Hey, all. I really don't know what to do with a post like this.
I'm being sincere when I say (as I've said on two threads here) that this genre of posts --- "I'm leaving this company I've been very publicly associated with, and here's the new thing I'm doing" --- is deeply cursed. There's no way to say anything interesting without it just stinking like an ad for the new thing.
Obviously, anything at all you say about a commercial project you're working on is easily read as promotional. And you're right, this kind of writing almost always is promotional. But there's a way to do it where at least you're trying to be in conversation with your peers, rather than hitting people over the head with how awesome you think the project is.
But I don't know how to do that in a post like this. I think the only way to read it is as, like, an investor memo. Not my goal, but I don't make the rules.
So my strategy here is just to stay kind of vague, and talk about where I think the world is going, rather than the specific thing we're doing. I can talk your ears off about capability systems, datalog, models driving hardware, virtualization, whatever. Those are fun conversations and I'm very psyched to have them; it's what lights me up about the work we're doing now.
But I don't think it can work here. I didn't submit this post and I didn't upvote it. I wrote it because I didn't want the whole thing I'm leaving Fly.io for to be wrapped up in some dumb Twitter thread.
If you're unsatisfied with the post, I don't blame you, but it's less a bid for the front page of HN than it is an update to my "about me" page. I'd literally rather talk about HN meta, and how to write for HN, than I would about operating systems at this moment. I truly appreciate the interest though.
You are overthinking it :) You are one of the most interesting people on HN. I am sure others think along those lines too and would love to learn more about what you are building and where the industry is going. Others can just tune it out.
I completely agree that we haven’t even gotten past the first post yet in terms of the magnitude of change that is likely to occur. People fail to comprehend that we’re still assimilating the Internet as a means of communication. Cell phones. It will be decades before what we have today is fully utilized, let alone what is likely to be developed. Anyway, thanks for the article.
Personally, I'd like to hear the full pitch, and I suspect many others would too. If the ideas are interesting enough for you to dedicate the next N years of your life to, I'm probably willing to spend 10 minutes reading the blog post. Particularly from you, a known-good writer.
We were on a call recently where I mentioned that as a fly user SDLC and declarative config was more important than ever, and so I was pushing for sprites to have OCI images.
I'm pretty AI pilled, and I'm building a local-first personal software platform, but I dunno... I've been feeling like i need to cram the "AI spirit" into very specifically shaped boxes to actually build robust software tastefully at speed. Like software on the outside and AI on the inside instead of the other way around.
I do think the future of computing will definitely be more like the Linux diaspora than the apple ecosystem, but also.. it is easy to get "duped" by AI without actually shipping. I'm starting to want "KPIs" to help me measure if my time is actually well spent against the things i care about.
I guess I'm extrapolating from your post, but even if the programming language is English, we'll probably still want to spend a large amount of additional compute to produce "artifacts" that have things like performant code, security hardening, and certified taste.
I do think the future of computing will definitely be more like the Linux diaspora than the apple ecosystem, but also... when i first got into linux i spent way too long trying to get my desktop perfectly aesthetic because it was all suddenly malleable. Now i vibe more with Debian.
I wish you luck, and I'm looking forward to seeing what you come up with! Fun times ahead for sure.
> I've been feeling like i need to cram the "AI spirit" into very specifically shaped boxes to actually build robust software tastefully at speed.
This is such incredible verbiage. Thank you for saying something so clearly that I've been saying myself in many other ways. I've been using terms such as "providing structure" and "applying constraints" and "reducing degrees of freedom".
I think very specifically the idea of "taste" is going to be extremely important in these AI-saturated days, and it's really where experienced humans will shine...at least for a while.
A post a few days ago turned funtional core, imperative shell into deterministic core, non-deterministic shell. And a comment [1] on that discussed turning that into "slop core, artisinal shell". Which makes a lot of sense actually despite feeling the wrong round initially. And maybe what an OS needs to do is to support that, at a higher level. What I think we will need is strong isolation - all the software might be coming from the "same place", but do we trust that place? Certainly not currently!
Hi Thomas, I think you're being too defensive. It's ok, I look forward to your technical posts, on what you imaging this OS powered by AI to be. Will be a fun ride.
If you're looking for a UX idea, when the phone boots, it asks "What are the top 5 things you do?". And that's how the phone experience starts.
All the best. (Ignore the silly haters, they're lost in their AI pessimism to just look forward to a new intersting thing).
1. When you write a comment in a discussion of an article, be upfront with being the author. It took me a while to understand that from your comment. "Author here" is a thing.
2. Your article itself doesn't have clear authorship stated. At the very bottom there is a copyright notice, but that's pretty generic and not a single person either. Just place your name at the top together with a date. Not everybody knows you or follows your blog.
3. When I read this ...
> When I was a little kid, a family friend sold us our first “computer” — scare quotes because I’m pretty sure it was a VTech Laser 200 clone, a chiclet keyboard Z80 that ran off cassette tapes and plugged into our TV.
.. I prepared to stop reading since .. how is that not a computer? After another paragraph or two which just felt like rambling, I gave up.
These are bad-faith arguments toward "What about self modifying software".
Self modifying software has no grounds in reality or predictable behavior. It is an unreasonable expectation that software is magic. You would not fly on a self-modifying airplane. This is not a future i want to live in.
The author, as a kid, believed computers where like a genie that could manifest anything they ask for. It seems to be a case of wishful thinking of a kid.
But it seems that he is falling for the LLM hype the same way. We all have the kid version of ourselves hiding somewhere within our grown versions, it seems...
First, the halting problem is only undecidable on arbitrary programs. Here we would be writing programs specifically to be easy to prove correct. In fact, we would be developing programs and proof at the same time.
Second, determinism is a nice property, but it's independent of whether you can proof software correct.
HyperLogLog only gives you a probabilistic approximation of the true number of distinct items. Here the proof would be about the probability that you are within some acceptable window around the true answer.
> "But it booted straight into BASIC. That’s all it did. I was, like, 8 years old. I wasn’t about to learn BASIC. What I learned instead: computers were not as awesome as I’d imagined."
You were the exception. Most kids felt awe. We started learning BASIC and creating "dumb games". That's the origin story of most people my age who ended up in this doomed industry.
I think you were the exception. Most kids never even knew what BASIC was. If there was a game already on the computer, they'd be happy to play it. If they really liked the game they might learn how to start the game if it wasn't already running, but they didn't care about doing anything else. Kids who were into computers looked at a computer and saw endless possibility. Everybody else played sports outside, watched TV, or hung out with their friends.
The kids who loved computers were the ones plugging in line after line of BASIC. Well, them and the ones who just really loved games. Games were a gateway drug for a lot of kids because for so long games could really make you work for it. Games had kids making changes to config.sys and setting BLASTER environment variables, connecting to a BBS or FTP sites for downloads, and troubleshooting network issues to get LAN games running with friends. You could get pretty good at computers just getting your fix. That took a pretty strong commitment though because if all you cared about were games, and if you had them, consoles made gaming very simple. Too simple if you valued the sense of learning and accomplishment that came from winning a wrestling match with an unforgiving computer keeping you from the game you wanted to play.
I was one of the kids who didn't have a computer at home but had friends with those, and would read the BASIC listings in the magazines many times over to figure exactly what they did and think about what I could change before I visited my friends and started typing it all on their computer.
Yeah, I did that too. That's what you did with these machines. Deeply unsatisfying! Like: a real lesson in how gnarly real computers actually were. This is my point. You got one of these thingies and thought you could do anything, but it turns out "anything" took on values ranging from "hours to many hours of typing in BASIC listings and then working through inscrutable errors until something showed up on the screen but then crashed". Computers sucked ass!
Different strokes, I guess. FWIW, I also started at 8 -- with a Commodore 64 and a Commodore Vic-20, both connected to a television -- and I found the process of trail-and-error learning my way through BASIC to be absolutely awesome.
I would start with a completely blank slate and then, gradually, be able to create something out of nothing. If I was clever enough, it could do whatever I wanted (as long as my wants were modest). It was magical and I have nothing but fond memories of the time I spent doing that.
Haha! I did, but not for a few years, until I took up Pascal and MacOS and then wanted to find a way around the 32k text limit that was a stubbornly difficult problem to solve for a while (and was the original reason that BBEdit became so popular).
Trying to correctly render a vertical blinking cursor on top of proportional fonts took me ages to get right...ish.
Aside from the Spectrum, the Amstrad CPC (CP/M!), at various times a secondhand BBC Model B and then later the family PC, I wrote complex programs for a Sharp PC-1246 (because I had the cassette interface).
This was the very opposite of unsatisfying. I don’t recognise this idea that computers sucked.
AI-assisted coding is, to me, way more soul-destroying than hitting the limits of a simple home computer. In most regards I feel like I have reached the end of my interest in computing, and the pivot I had in mind — teaching - looks like it has been joy-deadened too. Nobody is going to want to learn how anything that is not AI actually works, tech people insist that AI will replace (they say augment, they mean replace) teachers and trainers, politicians seem to be listening, and non-tech people now increasingly resent technology.
I barely recognise the industry I spent my life in; I do not recognise the evident enthusiasm for a world that is abandoning a whole, transformative pedagogy in favour of the programming equivalent of games consoles: button pushing vibes.
I am not at all of the opinion that AI has no value, though it does not have anywhere close to the value people insist it must have for me. I am learning, trying to find an ethical angle on it that doesn’t make me feel uncomfortable and exhausted by the hype.
But it is the final nail in the coffin of the joy of tech work for me, and as a single middle-aged man I perceive this as life changing and very likely life-shortening, in the way that psychologically unhealthy things are a health risk for older adults: unless I find something that gives my work life a sense of connected purpose again, I figure I will be gone before the Spectrum makes it to fifty.
> Nobody is going to want to learn how anything that is not AI actually works, tech people insist that AI will replace teachers and trainers, politicians seem to be listening, and non-tech people now resent technology
I disagree with most of your post, but this one the most.
Non-tech people already resent technology... Try and sit next to a non tech person use a desktop pc. It wasn't until the smart phone that the average person started to use technology like you or I recognise. And that's only because it's been simplified to the extreme for them. (I don't blame them btw).
AI is such a step change for learning and teaching... For anyone who is inquisitive. Ask it endless questions, explore ideas, get it to generate images to help you...
>You got one of these thingies and thought you could do anything..
Why did you think that? Did you see people use it like that? anything? really?
For me, computers became intriguing after I started programming in BASIC. Not before that. Before that I was just curious about what the big deal about these things are...
In the literature, there would be passages that describe how computer followed "instructions", but I never could make sense of that or understood what an "instruction" was. So it didn't really sway me either way. I just tried hard and remained curious to understand what this "instruction" was and how I could give it to computers...
As near as I can remember, my first real programming experience was when I was eight years old, writing basic programs on an Apple IIc at school in 3rd grade. (I was too poor to afford my own computer until my senior year of college). It was awesome.
I have never found computers disappointing, and I have spent most of my career/life learning about them - from processor design through to distributed systems. Mostly I've been disappointed in the disparity between the number of interesting things I'd like to do, and the number of lifetimes I have available to me.
Figuring out inscrutable errors seems highly educational. I guess I don't have enough data to determine whether your bad experience was more or less common (probably less common on HN though). I imagine some (like yourself at 8) found computers frustrating and bad, while others (some peer posters on HN, who might also have been a bit older when they started) found them frustrating and good!
I did come to appreciate that. I was a professional C programmer for a lot of years! A hand tool woodworker, so to speak. It's great. But it's not the same as what I had thought computers were when I was a little kid.
You got a computer in that era because you thought you talked to them and they did what you asked them to do.
Then you got the computer and when it started up you got a flashing K. It was a bit of a disappointment. But you flicked open the hefty manual, started reading and before long you were hooked.
One career later, I now have a computer that is very, very close to what I wanted all those years ago but with a significant and irritating genie attitude to granting my wishes.
It was probably a function of not just personality type but education. For whatever reason I was taught basic in elementary school as a millennial and it left an impression.
Education has to help demystify computers and give people an idea of what they could do with one. Basic seems like a weird choice for a millennial, but at least it had you actually write code instead of abstracting it away into pushing around icons like some of the systems I've seen for teaching young kids about software or even just giving directions to turtles.
Why is BASIC a weird choice? It wasn't really a choice at all. All the computers in my school ran it until I was 8 or 9 and the school board could afford to upgrade to Windows machines.
If you define a millennial as someone born from 1982-1996, the majority of those kids by the time the had a computer lab (say by the time they were in middle school) would have been too old for BASIC. You are talking 92-06, Windows 95 took over the majority market share by the end of '96, and you are forgetting a whole lot of schools had Apples being pumped in, so a lot less access to BASIC. Unless you were in a programming adjacent class, the majority of millennials weren't learning basic. Ask your non tech millenials how much BASIC they learned and they would look at you like you grew an extra arm.
Depending on where you draw the line for millennials the school computers could have been running anything from Windows 95 to Windows XP. BASIC wouldn't have been included in XP or 2000, and depending on what exactly they were using, it may have been easier to use something with more features. QBasic for example wouldn't have syntax highlighting. Simple programs in languages like python or perl can be used to teach the same concepts, with fairly similar syntax, and would have been more applicable to software people were actually writing at the time.
Your priors are all out of whack. Most people never learned BASIC, let alone at 8. Programming is much more accessible now than then, but the overwhelming majority of people still don’t know any JavaScript or python or anything else.
Programming (by humans) is less accessible in 2026 because most personal computing devices don't boot into a programming environment and they don't include a user-friendly programming instruction manual (graphing calculators may be an exception however). And before the era of "there's an app for that" it was assumed that users would need to write their own programs, so systems were designed to support that. Vibe programming (by LLMs) is more accessible now however.
BASIC was designed as a simplified, interactive version of FORTRAN that non-experts could learn in an afternoon, but which was still powerful enough for a wide range of applications including numerical methods, simulations, data processing, and games. It was also small enough to implement effectively on a timeshared minicomputer or an 8-bit microcomputer.
Even though the Mac was GUI-based, for a time it shipped with, and promoted, HyperCard, which led to a brief explosion of user-created programs developed as HyperCard stacks.
>most personal computing devices don't boot into a programming environment
Yes, but now everyone and their mum has a personal computing device, and there's tutorials readily accessible, and five seconds of googling can bring you to an online programming environment.
All true, but the onramp isn't included in the box. When you get a smartphone, how do you know that you could even program it? Is there any encouragement to do so vs. just buying apps? They are more like game consoles, in more ways than one.
