specc'd CRUD endpoints
I’m really tired of this trope. I’ve spent my whole career on “boring CRUD” and the number of relational db backed apps I’ve seen written by devs who’ve never heard of isolation levels is concerning (including myself for a time).
Coincidentally, as soon as these apps see any scale issues pop up.
On the other hand, that extra money can be used to expand the business in other ways, plus most kids coming out of college these days are going to be experts in getting jobs done with AI (although they will need a lot of training in writing actual secure and maintainable code).
> plus most kids coming out of college these days are going to be experts in getting jobs done with AI
“You won’t lose your job to AI, you’ll lose it to someone who uses AI better than you do”
Even the highest ranking engineers should be experts. I don’t understand why there’s this focus on juniors as the people who know AI best.
Using AI isn’t rocket science. Like you’re talking about using AI as if typing a prompt in English is some kind of hard to learn skill. Do you know English? Check. Can you give instructions? Check. Can you clarify instructions? Check.
> I don’t understand why there’s this focus on juniors as the people who know AI best.
Because junior engineers have no problem with wholeheartedly embracing AI - they don't have enough experience to know what doesn't work yet.
In my personal experience, engineers who have experience are much more hesitant to embrace AI and learn everything about it, because they've seen that there are no magic bullets out there. Or they're just set in their ways.
To management that's AI obsessed, they want those juniors over anyone that would say "Maybe AI isn't everything it's cracked up to be." And it really, really helps that junior engineers are the cheapest to hire.
Sure. First line tech support as well. In many situations customers will get vastly superior service if AI agent answers the call.
At least in my personal case, struggling with renewal at Virgin Broadband, multiple humans wasted probably an hour of everyone's time overall on the phone bouncing me around departments, unable to comprehend my request, trying to upsell and pitch irrelevant services, applying contextually inappropriate talking scripts while never approaching what I was asking them in the first place. Giving up on those brainless meat bags and engaging with their chat bot, I was able to resolve what I needed in 10 minutes.
> Sure. First line tech support as well. In many situations customers will get vastly superior service if AI agent answers the call.
In IT, if at a minimum, AI would triage the problem intelligently (and not sound like a bot while doing it), that would save my more expensive engineers a lot more time.
This is mostly because CS folks are given such sales and retention targets; and while I’ve never encountered a helpful support bot even in the age of LLMs, I presume in your case the company management was just happy to have a support bot talking to people without said metrics.
Its strange you have to write this.
In India most of the banks now have apps that do nearly all the banking you can do by visiting a branch personally. To that extent this future is already here.
When I had to close my loan and had to visit a branch nearly a few times, the manager tells me, significant portion of his people's time now goes into actual banking- which according to him was selling products(fixed deposits, insurances, credit cards) and not customer support(which the bank thinks is not its job and has to because there is no other alternative to it currently).
“brainless meat bags” have you ever thought they are instructed to do so to achieve product selling quotas?
Anyone who blindly follows orders is a brainless meat bag too.
Again, you assume those people have choice, you definitely should search more how people on these jobs are pressured to reach quotas and are abused in many ways. A simple search on Reddit you can see plenty of reports about it:
https://www.reddit.com/r/callcentres/comments/1iiqbxh/the_re...
Abuses done by customers: https://www.bbc.com/news/business-59577351
but wait, what about my Nvidia stocks bro? Can we keep the AI hype going bro. Pls bro just make another AI assistant editor bro.
It's refreshing to finally see CEOs and other business leaders coming around to what experienced, skeptical engineers have been saying for this entire hype cycle.
I assumed it would happen at some point, but I am relieved that the change in sentiment has started before the bubble pops - maybe this will lesson the economic impact.
Yeah, the whole AI thing has very unpleasant similarities to the dot com bubble that burst to the massive detriment of the careers of the people that were working back then.
The parallels in how industry members talk about it is similar as well. No one denies that the internet boom was important and impactful, but it's also undeniable that companies wasted unfathomable amounts of cash for no return at the cost of worker well being.