As I mentioned, I think graphing calculators might be an exception.
In the mid 80s, something like 10% of US households had any computer at all. That the onramp is in the box doesn't really matter to the people who never saw the box.
I guess the question is how you measure accessibility - per device, or across the population as a whole? In 2026 several billion people have personal computing devices, but the devices no longer provide that programming onramp. Interestingly enough, it looks like ~25% of the world's population is still offline.
Unix/Linux historically gave gave you a shell prompt (allowing you to type in shell programs) after login and a C compiler in the box. Modern Ubuntu or macOS technically still have terminal apps and shell prompts, but usually boot into a desktop GUI.
You were the exception too. A PC was one of the most expensive items in my house when we bought it. The computer table it sat on was the second most. I was so afraid of doing anything on it that'd break it. Heck I freaked out one time when I shut it down from the power button directly.
My Acorn BBC Model B booted straight into BBC BASIC as well, and I too thought it was awesome and taught myself BASIC straight away, and then 6502 assembler and latterly BCPL and so on. I was even younger than OP.
But I think we were the exception, even if we didn’t realise it at the time. In hindsight, my interest was most definitely not shared by my peers.
I remember getting magazines with basic games, where the full syntax was printed out. I would type it in and save to floppies on my Apple 2e - which my mother purchased second hand.
I'm not even that old, but totally agree with this sentiment. Going to miss those times. It's amazing being able to build whatever you want now, but it's bittersweet because if you truly wanted to learn how to do it, you could do it yourself.
I don't understand this. Back in the early days of the internet, you absolutely couldn't learn 'how to do it' without finding obscure books or knowing someone. Now though? You can learn all of it, and find the resources almost instantly. Want to build a computer, self host, program, you have millions of resources at your hand.
Things that are easy to do don't necessarily mean that's the right way to do it. Grueling work sometimes helped you learn far more than just getting the answer to something easy.
Back in the early days of the internet, there were FAQs written about everything. You could talk to people who actually knew what they were doing, as opposed to today.
I ended up a systems programmer in this doomed industry. BASIC on a Z80 didn't get me there (exploits on a DG/UX at a library system did.) Different people, different vectors.
I feel like the future is bright precisely because AI sucks so bad while also lowering the bar.
It made people not take software and software engineers for granted. Public perception of software has also finally matured. Boomers and tech CEOs fell into a tarpit that harshly exposes their stupidity while laypeople now feel like they can change the system.
This is all exciting in the best ways. My usage of AI at work and for personal projects goes as far as whatever the AI summary is on google. That's more than enough for me to get enough of a productivity boost on par with those who embarrassingly thrust themselves balls deep and refuse to pull out. Not everyone has to push the "I'm feeling lucky" button to get the same work done at the same pace.
There's nothing to be excited for. Even if AI works as well as people say (and to be clear: it very much doesn't), that future sucks ass. No more fun making things, no fun solving problems, just hand it off to the machine that does things for you. That's an awful future where every shred of joy is gone from computing.
I think that it should not need to be mutually exclusive. You can make it you can enter the programs yourself (if you prefer to do for whatever reason, including, it does not work well (like you mention), or because of Dijkstra objection or other objections to the use of AI for programming the computer or AI in general), but that AI can also be possible.
Are you telling us the only reason why making things was fun is because it made you "special"? And now that anyone can do it it isn't worth it for you anymore?
For me, I made things because I liked the process of making things, it was never really about the result. Much like solving crossword puzzles is about the process itself, not about getting the answer.
You are completely right. I am not entitled to anything. But before I was paid to have fun, I spent years trying to get better at it and then suddenly it completely lost its value. From fun of solving crossword puzzles to the dread of validating if crossword puzzles bot solved crossword puzzle correctly, with worse pay, less job security or opportunities.
I am not entitled for things to continue as they were, but I can still feel sad about it, you know.
I have never been able to understand this perspective. It's come up so often that it almost feels like a pro-AI propaganda strawman in the style of the "virgin vs chad" meme.
You can and should still write most of the code yourself and use AI to assist when you get stuck. Apart from saving yourself some hours because you no longer hit dead ends on google, I don't understand how that's meaningfully different from using a search engine and finding solutions on StackOverflow and GitHub. I don't even understand how people think this traditional workflow I'm suggesting is any slower than reviewing LLM output either. It takes the same cognitive load and the same time in the end. The only people who believe they can go faster and reduce cognitive load aren't just inexperienced, but also seem to have ADHD or something. Like they just wanna bask in the vibes of the machine-generated text scrolling by.
If your goal is to make something you're going to live with and maintain long term, or is more generally the foundation of a bigger idea, you can't ignore any part of the code or delegate it away. It's still writing that you're bound by like any other document.
I don't know what's wrong with my perspective. I cannot relate to this at all.
I am at a job where I am paid well to build software to do things in a distributed system that processes information across various external systems (some with physical world impacts). The outcome of that computing is revenue for the company. Using agent-wrapped LLMs makes the software authorship component of the job really fast. The rest is virtually the same speed.
For Thomas's mobile computing examples, these are done really well by my phone's operating system's built-in voice command system, mature for a decade by now.
At home, I use software that is well-crafted to do some thing. Or I write a series of scripts to do narrow things that for some reason aren't in Aptitude or Snap, for example "copy all the photos off my iPhone and rename them and sort them by date in my filesystem".
Perhaps I have no imagination, but I don't feel like having an LLM make applications on my phone. I want good software that someone that I trust wrote.
Thomas has been a successful cryptographer, vulnerability analyst, and entrepreneur, so he probably knows what he's doing. I just don't get it.
> At home, I use software that is well-crafted to do some thing. Or I write a series of scripts to do narrow things that for some reason aren't in Aptitude or Snap, for example "copy all the photos off my iPhone and rename them and sort them by date in my filesystem".
> Perhaps I have no imagination, but I don't feel like having an LLM make applications on my phone. I want good software that someone that I trust wrote.
I do not use LLM either, and do not intend to do. I can write my own programs. Even if someone does want to use LLM, it should neither be required nor expected nor the default (and there are also reasons why I think that it should be discouraged to use LLM too much); you should be able to write the scripts and programs by yourself if you want to do it that way (and the computer made to make this possible to do without LLM), even if some people can make LLM to do it at their option, but it should not to require LLM and AI.
I also think that the computer that can start with a programming language such as BASIC or Forth (without first needing menus and GUI to access all of the programs) is useful, and that it is not as good that many modern computers do not do like that.
In my opinion, this post misses the forest for the trees. Yes, the model of an OS running various off-the-shelf apps for various tasks will soon become obsolete. But it's not the idea of the OS that's outdated - it's the idea of apps. Why would you ever want to ask an AI to make an app for you to complete some task, if you can instead ask the AI to complete the task directly? The future isn't a bunch of extremely personalized apps - the future is a single AI assistant doing all the work.
I've seen a few posts of the form "why would you ever want to use anything else than AI from now on?" by now. My counterpoints would be:
Unless you have a local model and the appropriate hardware for it, your agent is Somebody Else's Computer. Do you really want to send all your data and make your entire computing experience dependent on whatever OpenAI or Anthropic or whoever else is planning in this moment?
That's not even starting with inference time and token cost. Despite all the incredible advances in inference, it still takes more time than most non-AI computer functionality. Do you really want to wait a few minutes and pay money for something that you could also do with a few clicks fully locally on your PC?
But the most important thing: User interfaces. Right now, we're basically cramming everything you could possibly want to do at a computer into a chat interface. But there are lots of applications fir which specialized graphical interfaces are much more suitable. Why would you want to get rid of them?
Somewhat connected to that: Repetition. If you have to do the same task again every week or every day, it seems wasteful to ask the AI for it every time: Not just are you wasting a lot of time, energy and tokens, you're also at risk of getting inconsistent results, if the agent from today's session will interpret the requirements slightly differently than the agent from yesterday.
You can circumvent all those things by having the AI write you a custom app, but then using the app without AI to do the task.
Those are all techbical limitations that will be solved with time. In 5 years I would be surprised if we do NOT have an AI as smart as today's frontier models, that runs completely locally, and has a real time voice interface.
I'm not a Treky but I think Star Trek has the right interface here. For basic operations you just use voice and tell it your intent or what you need to know "Computer, what's the fastest route for me to get home?". "Madison is fastest right now, it's 3 minutes faster than Lake, it traffic is building up".
This isn't just whatever Waze gives you, it knows about your constraints and preferences. So you need to program that somehow. Imagine taking it one step further "Computer, what's the fastest route to pickup my daughter and get home?" or even it just responding to the first prompt "Take Roosevelt, and today you need to pickup your daughter from volleyball".
In Star Trek they often do this when they create "simulations". They tell the computer constraints and then it refines it. No writing programs, but effectively that's what you are doing but with the natural language voice interface.
"Computer let's create a simulation to figure out the fastest way for me to get home. I want to take a route along the expressway and know whether to get off at Roosevelt, Madison, or Lake. I need to pickup my daughter when she has volleyball." You don't need to repeat the whole instruction each time, and behind the scenes it is probably turning it into some deterministic code, so next time you trigger that program it's not going to hallucinate some nonsense.
We live in a "capitalist" society. Moats will be built, regulations will be enacted to ensure the American Corporation controls the market segment. No one has been able to unseat Amazon, no one has been able to unseat Google, no one has been able to unseat Microsoft, no one has been able to seat Twitter, Facebook, etc. The modern market is segments that are controlled by a single entity. There will be no local models.
> "Computer let's create a simulation to figure out the fastest way for me to get home. I want to take a route along the expressway and know whether to get off at Roosevelt, Madison, or Lake. I need to pickup my daughter when she has volleyball."
Imagine saying that to a random dude with a map and that should give you a good idea on how many info you're leaving out. You can say that the AI will infere those but that seems to be the common fallacy of AI enjoyers: They're always assuming the AI will magically find out the missing information from their prompt somehow.
I have Organic Maps installed on my phone and within a few minutes (faster than my car warming up) I can set a multi stop route (you can bookmark places). Training a random user to use such apps is also equally fast. The same happens with various pro tools: once trained, a user can be very fast with them.
> I have Organic Maps installed on my phone and within a few minutes (faster than my car warming up) I can set a multi stop route (you can bookmark places).
Wow that's slow. I can do the same in Google Maps in seconds. Can you tell me what I'm missing? I'm definitely missing something.
Much tighter electronic control of fuel, ignition and even cooling system combined with tighter tolerances and modern oil formulations. Wait for the cat to light off and the revs drop and then put it in gear and take it easy for the next couple of minutes of driving.
We are talking about a hypothetical future where coding AIs are good enough, cheap enough and ubiquitous enough that not just expert programmers, but everybody is capable of vibe coding exactly what they need, to the point they actually reach for coding AIs and not the app store with ready-made apps. That future is still far away, at least 5 years, probably closer to 10-15. In that hypothetical future, most of the problems you're talking about would be solved, or else we wouldn't have that future yet.
Sometimes it is wild to me how two people can look at the same information and come to completely different conclusions.
As best as I can tell, the only constraint the really matters right now is availability of compute, or as a buddy of mine put it "GPU GO BRRRRRRRRRRR!"
I imagine the future of computer interaction could evolve into a split model combining foundational capabilities that just make sense to be pre-built, tested and shared deterministic code - codecs, numerical solvers, geometry kernels, databases and transactional tools, compilers, theorem provers, cryptography, etc. - with a fluid interaction layer that accesses either local or cloud-provided versions of those foundational capabilities and is capable of synthesising any tools it needs where it would be helpful to build it first - either as a one-off or something to cache - for anything that can’t just be done as a task.
Applications are compressed bundles of some assumptions about how something should work and be interacted with. So you wouldn’t need VLC, it’s a shell around a foundational codec layer. Instagram is just a view over data in the Instagram social network, and you could customise it as you see fit. Transmit is a lovely application but you can delegate that to a fluid task layer that can do one-off tasks or make something you can re-use.
Everything in the fat ‘middle layer’ of software that is connecting a user to something foundational through interface assumptions might disappear, unless it’s more useful for it to exist as a permanent tool. Things that a lot of people share or that are more complex, Blender for example.
I also imagine there’d be a lot more focus on making explicit things like canonical records or schema, contracts, provenance, versioning, access policies, some kind of fixed views. Things that your local fluid interface would rely on the structure for when it came to connecting to your bank and doing whatever you want with your finances.
Not all software as we think of it might disappear, but the things that are lightweight layers over something more stable could all be subsumed with tools synthesised on demand.
> if you can instead ask the AI to complete the task directly?
Going to entirely depend on the nature of the task and where AI pricing goes in the future.
Would you like to spend $0.10 on every task or would you like to spend $1.00 to make an app that does it free for you forever?
I don't think it'll be one or the other.
But also I don't see things like video games ever being replaced by 1 off AI. If for no other reason then the fact that a lot of games are fun because of multiplayer capabilities. That's not something you can do a 1 and done with AI (unless you only want to play games with family and close friends).
If it costs $0.10 to do the task once, then you can be sure as hell it costs orders of magnitude more than $1 to vibe code a whole damn app that does that thing, especially if you want good user experience which is the whole point of custom vibe coded apps. Whereas the App Store will have an inferior, mass-market version of what you want actually priced $1.
That's not necessarily true. Vibecoding a simple JSON -> YAML converter to translate a 2MB JSON is going to be much cheaper than including that JSON in a prompt and burning hundreds of thousands of tokens. Besides, there's lots of cases where the math simply favors a bigger one-time cost rather than paying fees in perpetuity.
Even today, an advanced AI model will realize exactly what you said, and generate a script and run it behind your back to save the costs and ensure correctness. To me, that's still using the agent directly, not vibe coding. No app gets developed in the process, and if the agent doesn't tell you, you won't even know any coding took place.
With the caveat that literal today's models are not allowed to run code themselves but instead give you the code in the chat window and ask you to run it yourself for security and safety reasons. But the capability is there.
Right, which makes the one-shot task cost the same as the permanent solution in that case. So if you wanted to do this task even twice, you would still be better off creating the tool once and just using it. Imagine if proof assistants like Lean didn't exist. Research with AI would get a whole lot more expensive if it had to re-create Lean every single time you asked it to prove something.
If you want to do literally the exact same task in literally the exact same configuration and ideally the exact same input data, then yes, it costs the same. But that solution is as permanent as it is useless.
A tool for generating a report for September 2026 is no more expensive than an agent generating a report for September 2026. But that tool is useless for you in October 2026. You'll need to generate a whole new tool for October 2026, and generating 2 tools costs as much as generating 2 reports so you're not saving anything.
Making a tool that can generate a report for any single month can easily cost 10 times more than a single report, entirely due to the extra code flow analysis required to extract the right variables in the right places. And that's without any UI. If you want UI that's another several times more tokens because UI is hard. Also it will suck compared to a simple chat window where you can ask the assistant for anything and it will understand. If you want UI that's better than the chat, that alone costs you more than everything you spent so far.
All in all, your permanent solution starts paying for itself in about 4 to 6 years, assuming you never had to update how the report is generated in the meantime.
Or you could have a very simple, very cheap script that only gets run once then gets deleted (or even better, kept for future reference to make future report scripts even cheaper to generate).
Note that this is all in the hypothetical future where AI is actually capable of one shotting such a task. If it makes mistakes, then human-made apps are still the most cost-effective option and this whole discussion is moot.
An AI assistant can give different answers for the same inputs. An application written by an AI assistant, if it doesn't just call an LLM at runtime, will mostly likely produce the same answer for the same inputs. That is, unless there's something intrinsic to its business logic that makes it nondeterministic.
Sometimes you really want the latter even if the former is less effort.
And when applicable, the program would be much more efficient than the LLM.
So it would really be beneficial for the LLMs to be able to have their own environments to write and run code, but also to store those programs for later use automatically.
A *human* assistant can give different answers for the same inputs. At some point, we'll need to stop treating computers as instruction executors and start treating them as autonomous agents. Non-determinism isn't a bug to be fixed, it's just the reality of working with this new kind of machines, and operators of those machines need to embrace that fact, like they did for human assistants.
And we started adding rules for the humans to follow, until we have complex bureaucracy. We invented tons of formal notations systems and error detecting and correction mechanisms. We invented deterministic machines. All because we are deeply unsatisfied with the non-determinism! Because it means you can never trust some process, optimize it away and stop thinking about it. You always need to check, and there is always something missing.
A lot of our manufacturing ability improvement during the last 200 years, as well as our engineering ability improvement during the last 500 years, is not just the raw power we unlocked, but that we made it deterministic, reliable and reproducible and hence automatable and automatic.
But sometimes you do just want the same answer each time. Not this human woo crap, where the answer depends on god knows what, but a deterministic function of some kind, reliably producing a given output for a given input according to a fixed set of rules.
Yes, but sometimes (quite often actually) you really don't care. The entire subfield of audiovisual compression is built around the principle that you don't care. The data is actively being lost and the image distorted, and you're fine with it. Wanting deterministic results out of MP3 is like wanting AC units to be loud. Sure, they are loud, just like MP3 is deterministic, but why exactly is being loud one of your requirements when choosing an AC unit? Isn't it enough to produce cool air, or in case of MP3 - sound good enough?
Accuracy, precision and determinism are completely different things. Lossless compression reduces precision, and by that also accuracy, but it doesn't become undeterministic.
It would be like telling the computer to compress a music file containing Bach's Toccata and Fugue in D Minor, and getting back a file that contains Rick Astley's Never Gonna Give You Up. Maybe you don't care, but I do.
You are talking about AI outright failing to get anything remotely close to the correct result. That's a completely different argument and has nothing to do with non-determinism. Things can be deterministically wrong too.
I want an agent studio and execution sandbox where my LLM driven assistant builds and maintains my apps with a receipt or transcript like history of what it did to create each app. Determinism built per use case with non deterministic assistance orchestrating the build and management layer. Extra credit if I can trivially share and collaborate with others per “deterministic silo” via a shared link. Something like Claude Code + AWS Lambda|(Docker|Podman) + Tangled.org with LLM code generation for my phone and workstation.
The LLMs empower for building, the determinism improves output-expectation alignment. It’s fundamentally an on device software factory, hypervisor, and storage system.
Cattle can do complex tasks too. Including managing other cattle.
Think of all those bad sci-fi movies where aliens take over Earth, enslave all humans and treat them like cattle. Very competent cattle that's put in charge of local governments, scientific research etc. but very much still cattle where individuals are entirely interchangeable and killing them is just everyday work. This is me and AI assistants. (I'm the aliens.)
If the code works good enough, it works good enough. If good enough is insufficient for success criteria, you’ll have to trade app evolution velocity and flexibility for more determinism via more human effort (“artisanal hand crafted”), evaluated on a per use case basis. “Does this software have to be good? How good? Have I hit diminishing returns yet?”
Pros and cons to code being cheap to generate now, it’s always about tradeoffs. I don’t want to build cars, I want to drive cars.
Only in the sense that your build-a-car will be "good enough" as long as you never find out why your palms are always itchy, and you never press the brake so hard it stops working, and nobody determines which of the cars started the fireball during the accident.
People suck at "good enough". All major engineering disasters were something "good enough" for somebody. People burn down their houses with "good enough" electrical work.
> Yes, the model of an OS running various off-the-shelf apps for various tasks will soon become obsolete.
I doubt it. Most people would rather stab at their phone to download an app that just does what they want than explain to a chatbot over a period of minutes/hours and several reiterations to get it to spew out what they want, let alone what kind of software they want it to make for them. Most people can barely articulate what they want in the first place.
The market for applications isn't going anywhere and neither is the OS. I can't imagine booting a computer directly into a chat window where I have to explain to a bot what I wish would show up on my screen, where on my screen, and how it should look. What a total nightmare that would be!
Even if what passes for "AI" these days was as smart and capable as the computers on Star Trek (and it very much isn't) you wouldn't want to ask it for everything. Even the Enterprise had an OS that people interacted with. What they also had though was an OS with configurable interfaces that people could customize according to their current task and preferred workflow.
That sort of thing is maybe possible in the future for "AI" empowered operating systems we might live to see, but 90% of the people using computers aren't going to bother with any of that, just like they've never seen the settings/options in most of their applications or their operating system right now. Not because it couldn't make their time working the computer better, but because they just can't be bothered.
The writer of this piece remembers how, as a child, computers promised to let him write his own software and games and it seemed like there were endless possibilities until the reality that programing takes effort killed a little of that enthusiasm (although not enough that he didn't eventually become a programmer), but many people never see a computer as anything more than a means to an end or a device for mainly passive content consumption.
The cognitive dissonance is astounding in AI circles. The vast majority of people have a personal computer in their pocket and they use it the vast majority of the time to scroll Instagram, play games, and occasionally interact with their bank and other social services. People who have laptops and desktop computers have them for specialized tasks they already do. None of the vast majority of users who buy their phone when the screen cracks and the battery stops working care about the agentic AI world because their phone already does everything they want to do.
I think the rise of Claw's, Grok Bot and Muse is a good counterpoint to this. Yes they are still niche specialized interfaces right now, but they are basically allowing non-technical people to do things that have been easily possible with scripts and cronjobs for a long time. Its basically the Ford and faster houses analogy.
When I look around me, the only time people get their phone out of their pockets is either for youtube, some social media, or calling/chatting with people. It's very rare to even see them set reminders or use the calculator. It's look like you're in some tech bubble. People outside of that don't scripts or do repetitive and formalized stuff. The people that do create apps and services to sell.
My wife is as non-technical as it gets, but she still spends hours with ChatGPT everyday. Calorie counting, parenting tips, gardening help, fact checks, looking up celebrities, road trip planning, meal planning, health related questions, all with ChatGPT. I really don't think she's the only one. It wasn't the most downloaded iPhone app of 2025 for nothing.
Okay, I went too far with that one sentence. Can we all pretend I have instead said "this model will be obsolete for most, not all, but most productivity apps and most, not all, but most office tools"?
It is not clear whether models will be both capable enough and cost effective enough but not as capable to wipe us all out. The outcome is a tiny sliver of possible futures and even then likely short-lived.
Current capabilities are not there. You need software engineers to keep it going. And even then as intoxicating as it might be to be able to bring to life all the side projects you often had no energy/skill to finish, today it is still may be cost / “cognitive load” -effective to buy off the shelf apps that somebody else has to maintain (even if they are not as personal / less customized for your use-case).
Apps will become barely visible data providers. Some will provide specialized widgets for those cases where a specialized UI is needed. And data will be designed to inter operate.
Why would you ever want a nice word processor to write a letter to someone when you could just ask AI to write the letter? Why would you ever want photo editing software when you could just tell the AI to make your picture look good? Why would you want audio recording and mixing software when you could just tell the AI to output the next Billboard 100 hit? Why would you want a camera when you could just ask AI to generate photos for you?
Like browsing TikTok or messaging your friends? People want their devices to work like their friends’ devices do. Your prediction seems to be out of touch with how and why people actually use computing devices. Nobody actually uses them to "perform tasks".
Maybe we need an app store just for agents to make purchases. Imagine a world where startups have their valuations determined by how many agents are using their apps. Those startups... Also operated by agents. We don't even need users anymore!!!
Inputs for the ui might be me talking, or typing, or having someone else talk or type. They might also be GPS, temperature, BLE devices nearby. Or maybe a button I smash because I don't feel like re explaining something.
What happens after an input is still the best part of a UI. We like seeing/hearing/feeling things in different ways.
The fun part about AI assistant is that you can straight up tell it how you want data presented, and it will do its best to present it that way (within limitations of its capabilities and available tools).
It’s funny you say this because Apple resisted an App Store for apps. They originally expected web based apps. The market demanded an App Store. Remember web snippets where one could clip a portal to view a small section of a website?
There is a big misconception in your answer in that you seem to believe that an LLM would always produce (machine) code that does what the user intended, in a correct and safe manner. Neither of these assumptions is true. If you knew how LLMs are built and operate, you would know they are not reliable at all. What you might ask from an AI interpreter OS might be unique and thus might be absent from its training set and might not follow a pattern inferred from its training set either.
LLMs are the first machine learning models that blatantly and regularly produce incorrect output and we have been brainwashed into accepting that. An application on the other hand can be exhaustively verified. There is no comparison.
First and foremost, I'm comparing specifically just having a suite of heavily personalized vibecoded apps like in OP blog post, versus using an agent directly. I'm comparing just these two options and nothing else.
If we don't assume up front that AIs are good enough for at least one of those things, then there's no conversation to be had. Personally I'd rather have a conversation than not have a conversation, but you do you.
Had this thought this morning after seeing DHH announcing that Basecamp isn't writing code by hand anymore and they're using AI to write Rust for Hey's backend.
To me they're failing to see the even bigger picture that maybe even Hey itself is outdated in the Ai era and not only is its Ruby framework antiquated... the whole Hey service is too.
> The future isn't a bunch of extremely personalized apps - the future is a single AI assistant doing all the work.
I suppose if you break your glasses, squirt real hard, and drop down about 30 IQ points, then sure.
We're extremely far from anyone, even those with the most interest and capability, giving a practical demo of this future you're suggesting.
What we all see irrefutably happening is some new people are just barely able to do slightly more than they used to. That's the impact we'd expect from a better search engine and a better "I'm feeling lucky" button which, of course, is what this era of AI is.
> We're extremely far from anyone, even those with the most interest and capability, giving a practical demo of this future you're suggesting.
Yes, absolutely. This is a very far future. But that future is (maybe, possibly, hopefully) coming, unlike this other future which is absolutely never coming.
Having AI "do the task directly" sounds just as miserable as anything else people keep insisting is the future. We can't rely on probabilistic methods. We will still need humans in the loop to pin down the deterministic parts and also the parts we intended (often the same, but not always).
Humanity has dreamed of being able to control computers by having actual conversations for the last 80 years or so based on nearly all sci-fi movies ever released, and that dream is finally becoming reality. And it's not a coincidence - talking is the most natural, the most human way to interact with a computer, and the only reason we haven't done that successfully before was technical limitations. By which I mean, if not for technical limitations, we'd have been talking to our computers a long freaking time ago. Wait, how old is Siri now?
If AI keeps getting better, then it is inevitable that voice conversations will become the dominant way in which mostbpeople use most computers for most tasks most of the time. It's just part of human nature and there's no way around it. The only way this doesn't happen is if AI never becomes good enough to support it.
Personally, I'm not a fan of voice input. But I want AI to get better. And the future where voice conversations work well enough to be used as the main HCI method is inevitably also the future where voice conversations are used as the main HCI method by most people. I'm hopeful for technological progress, that's all. It's just that you cannot have this progress but not have people take advantage of it in this exact way.
Compare it to the future where most people vibe code their own apps, which has zero chance of happening no matter how good AI gets.
> the model of an OS running various off-the-shelf apps for various tasks will soon become obsolete
"ChatGPT, please send <username> a message with the following contents: <content>. Oh, what platform? Create the platform to send the message for me. That username doesn't mean anything on this brand new platform? Uhhhh" [insert step where the AI-psychotic decides to give up on communicating with humans and tells ChatGPT to roleplay as the person they intended to message, I guess]
I hate that web platforms and other low level cloud infrastructure are called "apps" these days. Makes it impossible to talk about actual end user apps, vs. all the other code that's also running in addition to actual end user apps, which will remain in place even if actual end user apps become obsolete.
In other news, switching from gas guzzlers to electric cars had the unfortunate side effect of highways disappearing.
> the future is a single AI assistant doing all the work.
Sounds more realistic than what this blog post is proposing. Meta also tried releasing a phone, how did that go?
Muse on the other hand is the polished version of what OpenClaw should have been and Meta appears to have given a lot of thought about it more than this blog post.
He talks about applications he generated without specifying what they are or do. Does he really believe everybody is going to generate their own text editor, paint program, file explorer or graphical windowing environment with A.I. just because they can?
I believe the author is misguided about A.I.'s capabilities just like many analysts and investors are.
A.I. (or should I say LLMs) is useful for specific problems but contrary to its name it has no "intelligence."
> That makes a lot of sense in a world where we’re importing all our software from strangers. It makes less sense in the world we’re heading to, where most of the software we’re carving up fiefdoms for has the same provenance.
If I'm running software written by an LLM, even if I was the one who prompted the LLM, I would still want my OS to treat it as if written by a stranger.
I know he's not necessarily talking about relaxing security models here, but he's not being very specific about his vision otherwise.
I think OP may be onto something real, while you are definitely right if we translate concepts directly. But maybe the challenge, with current platforms included, is because we try to cram identities by reference. If a phone had its own identity, with which you transact as with any party then lots of the security concerns would dissolve. To tone down crankiness, I refer to identity in the technical sense, not personality etc. This doesn’t solve the challenge but it might point to a different foundation for a security model.
Ownership does not imply co-identity. If my phone harms someone else, without me being responsible but e.g. the brand that updated its firmware? Your questions are good, and I don’t have the answers. I just think our current model doesn’t either.
I mean, everything should eh sandboxed imo. Even OS included software.
Problem with sandboxes is that it doesn't solve the "user who doesn't know or care" problem.
If you have to give access to particular files to an app, your average user will just give access to all. Someone installing tiktok doesn't look at the permissions...
The best thing would be for apps to operate on their own copy of a file regardless. But that doesn't stop bad apps from leaking importsnt files. Unless files can be signed as being for x thing requiring y special permissions as granted by the originator and not the user.
The crowd on here we can't to be able to truly own our devices. But I think for the safety of the average person they should be locked down by default, same as dev mode on android devices where you accept liability for doing silly things.
Nope. There has never been a "good" post written in this genre ("I'm leaving my last team, here's my new project") and I'm not going to waste interesting bits on my pro-forma transition post.
I didn't submit this! Happy to chat about it, but I feel like I was pretty clear in the post that I wasn't bidding for the front page.
I doubt that we are entering a world where people will be using AI to write their own little apps, and that is coming from someone who does it.
All you need to do is look at history to realise that. We had things like BASIC, HyperCard, and the early web that attracted droves of people who were excited by the opportunities. While a lot of those people went into industry to write software for other people, a lot did not. Most people, I suspect, didn't even explore the potential because they simply don't think of computers as programmable machines -- even if they were aware that there are people who program computers for a living.
The barrier to entry is lower now because of AI. Even though I code, I ask the AI to write a lot of little helper apps for my wife who homeschools our kids, and I don't code review many of them. I only do enough user testing to ensure they function as requested.
Armed with this knowledge, I have tried in vain to get various family and friends who are not technically minded to do anything like this. In some cases, they just tell me that they can't think of anything the AI can help with.
Many things that seem simple to them are more complex, and they can't tell the difference between simple apps that the AI can knock out in a few minutes and larger projects that require internet hosting and/or vastly more AI credits than they would ever be willing to pay for.
By far the biggest problem is that many of them are not very good at describing what they want. For simple apps, you can go back and forth with the AI to ask for changes, but the more complex the solution, the more its becomes like a traditional engineering design project where carefully though out design of features and their interactions are critical for success. The new "code" may be more like human language, but the user facing details must still be specified.
The thing is the LLM interfaces encourage people to write their own apps.
Claude especially, where you might say “how can I do task xyz more quickly” and Claude responds with a little JS app. This creates an expectation that it’s possible to do anything if you ask Claude right.
I have seen this first hand where my partner - a deeply non technical person - came to ask what an API token is and how she can get one for Canva. Claude was busy automating her Canva -> Pinterest workflow.
She now has a half dozen of these things and hunts through her Claude chats to reuse them.
In my opinion, an OS that hosts a local agent can produce things for you on the fly in response to a particular task in addition to the baseline set of features.
Too often have I ran into "you can't do that" in operating systems, especially Macos where you either have to buy a paid app to implement what they should have already, or what you want is niche enough that you have to build it yourself.
Plus and agent can operate across your hardware including the cloud. I could ask something like "graphics cards are expensive and always out of stock so when some pop up can you notify me and show me the price etc" and then receive a rich alert on whatever device I happen to be using at the time it happens, once agents are aware of where we are and what we're doing (yes, it will be pretty dystopian with current governments and the drooling animals who vote for tribalism).
Or being able to be like "I want my photo gallery to show animated thumbnails when I've taken a series of photos of the same subject. Oh, little bit slower like one every second thanks" boom done.
And much UI will becomes defunct in the first place, UI only exists because of many annoying discrete steps we have to do to get something done. If I can just ask an agent to do the thing I want then I don't care about that.
It's like how I only need to make a PowerPoint now and then and every time I open it I've completely forgotten where all of the stupid fucking buttons are in the stupid fucking ribbon bullshit they made.
I don’t think the post has a lot to offer, but I will share thoughts on how a modern operating system can add _a lot of value_ both for non-agentic users and for agents.
Looking towards macOS and iOS for example, an underrated feature for the “it just runs your web browser” crowd is all of the system-provided databases. These turn rather specialized systems - photo libraries, music streaming, calendaring, contacts management, and mail into shared databases that link applications together.
On iOS it goes much further - because of HealthKit, my Withings scale, my Garmin (when I had one), my MyNetDiary nutrition tracker, my continuous Glucose Monitor, and my third-party analytics apps share the same access to my data. And it is done securely - I can choose exactly which part each sees. This is a tremendous resource for both developers and LLMs.
My calendar, hosted on a personal server but available through Calendar, is transparently available to my to-do list app - so I can see my appointments, my tasks, and even my workouts (thanks to health kit) in one view.
Weather is a system service on iOS. Calendar is a system service. Contacts is a system service. Health data is a system service. Even music - third-party apps can control and consume your music library on iOS - is a system service (if you use it).
My Mac knows how to develop RAW files out of the box, and it can connect with integrations to get them printed, and use my library metadata to sync my favorites to my phone. I have a third party app that syncs my Apple photos library to my NAS.
A good OS, like iOS or macOS, could be the fabric that holds all your vibe-coded and ephemeral LLM apps together. I see it making a whole new category of things possible.
For the record: I didn't suggest this post did have a lot to offer. I didn't submit or upvote it. I pretty conscientiously kept it simple and abstract, because (as I said in a different thread) this genre of posts ("I just left the company I'm highly associated with, to do this new thing") is deeply cursed.
You still gotta write the post, though, because otherwise what you get is a circulating Twitter post.
I'll write something more fulfilling about this next week, I think.
But so much of macOS is also uninstrumentable except via undocumented APIs that are unreliable at best. Why wouldn't Linux or even FreeBSD be even better? They're not limited to content but even the source that can be updated and rebuilt to do literally anything.
Honestly, as a long time Emacs user, agents are amazing because when all your functionality and data lives in open source code and text files, the world is literally your oyster.
Linux and the BSDs might be more instrumentable, but the rest of the experience is so fragmented in comparison. Regardless of category, there is no single service or data format that everybody has agreed to use and build on top of.
The advantage of the Apple approach is that the user’s data is instantly available across third party apps. For example, instead of needing to build (or import) a whole CalDAV apparatus to be able to build a little calendar-based todo app, you can just use the system calendar API and let the OS do the heavy lifting, making your app much more lightweight and instantly usable with almost no user configuration. This naturally applies for scripts and agent skills, too.
> Why wouldn't Linux or even FreeBSD be even better?
Because Linux doesn’t offer a rich set of system services built around the needs of desktop users and exposed through curated APIs.
Linux can’t be better at providing a shared calendar API or a shared photos library API, or one of dozens of other APIs because Linux just doesn’t have those features.
Some particular DE could offer those features, but then who is going to build the app ecosystem around that? And isn’t it going to suck when you want to make a Linux native photos app but it doesn’t work on Ubuntu because your LLM was trained on KDE?
> Forget the map, just tell me if I should get off the Eisenhower Expressway and take Roosevelt, Madison, or Lake back home?
> Is there a meeting happening anywhere right this minute that I’m supposed to be in?
> You can go on and on like this. These all sound trivial. That’s the point. They’re normal-life problems nobody would have built serious programs around, because the audience sounds too small.
Waze does not answer my commute question for me. Waze finds what it believes to be the quickest route from west loop to my house, but that route is virtually never reasonable. There's only 4 reasonable routes; I just need to know which one to take.
It’s not the fastest route, because the algorithm is stupid as a result of trying to be smart.
It sees a small side road and thinks “I will take the user down here because there’s no traffic”
But it doesn’t consider that the side road is empty because you’ll have to turn across four lanes of traffic in rush hour. So you get stuck for twenty minutes.
You get that this is a commute, right? Not charting a course through unfamiliar territory. A drive I've done thousands of times. I don't care if Waze thinks I should be on Harrison or Race. Fuck you, Waze. There's 4 reasonable routes; I just want to know which one.
It's an app with an audience of just me. That's the point. You're not supposed to want it. You have your own problems.
Apple literally has automatic commute notifications that tell you the optimal route (15 minutes via Easy St.), in this hypothetical world where your agent knows enough about you to suggest today's route, why doesn't your digital panopticon purveyor (be it Apple, Google, your car manufacturer) automatically pull that info the second you get into your car?
we could spend time answering that question or, as tptacek proposes, diy the solution because the app making app is right there, and answers to your input not Apple/Google's
The idea of software provided by domain experts has to continue even if you can write software to do what you want by chatting to AI.
"Write an app to provide psychological advice to people with mild anxiety, then market it and sell it to people with mild anxiety."
If an entrepreneur did that with no domain expertise, imagine the potential liability.
How do you know the output is correct for anything that matters more than completing simple and repetitive tasks? You still have to know an expert in a domain has vetted the software. The result of applying AI to OS development is making itself fairly clear with MS I believe.
> Is there a meeting happening anywhere right this minute that I’m supposed to be in?
Nothing about this is groundbreaking and could have been written in BASIC.
I grew up on BASIC using a BBC Micro in the 80s, using a QWERTY keyboard and a screen. Some decades later I'm using a QWERTY keyboard and a screen with a few more pixels. I'm betting nothing will change much. An OS will do OS things, just a lot faster and more securely.
Spreadsheets will still be useful for making deterministic calculations about finances. It will just no longer take 100 engineers to build one.
GIS software will continue to use deterministic algorithms for calculating areas because it's a useful function but, hopefully it will get more user friendly.
Most people don't have specialised computing / software needs or uses. In fact, they're generally Overwhelmed by the current software landscape. They want less, not more tools.
And to them software is generally a not much more than a home-screen bookmark to some client that could be a website.
> It’s not simply to “provide access to hardware resources”; plenty of embedded systems do that with runtimes that have no business calling themselves an “OS”. No, the core purpose of a modern operating system is to partition different applications off from each other, and carefully control how they can communicate.
First of all the initial definition is totally correct, and if your embedded function needs to call an intermediate layer to access the hardware that layer is an OS. Second of all, all those "purposes" are exactly achieved by providing _managed_ access to hardware resources.
And that is valuable in whatever world we end up living in. Sure, 10 years from now my bank app and my Hacker News browser may both be written by "my LLM" and tailored to my use cases and desires. Will I want them to share memory? I don't think so.
In recent times, I've palpably felt the excitement of being able to imagine like I was a kid again. BUT here's the reality I've simultaneously been hit with:
Most people, and I wouldn't be surprised if it was 90%, who've had any interaction with AI are still stuck on chatting with a chatbot. They've got no clue how far you can now go with a few well written markdown files and/or prompts, all in natural language!
So while I mostly agree on your outlook, and think it's super exciting, I've also felt firsthand that building with AI hasn't yet diffused as much I'd hoped it would. It's still concentrated in the hands of people who have an aptitude for building in some way, shape or form.
The question is: How do you get more people, who aren't power users or builders, to start building with AI? Maybe it starts with getting them to see that building to solve their own life's problems makes life more interesting?
I look forward to following your and Kurt's new adventure. I love Fly in no small part to your writing for its blog. Wild to see what feels like a complete changing of the guard at Fly, at least from the outside. Your post resonates; reminiscent of the Ink&Switch talk from local first conf that I watched today.
I'm tired of reading about the awesome future that could happen, but doesn't. If you think this future can happen, go and make it. Surely it's just a few Claude prompts away? Or surely you already live in this time you describe, so it's not the future, but the present?
They don't really have to fight, most people these days accept their role as a guest on their hand-held computer.
The hand-held computer tells them what they're permitted to do, and they conform.
Sometimes, one of the users has a non-conforming idea and says "I'd like to be a superuser on my computer" and they're shouted down with a chorus of "you'll be hacked" and "nobody will be able to trust your computer and you'll be excluded from online services". Sometimes the organisations saying this even use the excuse that "restricting the user is essential to maintaining the user's privacy".
The OS is the distilled wisdom of other experts. Wisdom that neither I, nor the current generation of next-word-guessers, could possibly keep pace with; no matter how many reddit bots the frontier lab operates in its attempt to extract complex concepts into simpler vectors. It will be the collection of things that are so unbelievably expensive to reinvent and re-learn for each machine and user, that they are not subject to to the market forces that come with making the slop cheaper. Any well-maintained OS will maintain that role. And be ever more important in the future, as apps diverge into even lower quality and even less shared benefit from improving them (for the other approx. 2 users).
The OS is a giant collection of legacy compatibility shims which many people agree isn’t how you would design things if you were starting from a clean sheet of paper today.
If you look at software like sophisticated database kernels they bypass the OS almost completely because they can reimplement a better OS in userspace, using the kernel as little more than a device driver. Over time this has become more common, not less. OS like Linux have actively embraced this role.
Being able to effectively specialize the OS for specific applications has large benefits. It is a specialized skill but not unusually expensive to implement.
I agree with the first part of the thesis (AI means unlimited custom software for power users who are able to describe exactly what they want) and strongly disagree with the second part (this means we don’t need any of the technology we’ve developed for software development, or distribution, resource management, etc.)
I love the idea that normal people building software to solve real problems in their life is, necessarily, a toy; the only real software is blessed by our sacred priesthood of PLT experts.
> Don't be discouraged if what you produce initially is something other people dismiss as a toy. In fact, that's a good sign. That's probably why everyone else has been overlooking the idea. The first microcomputers were dismissed as toys
There won't be an OS anymore. All you need will be a network enabled graphics driver (Hello X11) for the AI to connect to and draw on.
Then the AI will just create the image that resembles a spreadsheet or whatever on the fly and write straight to the framebuffer.
All other tasks that go with the spreadsheet will also be done on the fly.
But I could imagine our language will change and certain words will basically act like macros and create a set of assumptions and specifications for the prompt that you're using.
Uncharitably... because eventually they hope to produce a phone that isn't running iOS or Android?
More charitably because the title is actually an important question right now in the abstract for almost everything, not just operating systems.
But the forces that shape the question are so large and hard to get a handle on when we are living through the moment that you have to start thinking about it somewhere, why not OSes?
I think people will still buy templated software - applications that come with source your AI modifies or plugins your AI writes. Most people need Excel, not a badly written clone, but they may want a bunch of custom plugins that can deeply modify the UI for their use case.
Is this how the internet finally dies? Not with a bang but with the whimper of thousands of AI slop apps scrapping websites into oblivion?
Most of these problems the slop apps can solve are dependent on gathering information to answer questions. If no one ever visits weather.com to get weather information, ad revenue will completely dry up and the site will go under. Then the slop stops slopping and I can’t get my hyper local weather app to work.
I’m not going to go as far as to say this article is wrong, in fact in this cursed timeline it’s probably right. I just don’t see how it’s good.
Ironically this may finally force the fulfillment of the Web 2.0 paid API service dream. Remember when people were pushing companies that just provided a paid API?
There might very well be more of that. In a world where everyone has agents filtering data on their behalf, it seems more complicated to figure out a way to force a user to view certain content along with their desired data than to simply refuse to supply data until a small transaction completes.
On the other hand, that’s just traditional banner/video style advertising where the goal is to force specific content on the user. If someone wants to spread the word about their product or service, I’m sure they’ll pay money to advertising services that use agents to inject their message into the global context. This is happening already, we just talk about it in the frame of “dead internet theory” and AI influencers, etc. Anywhere that content is being produced and consumed is an opportunity for the message to be nudged in favor of the person paying money.
So we go from the days when we have to pay our ISP so that we could go to weather.com to see it if it's raining outside, to a future where we have to pay for our ISP, so that we can connect to the AI company who we also have to pay, so that we can have their AI connect to the API at weather.com which we also have to pay for, so that we can see if it's raining outside? I'm not convinced that'd be making our lives better through technological progress.
Oh, ok. I guess computer science is finished, then. I should have known. It's not like there's any evidence that it's been blown wide open recently.
(The irony is that all the reading that's gone into our design is like, old CSAIL papers. Maybe it's true! Maybe everything we're doing is just crate-digging through old settled computer science. That would be on brand. I should find a way to work Athena into it. Maybe we'll use Zephyr for messaging.)
I guess it really is over if this is what professionals (supposedly) seem to think.
OP is right though, an OS remains an OS, same as it’s always been. What’s changed is how we interact with it and what we lay on top of it. Nothing magical or philosophical about it.
I think operating systems have changed. They don't interact directly with hardware much anymore, and the hardware is filled with SoCs all running systems of their own, sometimes with entire network stacks, communicating with devices and each other, collecting and processing data, often totally invisible to the operating system the user gets to work with. The operating systems most people use have gone from giving users control over their computer to being a sandboxed interface to manage files and some of what their computer is doing.
Could you elaborate on what you mean by "don't interact with hardware much anymore"? If you compare modern drivers with similar ones from 20 years ago, for example, you'll find the modern ones tend to be quite a bit more complex.
No directly anyway, for the OS it's all abstractions and isolation. There was a time when the OS could basically see and control everything, but now there are subsystems all over the place with their own memory, their own software, all outside of the control of the OS. CSME runs while the computer is powered off.
There could be some interesting actual OS design innovations for new permissions, access control, and sandboxing systems that are designed from the ground up for these "conjurable" experiences (or even OS components themself).
Finding some balance between access and safety for vibe-conjuring parts of the OS in a way that's still powerful but won't accidentally nuke or corrupt itself.
The historical baggage is so heavy I don't think anyone can realistically throw it away, so what we'll end up getting are almost certainly incremental improvements to existing foundations.
I do think there's some interesting space to be explored in capability-based systems, but even then I think it would mostly end up looking similar to how existing systems are, just with one more element added
The more AI is wideapread and is used by power users to build apps, the more devices will be locked from running any custom ("non-certified / non-licensed") software. For your safety and security, of course.
If you give an ape a slingshot he might think he can take over the entire world. But he doesn't even know what rubber bands are, or what it takes to make them.
What I did on my computer, any power user can do on theirs.
Back when Google was the newest thing, I would Google all the obscure things that I could think of, trying to fill in gaps in my knowledge. Trying to verify urban myths but usually debunking them, etc.
Any power user or even any user, could have done that. But most people who considered themselves critically thinkers actually did the opposite, digging deeper into all the harmful rabbit holes we see today.
A lot of things can done "with just a few steps" with AI but I think it's underestimated how hard it is to come up with those steps.
Software in the future will have to be building blocks for agents to extend. For my projects I decided to generate LLM targeted guides in the docs/ folder, to teach agents quickly "how to" use the project as a framework.
How to implement an MVC view, where to place the files, a guide how to write a good backend route, how to extend the schemas, and other articles like that.
In my opinion we've already reached "Star Trek like computers". Our OS will have to accommodate for better sandboxing and better permission management and better role separation policies, because currently that whole security sector assumes that programs will be static and won't change before/after an approval.
We need to figure out how to better sandbox programs in terms of not only control flow, but also in terms of access to _data_ on a user's behalf.
It's the thing that runs your AI agent, that runs your AI pocket bot (AIBI looks cute btw), that runs the software doing the training of AI models, that runs the datacenters doing the inference, etc.
There's still data. There are still bits moving. Filesystems and networking aren't going anywhere.
In a charitable reading the title is something like "what is even a desktop OS now?" or "what is even a consumer OS now?".
As for the new project, I've got one bet: there's Linux involved to host it.
Linux is an OS.
P.S: as a sidenote the entire "users don't install apps anymore" was already true to some extent. A shitload of users are totally fine with an OS that runs a browser and that's it. And they're fully operational. And we still all know what an OS is, even for, say, a Chromebook. Heck, I'm not even sure "installing apps" as ever been, except during the first install of the OS, what an OS was ever about.
This reminds me of WWDC 2010 when Steve Jobs said developers using private APIs 'know exactly what they're doing'. Good luck with your project, but I think you're vastly underestimating the problem. And many OSes are already integrating user generated AI apps.
Taking it at face value, it seems to me this will also collapse and get folded into the model.
Unclear what will be left next to it. It will just live on your phone and take care of everything. The apps will be markdown files describing what they do, probably compiled ahead of time for latency.
But the model will be in control, so what else is left to implement besides giving it bash, a compiler and a display?
I guess there is value in the year before that happens.
I agree with the gist of how weird everything is going to get, and how we are living in a time where people haven't come close to internalizing the shift.
IMO even this AI phone is doomed in the slightly longer run.
One more step into the weirdness and people will just be working with AI to build their own bespoke phone with the specs sent over to a manufacturer that is "like PCBWay, but spits out whole devices".
I have no idea what this does to society, the economy, etc when we reach a point where nobody really "needs" any kind of middleman for almost anything (well, except for things they actually really need, like food and shelter), but it looks like I'll see it happen for a bit before I die.
Hardware is still constrained by compromise. If I want a screen for movies with a battery that lasts and doesn't burn my leg when it's in my pocket, I give up a physical keyboard and nice speakers and a bigass physical camera lens.
The physical device that is best for me is probably the best car compromise for several million other people.
if my ex calls and I want my glass brick to flash the "help me" in SOS on the screen, play Hector Berlioz, and dim all the lights in the room – I can have that. I don't have to want the same thing a million other people do.
> Consider: what does computing look like in a world where many (maybe most) applications have an audience of just 1-2 people? How is software distributed? What does it run on?
The same. The slop apps remain just that. They may do something useful for you, but that's it. A bit like a branch in a repository.
It's the same thing as looking down your nose at people who use Excel as a database because it's not a "real" database.
Yes, I'm sure we all have horror stories about spreadsheets that got out of hand, but that ignores the thousands (or more) spreadsheets that are silently productive everyday.
That's a pretty good simile in fact. Excel programs are very useful to the people who write them (including myself sometimes); and yet - almost none of them are widely shared, becoming fundamental tools.
Hey, all. I really don't know what to do with a post like this.
I'm being sincere when I say (as I've said on two threads here) that this genre of posts --- "I'm leaving this company I've been very publicly associated with, and here's the new thing I'm doing" --- is deeply cursed. There's no way to say anything interesting without it just stinking like an ad for the new thing.
Obviously, anything at all you say about a commercial project you're working on is easily read as promotional. And you're right, this kind of writing almost always is promotional. But there's a way to do it where at least you're trying to be in conversation with your peers, rather than hitting people over the head with how awesome you think the project is.
But I don't know how to do that in a post like this. I think the only way to read it is as, like, an investor memo. Not my goal, but I don't make the rules.
So my strategy here is just to stay kind of vague, and talk about where I think the world is going, rather than the specific thing we're doing. I can talk your ears off about capability systems, datalog, models driving hardware, virtualization, whatever. Those are fun conversations and I'm very psyched to have them; it's what lights me up about the work we're doing now.
But I don't think it can work here. I didn't submit this post and I didn't upvote it. I wrote it because I didn't want the whole thing I'm leaving Fly.io for to be wrapped up in some dumb Twitter thread.
If you're unsatisfied with the post, I don't blame you, but it's less a bid for the front page of HN than it is an update to my "about me" page. I'd literally rather talk about HN meta, and how to write for HN, than I would about operating systems at this moment. I truly appreciate the interest though.
You are overthinking it :) You are one of the most interesting people on HN. I am sure others think along those lines too and would love to learn more about what you are building and where the industry is going. Others can just tune it out.
I really appreciate the post.
I completely agree that we haven’t even gotten past the first post yet in terms of the magnitude of change that is likely to occur. People fail to comprehend that we’re still assimilating the Internet as a means of communication. Cell phones. It will be decades before what we have today is fully utilized, let alone what is likely to be developed. Anyway, thanks for the article.
Personally, I'd like to hear the full pitch, and I suspect many others would too. If the ideas are interesting enough for you to dedicate the next N years of your life to, I'm probably willing to spend 10 minutes reading the blog post. Particularly from you, a known-good writer.
We were on a call recently where I mentioned that as a fly user SDLC and declarative config was more important than ever, and so I was pushing for sprites to have OCI images.
I'm pretty AI pilled, and I'm building a local-first personal software platform, but I dunno... I've been feeling like i need to cram the "AI spirit" into very specifically shaped boxes to actually build robust software tastefully at speed. Like software on the outside and AI on the inside instead of the other way around.
I do think the future of computing will definitely be more like the Linux diaspora than the apple ecosystem, but also.. it is easy to get "duped" by AI without actually shipping. I'm starting to want "KPIs" to help me measure if my time is actually well spent against the things i care about.
I guess I'm extrapolating from your post, but even if the programming language is English, we'll probably still want to spend a large amount of additional compute to produce "artifacts" that have things like performant code, security hardening, and certified taste.
I do think the future of computing will definitely be more like the Linux diaspora than the apple ecosystem, but also... when i first got into linux i spent way too long trying to get my desktop perfectly aesthetic because it was all suddenly malleable. Now i vibe more with Debian.
I wish you luck, and I'm looking forward to seeing what you come up with! Fun times ahead for sure.
> I've been feeling like i need to cram the "AI spirit" into very specifically shaped boxes to actually build robust software tastefully at speed.
This is such incredible verbiage. Thank you for saying something so clearly that I've been saying myself in many other ways. I've been using terms such as "providing structure" and "applying constraints" and "reducing degrees of freedom".
I think very specifically the idea of "taste" is going to be extremely important in these AI-saturated days, and it's really where experienced humans will shine...at least for a while.
A post a few days ago turned funtional core, imperative shell into deterministic core, non-deterministic shell. And a comment [1] on that discussed turning that into "slop core, artisinal shell". Which makes a lot of sense actually despite feeling the wrong round initially. And maybe what an OS needs to do is to support that, at a higher level. What I think we will need is strong isolation - all the software might be coming from the "same place", but do we trust that place? Certainly not currently!
1. https://news.ycombinator.com/item?id=49784704
Hi Thomas, I think you're being too defensive. It's ok, I look forward to your technical posts, on what you imaging this OS powered by AI to be. Will be a fun ride.
If you're looking for a UX idea, when the phone boots, it asks "What are the top 5 things you do?". And that's how the phone experience starts.
All the best. (Ignore the silly haters, they're lost in their AI pessimism to just look forward to a new intersting thing).
Oh, I'm not worried about the haters, they're the air we all breathe; I'm just neurotic about my writing.
But as an AI skeptic, you optimists are the air that I breathe. So we’re just breathing each others?
My 3ct:
1. When you write a comment in a discussion of an article, be upfront with being the author. It took me a while to understand that from your comment. "Author here" is a thing.
2. Your article itself doesn't have clear authorship stated. At the very bottom there is a copyright notice, but that's pretty generic and not a single person either. Just place your name at the top together with a date. Not everybody knows you or follows your blog.
3. When I read this ...
> When I was a little kid, a family friend sold us our first “computer” — scare quotes because I’m pretty sure it was a VTech Laser 200 clone, a chiclet keyboard Z80 that ran off cassette tapes and plugged into our TV.
.. I prepared to stop reading since .. how is that not a computer? After another paragraph or two which just felt like rambling, I gave up.
no idea you were working on fly !
As someone who's building its news project on it, it's both reassuring, but also not since you're leaving..
This is absolute bullshit.
These are bad-faith arguments toward "What about self modifying software".
Self modifying software has no grounds in reality or predictable behavior. It is an unreasonable expectation that software is magic. You would not fly on a self-modifying airplane. This is not a future i want to live in.
No; sorry, no.
The author, as a kid, believed computers where like a genie that could manifest anything they ask for. It seems to be a case of wishful thinking of a kid.
But it seems that he is falling for the LLM hype the same way. We all have the kid version of ourselves hiding somewhere within our grown versions, it seems...
As long as the software is proven correct, who cares how it achieves that?
How can it be proven correct if it's not deterministic? Isn't this the halting problem?
You seem a bit confused?
First, the halting problem is only undecidable on arbitrary programs. Here we would be writing programs specifically to be easy to prove correct. In fact, we would be developing programs and proof at the same time.
Second, determinism is a nice property, but it's independent of whether you can proof software correct.
Have a look at eg randomised quicksort or the 'expected linear time minimum-spanning-tree algorithm' https://en.wikipedia.org/wiki/Expected_linear_time_MST_algor...
Or have a look at HyperLogLog. https://en.wikipedia.org/wiki/HyperLogLog
HyperLogLog only gives you a probabilistic approximation of the true number of distinct items. Here the proof would be about the probability that you are within some acceptable window around the true answer.
Can't the non-deterministic LLM produce deterministic software? Assuming it is not allowed to modify the theorem proving software.
Agreed. Read the article through and felt cringe most of the way through.
> "But it booted straight into BASIC. That’s all it did. I was, like, 8 years old. I wasn’t about to learn BASIC. What I learned instead: computers were not as awesome as I’d imagined."
You were the exception. Most kids felt awe. We started learning BASIC and creating "dumb games". That's the origin story of most people my age who ended up in this doomed industry.
I think you were the exception. Most kids never even knew what BASIC was. If there was a game already on the computer, they'd be happy to play it. If they really liked the game they might learn how to start the game if it wasn't already running, but they didn't care about doing anything else. Kids who were into computers looked at a computer and saw endless possibility. Everybody else played sports outside, watched TV, or hung out with their friends.
> Most kids never even knew what BASIC was
Yet magazines and books commonly published program listings for games in BASIC which many kids somehow managed to type in and run.
The kids who loved computers were the ones plugging in line after line of BASIC. Well, them and the ones who just really loved games. Games were a gateway drug for a lot of kids because for so long games could really make you work for it. Games had kids making changes to config.sys and setting BLASTER environment variables, connecting to a BBS or FTP sites for downloads, and troubleshooting network issues to get LAN games running with friends. You could get pretty good at computers just getting your fix. That took a pretty strong commitment though because if all you cared about were games, and if you had them, consoles made gaming very simple. Too simple if you valued the sense of learning and accomplishment that came from winning a wrestling match with an unforgiving computer keeping you from the game you wanted to play.
I was one of the kids who didn't have a computer at home but had friends with those, and would read the BASIC listings in the magazines many times over to figure exactly what they did and think about what I could change before I visited my friends and started typing it all on their computer.
Yeah, I did that too. That's what you did with these machines. Deeply unsatisfying! Like: a real lesson in how gnarly real computers actually were. This is my point. You got one of these thingies and thought you could do anything, but it turns out "anything" took on values ranging from "hours to many hours of typing in BASIC listings and then working through inscrutable errors until something showed up on the screen but then crashed". Computers sucked ass!
Until like a year ago.
Different strokes, I guess. FWIW, I also started at 8 -- with a Commodore 64 and a Commodore Vic-20, both connected to a television -- and I found the process of trail-and-error learning my way through BASIC to be absolutely awesome.
I would start with a completely blank slate and then, gradually, be able to create something out of nothing. If I was clever enough, it could do whatever I wanted (as long as my wants were modest). It was magical and I have nothing but fond memories of the time I spent doing that.
Did you end up writing a word processor? :)
Haha! I did, but not for a few years, until I took up Pascal and MacOS and then wanted to find a way around the 32k text limit that was a stubbornly difficult problem to solve for a while (and was the original reason that BBEdit became so popular).
Trying to correctly render a vertical blinking cursor on top of proportional fonts took me ages to get right...ish.
That is some next-level motivated reasoning
Aside from the Spectrum, the Amstrad CPC (CP/M!), at various times a secondhand BBC Model B and then later the family PC, I wrote complex programs for a Sharp PC-1246 (because I had the cassette interface).
https://www.computinghistory.org.uk/det/1421/sharp-pc-1246/
This was the very opposite of unsatisfying. I don’t recognise this idea that computers sucked.
AI-assisted coding is, to me, way more soul-destroying than hitting the limits of a simple home computer. In most regards I feel like I have reached the end of my interest in computing, and the pivot I had in mind — teaching - looks like it has been joy-deadened too. Nobody is going to want to learn how anything that is not AI actually works, tech people insist that AI will replace (they say augment, they mean replace) teachers and trainers, politicians seem to be listening, and non-tech people now increasingly resent technology.
I barely recognise the industry I spent my life in; I do not recognise the evident enthusiasm for a world that is abandoning a whole, transformative pedagogy in favour of the programming equivalent of games consoles: button pushing vibes.
I am not at all of the opinion that AI has no value, though it does not have anywhere close to the value people insist it must have for me. I am learning, trying to find an ethical angle on it that doesn’t make me feel uncomfortable and exhausted by the hype.
But it is the final nail in the coffin of the joy of tech work for me, and as a single middle-aged man I perceive this as life changing and very likely life-shortening, in the way that psychologically unhealthy things are a health risk for older adults: unless I find something that gives my work life a sense of connected purpose again, I figure I will be gone before the Spectrum makes it to fifty.
So yeah, no. No.
> Nobody is going to want to learn how anything that is not AI actually works, tech people insist that AI will replace teachers and trainers, politicians seem to be listening, and non-tech people now resent technology
I disagree with most of your post, but this one the most.
Non-tech people already resent technology... Try and sit next to a non tech person use a desktop pc. It wasn't until the smart phone that the average person started to use technology like you or I recognise. And that's only because it's been simplified to the extreme for them. (I don't blame them btw).
AI is such a step change for learning and teaching... For anyone who is inquisitive. Ask it endless questions, explore ideas, get it to generate images to help you...
> I don’t recognise this idea that computers sucked..
Same here. I can understand indifference, but this I don't get.
FFS I even loved playing with calculators!
>You got one of these thingies and thought you could do anything..
Why did you think that? Did you see people use it like that? anything? really?
For me, computers became intriguing after I started programming in BASIC. Not before that. Before that I was just curious about what the big deal about these things are...
In the literature, there would be passages that describe how computer followed "instructions", but I never could make sense of that or understood what an "instruction" was. So it didn't really sway me either way. I just tried hard and remained curious to understand what this "instruction" was and how I could give it to computers...
May be your falling the same way for LLMs now..
I don't know probably because I was eight years old.
Good point - what might become fun and engaging at 10 or 11 could be more frustrating at 8, especially on your own.
As near as I can remember, my first real programming experience was when I was eight years old, writing basic programs on an Apple IIc at school in 3rd grade. (I was too poor to afford my own computer until my senior year of college). It was awesome.
I have never found computers disappointing, and I have spent most of my career/life learning about them - from processor design through to distributed systems. Mostly I've been disappointed in the disparity between the number of interesting things I'd like to do, and the number of lifetimes I have available to me.
Different strokes for different folks, I guess.
Figuring out inscrutable errors seems highly educational. I guess I don't have enough data to determine whether your bad experience was more or less common (probably less common on HN though). I imagine some (like yourself at 8) found computers frustrating and bad, while others (some peer posters on HN, who might also have been a bit older when they started) found them frustrating and good!
I did come to appreciate that. I was a professional C programmer for a lot of years! A hand tool woodworker, so to speak. It's great. But it's not the same as what I had thought computers were when I was a little kid.
Why would you get a computer in that era if you didn't want to learn BASIC?
You got a computer in that era because you thought you talked to them and they did what you asked them to do.
Then you got the computer and when it started up you got a flashing K. It was a bit of a disappointment. But you flicked open the hefty manual, started reading and before long you were hooked.
One career later, I now have a computer that is very, very close to what I wanted all those years ago but with a significant and irritating genie attitude to granting my wishes.
Now if we can just tone down the genie stuff…
Kids weren’t getting computers, their parents were.
Yes, there was no good reason for us to have had that computer. A friend of the family sold it to us. I had no say! I was like 8.
To play the games that were included in books and magazines as BASIC program listings that you could type in and play.
It was probably a function of not just personality type but education. For whatever reason I was taught basic in elementary school as a millennial and it left an impression.
Education has to help demystify computers and give people an idea of what they could do with one. Basic seems like a weird choice for a millennial, but at least it had you actually write code instead of abstracting it away into pushing around icons like some of the systems I've seen for teaching young kids about software or even just giving directions to turtles.
Why is BASIC a weird choice? It wasn't really a choice at all. All the computers in my school ran it until I was 8 or 9 and the school board could afford to upgrade to Windows machines.
If you define a millennial as someone born from 1982-1996, the majority of those kids by the time the had a computer lab (say by the time they were in middle school) would have been too old for BASIC. You are talking 92-06, Windows 95 took over the majority market share by the end of '96, and you are forgetting a whole lot of schools had Apples being pumped in, so a lot less access to BASIC. Unless you were in a programming adjacent class, the majority of millennials weren't learning basic. Ask your non tech millenials how much BASIC they learned and they would look at you like you grew an extra arm.
In India, I think we still had BASIC taught in schools and private coaching centers in 95-98.
Depending on where you draw the line for millennials the school computers could have been running anything from Windows 95 to Windows XP. BASIC wouldn't have been included in XP or 2000, and depending on what exactly they were using, it may have been easier to use something with more features. QBasic for example wouldn't have syntax highlighting. Simple programs in languages like python or perl can be used to teach the same concepts, with fairly similar syntax, and would have been more applicable to software people were actually writing at the time.
When I was in high school we used VB6 and VB.NET on Windows XP. Does Visual Basic not count?
10 print "hello"
20 goto 10
10 print "<brother's name> is dumb"
20 goto 10
s/brother/teacher
Your priors are all out of whack. Most people never learned BASIC, let alone at 8. Programming is much more accessible now than then, but the overwhelming majority of people still don’t know any JavaScript or python or anything else.
Programming (by humans) is less accessible in 2026 because most personal computing devices don't boot into a programming environment and they don't include a user-friendly programming instruction manual (graphing calculators may be an exception however). And before the era of "there's an app for that" it was assumed that users would need to write their own programs, so systems were designed to support that. Vibe programming (by LLMs) is more accessible now however.
BASIC was designed as a simplified, interactive version of FORTRAN that non-experts could learn in an afternoon, but which was still powerful enough for a wide range of applications including numerical methods, simulations, data processing, and games. It was also small enough to implement effectively on a timeshared minicomputer or an 8-bit microcomputer.
Even though the Mac was GUI-based, for a time it shipped with, and promoted, HyperCard, which led to a brief explosion of user-created programs developed as HyperCard stacks.
>most personal computing devices don't boot into a programming environment
Yes, but now everyone and their mum has a personal computing device, and there's tutorials readily accessible, and five seconds of googling can bring you to an online programming environment.
All true, but the onramp isn't included in the box. When you get a smartphone, how do you know that you could even program it? Is there any encouragement to do so vs. just buying apps? They are more like game consoles, in more ways than one.
As I mentioned, I think graphing calculators might be an exception.
In the mid 80s, something like 10% of US households had any computer at all. That the onramp is in the box doesn't really matter to the people who never saw the box.
I guess the question is how you measure accessibility - per device, or across the population as a whole? In 2026 several billion people have personal computing devices, but the devices no longer provide that programming onramp. Interestingly enough, it looks like ~25% of the world's population is still offline.
I had Turbo C as a young teenager, but really learned C at the csh prompt of a terminal connected to a giant DG/UX computer in my first summer job.
I actually still don't know BASIC.
Unix/Linux historically gave gave you a shell prompt (allowing you to type in shell programs) after login and a C compiler in the box. Modern Ubuntu or macOS technically still have terminal apps and shell prompts, but usually boot into a desktop GUI.
You were the exception too. A PC was one of the most expensive items in my house when we bought it. The computer table it sat on was the second most. I was so afraid of doing anything on it that'd break it. Heck I freaked out one time when I shut it down from the power button directly.
My Acorn BBC Model B booted straight into BBC BASIC as well, and I too thought it was awesome and taught myself BASIC straight away, and then 6502 assembler and latterly BCPL and so on. I was even younger than OP.
But I think we were the exception, even if we didn’t realise it at the time. In hindsight, my interest was most definitely not shared by my peers.
I remember getting magazines with basic games, where the full syntax was printed out. I would type it in and save to floppies on my Apple 2e - which my mother purchased second hand.
I'm not even that old, but totally agree with this sentiment. Going to miss those times. It's amazing being able to build whatever you want now, but it's bittersweet because if you truly wanted to learn how to do it, you could do it yourself.
I don't understand this. Back in the early days of the internet, you absolutely couldn't learn 'how to do it' without finding obscure books or knowing someone. Now though? You can learn all of it, and find the resources almost instantly. Want to build a computer, self host, program, you have millions of resources at your hand.
Things that are easy to do don't necessarily mean that's the right way to do it. Grueling work sometimes helped you learn far more than just getting the answer to something easy.
Back in the early days of the internet, there were FAQs written about everything. You could talk to people who actually knew what they were doing, as opposed to today.
>you absolutely couldn't learn 'how to do it' without finding obscure books...
QBasic's bultin help/documentation was good enough for me..
I ended up a systems programmer in this doomed industry. BASIC on a Z80 didn't get me there (exploits on a DG/UX at a library system did.) Different people, different vectors.
That was my origin story. I was 8. Tandy CoCo2. I didn’t have a cassette so I had to type all the programs from the books. Again and again.
I hope this makes you feel better: all industries are doomed. But few rise and fall in one lifetime.
Same here but with the first CoCo, TRS-80.
Anyone remember "The Treasure of Raaka-Tu"?
fond memories modifying the variable for velocity in GORILLA.BAS before I really knew what velocity was :)
I don't remember there being any filesystem at all, let alone FAT.
Yes! Paraphrasing my childhood:
“Mom, I’m bored will you buy this a new computer game?” “Lol no! Go outside.” “Well crap. I guess I’ll have to write my own game.”
This was me too, but objectively speaking we were in the minority. Most kids did not learn BASIC.
Nice attempt at slipping one past, but the industry is far from doomed. Where do you work now that you believe that?
Holdouts feel everything is doomed.
Adopters feel the future is brighter than ever.
Article author is excited. I am too.
I exist outside your dichotomy.
I feel like the future is bright precisely because AI sucks so bad while also lowering the bar.
It made people not take software and software engineers for granted. Public perception of software has also finally matured. Boomers and tech CEOs fell into a tarpit that harshly exposes their stupidity while laypeople now feel like they can change the system.
This is all exciting in the best ways. My usage of AI at work and for personal projects goes as far as whatever the AI summary is on google. That's more than enough for me to get enough of a productivity boost on par with those who embarrassingly thrust themselves balls deep and refuse to pull out. Not everyone has to push the "I'm feeling lucky" button to get the same work done at the same pace.
"I exist outside your dichotomy" is one for the ages.
There's nothing to be excited for. Even if AI works as well as people say (and to be clear: it very much doesn't), that future sucks ass. No more fun making things, no fun solving problems, just hand it off to the machine that does things for you. That's an awful future where every shred of joy is gone from computing.
It might genuinely suck for you. It's amazing for people who couldn't build anything before, but now can. And there's more of them than you.
I think that it should not need to be mutually exclusive. You can make it you can enter the programs yourself (if you prefer to do for whatever reason, including, it does not work well (like you mention), or because of Dijkstra objection or other objections to the use of AI for programming the computer or AI in general), but that AI can also be possible.
You're welcome to not make things.
Are you telling us the only reason why making things was fun is because it made you "special"? And now that anyone can do it it isn't worth it for you anymore?
That's the wrong reason to do something.
For me, I made things because I liked the process of making things, it was never really about the result. Much like solving crossword puzzles is about the process itself, not about getting the answer.
Then nothing has changed. You're free to continue doing things as you like.
Unless you're complaining you can't be paid to do it the old way. But none of us were paid to have fun, we were paid to deliver results.
You are completely right. I am not entitled to anything. But before I was paid to have fun, I spent years trying to get better at it and then suddenly it completely lost its value. From fun of solving crossword puzzles to the dread of validating if crossword puzzles bot solved crossword puzzle correctly, with worse pay, less job security or opportunities.
I am not entitled for things to continue as they were, but I can still feel sad about it, you know.
I have never been able to understand this perspective. It's come up so often that it almost feels like a pro-AI propaganda strawman in the style of the "virgin vs chad" meme.
You can and should still write most of the code yourself and use AI to assist when you get stuck. Apart from saving yourself some hours because you no longer hit dead ends on google, I don't understand how that's meaningfully different from using a search engine and finding solutions on StackOverflow and GitHub. I don't even understand how people think this traditional workflow I'm suggesting is any slower than reviewing LLM output either. It takes the same cognitive load and the same time in the end. The only people who believe they can go faster and reduce cognitive load aren't just inexperienced, but also seem to have ADHD or something. Like they just wanna bask in the vibes of the machine-generated text scrolling by.
If your goal is to make something you're going to live with and maintain long term, or is more generally the foundation of a bigger idea, you can't ignore any part of the code or delegate it away. It's still writing that you're bound by like any other document.
lucky people, as soon as I learn how something works it's pretty much impossible for me to keep any interest in it
> lucky people, as soon as I learn how something works it's pretty much impossible for me to keep any interest in it
I guess that means you are still interested in pretty much everything
Yes! I loved BASIC! It was what got me hooked on computers.
Naw, you had the knack.
"doomed industry?"
Please.
I don't know what's wrong with my perspective. I cannot relate to this at all.
I am at a job where I am paid well to build software to do things in a distributed system that processes information across various external systems (some with physical world impacts). The outcome of that computing is revenue for the company. Using agent-wrapped LLMs makes the software authorship component of the job really fast. The rest is virtually the same speed.
For Thomas's mobile computing examples, these are done really well by my phone's operating system's built-in voice command system, mature for a decade by now.
At home, I use software that is well-crafted to do some thing. Or I write a series of scripts to do narrow things that for some reason aren't in Aptitude or Snap, for example "copy all the photos off my iPhone and rename them and sort them by date in my filesystem".
Perhaps I have no imagination, but I don't feel like having an LLM make applications on my phone. I want good software that someone that I trust wrote.
Thomas has been a successful cryptographer, vulnerability analyst, and entrepreneur, so he probably knows what he's doing. I just don't get it.
> At home, I use software that is well-crafted to do some thing. Or I write a series of scripts to do narrow things that for some reason aren't in Aptitude or Snap, for example "copy all the photos off my iPhone and rename them and sort them by date in my filesystem".
> Perhaps I have no imagination, but I don't feel like having an LLM make applications on my phone. I want good software that someone that I trust wrote.
I do not use LLM either, and do not intend to do. I can write my own programs. Even if someone does want to use LLM, it should neither be required nor expected nor the default (and there are also reasons why I think that it should be discouraged to use LLM too much); you should be able to write the scripts and programs by yourself if you want to do it that way (and the computer made to make this possible to do without LLM), even if some people can make LLM to do it at their option, but it should not to require LLM and AI.
I also think that the computer that can start with a programming language such as BASIC or Forth (without first needing menus and GUI to access all of the programs) is useful, and that it is not as good that many modern computers do not do like that.
In my opinion, this post misses the forest for the trees. Yes, the model of an OS running various off-the-shelf apps for various tasks will soon become obsolete. But it's not the idea of the OS that's outdated - it's the idea of apps. Why would you ever want to ask an AI to make an app for you to complete some task, if you can instead ask the AI to complete the task directly? The future isn't a bunch of extremely personalized apps - the future is a single AI assistant doing all the work.
I've seen a few posts of the form "why would you ever want to use anything else than AI from now on?" by now. My counterpoints would be:
Unless you have a local model and the appropriate hardware for it, your agent is Somebody Else's Computer. Do you really want to send all your data and make your entire computing experience dependent on whatever OpenAI or Anthropic or whoever else is planning in this moment?
That's not even starting with inference time and token cost. Despite all the incredible advances in inference, it still takes more time than most non-AI computer functionality. Do you really want to wait a few minutes and pay money for something that you could also do with a few clicks fully locally on your PC?
But the most important thing: User interfaces. Right now, we're basically cramming everything you could possibly want to do at a computer into a chat interface. But there are lots of applications fir which specialized graphical interfaces are much more suitable. Why would you want to get rid of them?
Somewhat connected to that: Repetition. If you have to do the same task again every week or every day, it seems wasteful to ask the AI for it every time: Not just are you wasting a lot of time, energy and tokens, you're also at risk of getting inconsistent results, if the agent from today's session will interpret the requirements slightly differently than the agent from yesterday.
You can circumvent all those things by having the AI write you a custom app, but then using the app without AI to do the task.
Those are all techbical limitations that will be solved with time. In 5 years I would be surprised if we do NOT have an AI as smart as today's frontier models, that runs completely locally, and has a real time voice interface.
I'm not a Treky but I think Star Trek has the right interface here. For basic operations you just use voice and tell it your intent or what you need to know "Computer, what's the fastest route for me to get home?". "Madison is fastest right now, it's 3 minutes faster than Lake, it traffic is building up".
This isn't just whatever Waze gives you, it knows about your constraints and preferences. So you need to program that somehow. Imagine taking it one step further "Computer, what's the fastest route to pickup my daughter and get home?" or even it just responding to the first prompt "Take Roosevelt, and today you need to pickup your daughter from volleyball".
In Star Trek they often do this when they create "simulations". They tell the computer constraints and then it refines it. No writing programs, but effectively that's what you are doing but with the natural language voice interface.
"Computer let's create a simulation to figure out the fastest way for me to get home. I want to take a route along the expressway and know whether to get off at Roosevelt, Madison, or Lake. I need to pickup my daughter when she has volleyball." You don't need to repeat the whole instruction each time, and behind the scenes it is probably turning it into some deterministic code, so next time you trigger that program it's not going to hallucinate some nonsense.
Why don't you just use the transporter instead of your car?
And yet, even on Star Trek we can see people typing things into screens or onto pads. The helmsman doesn't fly the ship by talking to the computer.
My point is that text-based interfaces - and voice is just a convenience around text - are not the best interfaces for a lot of tasks.
We live in a "capitalist" society. Moats will be built, regulations will be enacted to ensure the American Corporation controls the market segment. No one has been able to unseat Amazon, no one has been able to unseat Google, no one has been able to unseat Microsoft, no one has been able to seat Twitter, Facebook, etc. The modern market is segments that are controlled by a single entity. There will be no local models.
> "Computer let's create a simulation to figure out the fastest way for me to get home. I want to take a route along the expressway and know whether to get off at Roosevelt, Madison, or Lake. I need to pickup my daughter when she has volleyball."
Imagine saying that to a random dude with a map and that should give you a good idea on how many info you're leaving out. You can say that the AI will infere those but that seems to be the common fallacy of AI enjoyers: They're always assuming the AI will magically find out the missing information from their prompt somehow.
I have Organic Maps installed on my phone and within a few minutes (faster than my car warming up) I can set a multi stop route (you can bookmark places). Training a random user to use such apps is also equally fast. The same happens with various pro tools: once trained, a user can be very fast with them.
> I have Organic Maps installed on my phone and within a few minutes (faster than my car warming up) I can set a multi stop route (you can bookmark places).
Wow that's slow. I can do the same in Google Maps in seconds. Can you tell me what I'm missing? I'm definitely missing something.
If your car is relatively new (couple decades) you don't need to warm it up, just drive easy for a couple minutes.
The important part is that you're not prevented from driving when it's still cold.
What engineering innovation in the recent decades has enabled oil-lubricated internal combustion engines to not require warming up?
Much tighter electronic control of fuel, ignition and even cooling system combined with tighter tolerances and modern oil formulations. Wait for the cat to light off and the revs drop and then put it in gear and take it easy for the next couple of minutes of driving.
Planned obsolescence.
Getting rid of the internal combustion engine is the main one, but your point stands.
Empirically, nobody has a problem with their stuff running on someone else's computer, with chat interfaces, or with repetition, or with repetition.
Loss of control/privacy, poor performance at high cost, bad UX.
Fortunately our industry has been getting people used to these for years.
Are you criticizing vibe coding or assistants here? I really can't tell.
We are talking about a hypothetical future where coding AIs are good enough, cheap enough and ubiquitous enough that not just expert programmers, but everybody is capable of vibe coding exactly what they need, to the point they actually reach for coding AIs and not the app store with ready-made apps. That future is still far away, at least 5 years, probably closer to 10-15. In that hypothetical future, most of the problems you're talking about would be solved, or else we wouldn't have that future yet.
Man, you seem like you're thinking linearly... the capability growth does not look to be linear... 3 years tops IMO
Actually I'm thinking sublinearly. Ceiling is coming.
You really believe that?
Sometimes it is wild to me how two people can look at the same information and come to completely different conclusions.
As best as I can tell, the only constraint the really matters right now is availability of compute, or as a buddy of mine put it "GPU GO BRRRRRRRRRRR!"
> Unless you have a local model and the appropriate hardware for it, your agent is Somebody Else's Computer
So, any Apple computer?
I imagine the future of computer interaction could evolve into a split model combining foundational capabilities that just make sense to be pre-built, tested and shared deterministic code - codecs, numerical solvers, geometry kernels, databases and transactional tools, compilers, theorem provers, cryptography, etc. - with a fluid interaction layer that accesses either local or cloud-provided versions of those foundational capabilities and is capable of synthesising any tools it needs where it would be helpful to build it first - either as a one-off or something to cache - for anything that can’t just be done as a task.
Applications are compressed bundles of some assumptions about how something should work and be interacted with. So you wouldn’t need VLC, it’s a shell around a foundational codec layer. Instagram is just a view over data in the Instagram social network, and you could customise it as you see fit. Transmit is a lovely application but you can delegate that to a fluid task layer that can do one-off tasks or make something you can re-use.
Everything in the fat ‘middle layer’ of software that is connecting a user to something foundational through interface assumptions might disappear, unless it’s more useful for it to exist as a permanent tool. Things that a lot of people share or that are more complex, Blender for example.
I also imagine there’d be a lot more focus on making explicit things like canonical records or schema, contracts, provenance, versioning, access policies, some kind of fixed views. Things that your local fluid interface would rely on the structure for when it came to connecting to your bank and doing whatever you want with your finances.
Not all software as we think of it might disappear, but the things that are lightweight layers over something more stable could all be subsumed with tools synthesised on demand.
Perhaps, at least.
> if you can instead ask the AI to complete the task directly?
Going to entirely depend on the nature of the task and where AI pricing goes in the future.
Would you like to spend $0.10 on every task or would you like to spend $1.00 to make an app that does it free for you forever?
I don't think it'll be one or the other.
But also I don't see things like video games ever being replaced by 1 off AI. If for no other reason then the fact that a lot of games are fun because of multiplayer capabilities. That's not something you can do a 1 and done with AI (unless you only want to play games with family and close friends).
If it costs $0.10 to do the task once, then you can be sure as hell it costs orders of magnitude more than $1 to vibe code a whole damn app that does that thing, especially if you want good user experience which is the whole point of custom vibe coded apps. Whereas the App Store will have an inferior, mass-market version of what you want actually priced $1.
That's not necessarily true. Vibecoding a simple JSON -> YAML converter to translate a 2MB JSON is going to be much cheaper than including that JSON in a prompt and burning hundreds of thousands of tokens. Besides, there's lots of cases where the math simply favors a bigger one-time cost rather than paying fees in perpetuity.
Even today, an advanced AI model will realize exactly what you said, and generate a script and run it behind your back to save the costs and ensure correctness. To me, that's still using the agent directly, not vibe coding. No app gets developed in the process, and if the agent doesn't tell you, you won't even know any coding took place.
With the caveat that literal today's models are not allowed to run code themselves but instead give you the code in the chat window and ask you to run it yourself for security and safety reasons. But the capability is there.
Right, which makes the one-shot task cost the same as the permanent solution in that case. So if you wanted to do this task even twice, you would still be better off creating the tool once and just using it. Imagine if proof assistants like Lean didn't exist. Research with AI would get a whole lot more expensive if it had to re-create Lean every single time you asked it to prove something.
If you want to do literally the exact same task in literally the exact same configuration and ideally the exact same input data, then yes, it costs the same. But that solution is as permanent as it is useless.
A tool for generating a report for September 2026 is no more expensive than an agent generating a report for September 2026. But that tool is useless for you in October 2026. You'll need to generate a whole new tool for October 2026, and generating 2 tools costs as much as generating 2 reports so you're not saving anything.
Making a tool that can generate a report for any single month can easily cost 10 times more than a single report, entirely due to the extra code flow analysis required to extract the right variables in the right places. And that's without any UI. If you want UI that's another several times more tokens because UI is hard. Also it will suck compared to a simple chat window where you can ask the assistant for anything and it will understand. If you want UI that's better than the chat, that alone costs you more than everything you spent so far.
All in all, your permanent solution starts paying for itself in about 4 to 6 years, assuming you never had to update how the report is generated in the meantime.
Or you could have a very simple, very cheap script that only gets run once then gets deleted (or even better, kept for future reference to make future report scripts even cheaper to generate).
Note that this is all in the hypothetical future where AI is actually capable of one shotting such a task. If it makes mistakes, then human-made apps are still the most cost-effective option and this whole discussion is moot.
What if all computer games from now on are AI Dungeon?
An AI assistant can give different answers for the same inputs. An application written by an AI assistant, if it doesn't just call an LLM at runtime, will mostly likely produce the same answer for the same inputs. That is, unless there's something intrinsic to its business logic that makes it nondeterministic.
Sometimes you really want the latter even if the former is less effort.
And when applicable, the program would be much more efficient than the LLM.
So it would really be beneficial for the LLMs to be able to have their own environments to write and run code, but also to store those programs for later use automatically.
A *human* assistant can give different answers for the same inputs. At some point, we'll need to stop treating computers as instruction executors and start treating them as autonomous agents. Non-determinism isn't a bug to be fixed, it's just the reality of working with this new kind of machines, and operators of those machines need to embrace that fact, like they did for human assistants.
And we started adding rules for the humans to follow, until we have complex bureaucracy. We invented tons of formal notations systems and error detecting and correction mechanisms. We invented deterministic machines. All because we are deeply unsatisfied with the non-determinism! Because it means you can never trust some process, optimize it away and stop thinking about it. You always need to check, and there is always something missing.
A lot of our manufacturing ability improvement during the last 200 years, as well as our engineering ability improvement during the last 500 years, is not just the raw power we unlocked, but that we made it deterministic, reliable and reproducible and hence automatable and automatic.
But sometimes you do just want the same answer each time. Not this human woo crap, where the answer depends on god knows what, but a deterministic function of some kind, reliably producing a given output for a given input according to a fixed set of rules.
Computers can do that!
Yes, but sometimes (quite often actually) you really don't care. The entire subfield of audiovisual compression is built around the principle that you don't care. The data is actively being lost and the image distorted, and you're fine with it. Wanting deterministic results out of MP3 is like wanting AC units to be loud. Sure, they are loud, just like MP3 is deterministic, but why exactly is being loud one of your requirements when choosing an AC unit? Isn't it enough to produce cool air, or in case of MP3 - sound good enough?
Accuracy, precision and determinism are completely different things. Lossless compression reduces precision, and by that also accuracy, but it doesn't become undeterministic.
It would be like telling the computer to compress a music file containing Bach's Toccata and Fugue in D Minor, and getting back a file that contains Rick Astley's Never Gonna Give You Up. Maybe you don't care, but I do.
You are talking about AI outright failing to get anything remotely close to the correct result. That's a completely different argument and has nothing to do with non-determinism. Things can be deterministically wrong too.
I want an agent studio and execution sandbox where my LLM driven assistant builds and maintains my apps with a receipt or transcript like history of what it did to create each app. Determinism built per use case with non deterministic assistance orchestrating the build and management layer. Extra credit if I can trivially share and collaborate with others per “deterministic silo” via a shared link. Something like Claude Code + AWS Lambda|(Docker|Podman) + Tangled.org with LLM code generation for my phone and workstation.
The LLMs empower for building, the determinism improves output-expectation alignment. It’s fundamentally an on device software factory, hypervisor, and storage system.
> my LLM driven assistant builds and maintains my apps
Remember when everyone was talking about treating computers as cattle rather than pets?
This seems like a weird inversion. Programs as pets instead of cattle. Toxoplasmosis, ahoy...
Cattle can do complex tasks too. Including managing other cattle.
Think of all those bad sci-fi movies where aliens take over Earth, enslave all humans and treat them like cattle. Very competent cattle that's put in charge of local governments, scientific research etc. but very much still cattle where individuals are entirely interchangeable and killing them is just everyday work. This is me and AI assistants. (I'm the aliens.)
If the code works good enough, it works good enough. If good enough is insufficient for success criteria, you’ll have to trade app evolution velocity and flexibility for more determinism via more human effort (“artisanal hand crafted”), evaluated on a per use case basis. “Does this software have to be good? How good? Have I hit diminishing returns yet?”
Pros and cons to code being cheap to generate now, it’s always about tradeoffs. I don’t want to build cars, I want to drive cars.
Only in the sense that your build-a-car will be "good enough" as long as you never find out why your palms are always itchy, and you never press the brake so hard it stops working, and nobody determines which of the cars started the fireball during the accident.
People suck at "good enough". All major engineering disasters were something "good enough" for somebody. People burn down their houses with "good enough" electrical work.
> Yes, the model of an OS running various off-the-shelf apps for various tasks will soon become obsolete.
I doubt it. Most people would rather stab at their phone to download an app that just does what they want than explain to a chatbot over a period of minutes/hours and several reiterations to get it to spew out what they want, let alone what kind of software they want it to make for them. Most people can barely articulate what they want in the first place.
The market for applications isn't going anywhere and neither is the OS. I can't imagine booting a computer directly into a chat window where I have to explain to a bot what I wish would show up on my screen, where on my screen, and how it should look. What a total nightmare that would be!
Even if what passes for "AI" these days was as smart and capable as the computers on Star Trek (and it very much isn't) you wouldn't want to ask it for everything. Even the Enterprise had an OS that people interacted with. What they also had though was an OS with configurable interfaces that people could customize according to their current task and preferred workflow.
That sort of thing is maybe possible in the future for "AI" empowered operating systems we might live to see, but 90% of the people using computers aren't going to bother with any of that, just like they've never seen the settings/options in most of their applications or their operating system right now. Not because it couldn't make their time working the computer better, but because they just can't be bothered.
The writer of this piece remembers how, as a child, computers promised to let him write his own software and games and it seemed like there were endless possibilities until the reality that programing takes effort killed a little of that enthusiasm (although not enough that he didn't eventually become a programmer), but many people never see a computer as anything more than a means to an end or a device for mainly passive content consumption.
The cognitive dissonance is astounding in AI circles. The vast majority of people have a personal computer in their pocket and they use it the vast majority of the time to scroll Instagram, play games, and occasionally interact with their bank and other social services. People who have laptops and desktop computers have them for specialized tasks they already do. None of the vast majority of users who buy their phone when the screen cracks and the battery stops working care about the agentic AI world because their phone already does everything they want to do.
I think the rise of Claw's, Grok Bot and Muse is a good counterpoint to this. Yes they are still niche specialized interfaces right now, but they are basically allowing non-technical people to do things that have been easily possible with scripts and cronjobs for a long time. Its basically the Ford and faster houses analogy.
Howl was faster houses, Ford was faster horses.
When I look around me, the only time people get their phone out of their pockets is either for youtube, some social media, or calling/chatting with people. It's very rare to even see them set reminders or use the calculator. It's look like you're in some tech bubble. People outside of that don't scripts or do repetitive and formalized stuff. The people that do create apps and services to sell.
My wife is as non-technical as it gets, but she still spends hours with ChatGPT everyday. Calorie counting, parenting tips, gardening help, fact checks, looking up celebrities, road trip planning, meal planning, health related questions, all with ChatGPT. I really don't think she's the only one. It wasn't the most downloaded iPhone app of 2025 for nothing.
Sounds like it's replaced Google Search for her, not much else.
Okay, I went too far with that one sentence. Can we all pretend I have instead said "this model will be obsolete for most, not all, but most productivity apps and most, not all, but most office tools"?
It is not clear whether models will be both capable enough and cost effective enough but not as capable to wipe us all out. The outcome is a tiny sliver of possible futures and even then likely short-lived.
Current capabilities are not there. You need software engineers to keep it going. And even then as intoxicating as it might be to be able to bring to life all the side projects you often had no energy/skill to finish, today it is still may be cost / “cognitive load” -effective to buy off the shelf apps that somebody else has to maintain (even if they are not as personal / less customized for your use-case).
“Go enjoy my family photos for me.”
“Go play video games for me”
“Go create music for me”
Idk. Seems like there is still use for apps.
Exactly my opinion too.
Apps will become barely visible data providers. Some will provide specialized widgets for those cases where a specialized UI is needed. And data will be designed to inter operate.
I have reached this conclusion from a specific example I'm working on, more details in this blog post if you pardon a link to a website of mine: https://cloudywithachanceoflatency.net/blog_02-how-llm-accel...
Why would you ever want a nice word processor to write a letter to someone when you could just ask AI to write the letter? Why would you ever want photo editing software when you could just tell the AI to make your picture look good? Why would you want audio recording and mixing software when you could just tell the AI to output the next Billboard 100 hit? Why would you want a camera when you could just ask AI to generate photos for you?
Like browsing TikTok or messaging your friends? People want their devices to work like their friends’ devices do. Your prediction seems to be out of touch with how and why people actually use computing devices. Nobody actually uses them to "perform tasks".
Maybe we need an app store just for agents to make purchases. Imagine a world where startups have their valuations determined by how many agents are using their apps. Those startups... Also operated by agents. We don't even need users anymore!!!
This is the part that many people are missing. Decades ago I heard "the ideal UI is a search box," and I agree today even more.
I don't understand why people are still thinking that clicking on stuff will still be a desirable thing in the future.
Inputs for the ui might be me talking, or typing, or having someone else talk or type. They might also be GPS, temperature, BLE devices nearby. Or maybe a button I smash because I don't feel like re explaining something.
What happens after an input is still the best part of a UI. We like seeing/hearing/feeling things in different ways.
The fun part about AI assistant is that you can straight up tell it how you want data presented, and it will do its best to present it that way (within limitations of its capabilities and available tools).
What you’re describing is basically a terminal. We’ve had those for a while.
I agree, which is why I don't understand why these are treated like magic tools by the laymen.
It’s funny you say this because Apple resisted an App Store for apps. They originally expected web based apps. The market demanded an App Store. Remember web snippets where one could clip a portal to view a small section of a website?
The current market very clearly demands AI assistants that can do everything. See ChatGPT app, or the new Muse app.
There is a big misconception in your answer in that you seem to believe that an LLM would always produce (machine) code that does what the user intended, in a correct and safe manner. Neither of these assumptions is true. If you knew how LLMs are built and operate, you would know they are not reliable at all. What you might ask from an AI interpreter OS might be unique and thus might be absent from its training set and might not follow a pattern inferred from its training set either.
LLMs are the first machine learning models that blatantly and regularly produce incorrect output and we have been brainwashed into accepting that. An application on the other hand can be exhaustively verified. There is no comparison.
First and foremost, I'm comparing specifically just having a suite of heavily personalized vibecoded apps like in OP blog post, versus using an agent directly. I'm comparing just these two options and nothing else.
If we don't assume up front that AIs are good enough for at least one of those things, then there's no conversation to be had. Personally I'd rather have a conversation than not have a conversation, but you do you.
Had this thought this morning after seeing DHH announcing that Basecamp isn't writing code by hand anymore and they're using AI to write Rust for Hey's backend.
To me they're failing to see the even bigger picture that maybe even Hey itself is outdated in the Ai era and not only is its Ruby framework antiquated... the whole Hey service is too.
> The future isn't a bunch of extremely personalized apps - the future is a single AI assistant doing all the work.
I suppose if you break your glasses, squirt real hard, and drop down about 30 IQ points, then sure.
We're extremely far from anyone, even those with the most interest and capability, giving a practical demo of this future you're suggesting.
What we all see irrefutably happening is some new people are just barely able to do slightly more than they used to. That's the impact we'd expect from a better search engine and a better "I'm feeling lucky" button which, of course, is what this era of AI is.
> We're extremely far from anyone, even those with the most interest and capability, giving a practical demo of this future you're suggesting.
Yes, absolutely. This is a very far future. But that future is (maybe, possibly, hopefully) coming, unlike this other future which is absolutely never coming.
Why "hopefully"?
Having AI "do the task directly" sounds just as miserable as anything else people keep insisting is the future. We can't rely on probabilistic methods. We will still need humans in the loop to pin down the deterministic parts and also the parts we intended (often the same, but not always).
What else did you have in mind?
Humanity has dreamed of being able to control computers by having actual conversations for the last 80 years or so based on nearly all sci-fi movies ever released, and that dream is finally becoming reality. And it's not a coincidence - talking is the most natural, the most human way to interact with a computer, and the only reason we haven't done that successfully before was technical limitations. By which I mean, if not for technical limitations, we'd have been talking to our computers a long freaking time ago. Wait, how old is Siri now?
If AI keeps getting better, then it is inevitable that voice conversations will become the dominant way in which mostbpeople use most computers for most tasks most of the time. It's just part of human nature and there's no way around it. The only way this doesn't happen is if AI never becomes good enough to support it.
Personally, I'm not a fan of voice input. But I want AI to get better. And the future where voice conversations work well enough to be used as the main HCI method is inevitably also the future where voice conversations are used as the main HCI method by most people. I'm hopeful for technological progress, that's all. It's just that you cannot have this progress but not have people take advantage of it in this exact way.
Compare it to the future where most people vibe code their own apps, which has zero chance of happening no matter how good AI gets.
> the model of an OS running various off-the-shelf apps for various tasks will soon become obsolete
"ChatGPT, please send <username> a message with the following contents: <content>. Oh, what platform? Create the platform to send the message for me. That username doesn't mean anything on this brand new platform? Uhhhh" [insert step where the AI-psychotic decides to give up on communicating with humans and tells ChatGPT to roleplay as the person they intended to message, I guess]
I hate that web platforms and other low level cloud infrastructure are called "apps" these days. Makes it impossible to talk about actual end user apps, vs. all the other code that's also running in addition to actual end user apps, which will remain in place even if actual end user apps become obsolete.
In other news, switching from gas guzzlers to electric cars had the unfortunate side effect of highways disappearing.
> the future is a single AI assistant doing all the work.
Sounds more realistic than what this blog post is proposing. Meta also tried releasing a phone, how did that go?
Muse on the other hand is the polished version of what OpenClaw should have been and Meta appears to have given a lot of thought about it more than this blog post.
I don't think we need yet another phone.
The author is missing fundamentals. Exactly what to expect from someone who thought the built is BASIC was lame.
He talks about applications he generated without specifying what they are or do. Does he really believe everybody is going to generate their own text editor, paint program, file explorer or graphical windowing environment with A.I. just because they can?
I believe the author is misguided about A.I.'s capabilities just like many analysts and investors are.
A.I. (or should I say LLMs) is useful for specific problems but contrary to its name it has no "intelligence."
Out of curiosity, what definition of "intelligence" are you using when you say that an LLM doesn't have it?
True reasoning capabilities and continuous learning.
And self-awareness for bonus points.
And then there are products that call themselves an "OS", but which they have nothing to do with computer operating systems...
> That makes a lot of sense in a world where we’re importing all our software from strangers. It makes less sense in the world we’re heading to, where most of the software we’re carving up fiefdoms for has the same provenance.
If I'm running software written by an LLM, even if I was the one who prompted the LLM, I would still want my OS to treat it as if written by a stranger.
I know he's not necessarily talking about relaxing security models here, but he's not being very specific about his vision otherwise.
I think OP may be onto something real, while you are definitely right if we translate concepts directly. But maybe the challenge, with current platforms included, is because we try to cram identities by reference. If a phone had its own identity, with which you transact as with any party then lots of the security concerns would dissolve. To tone down crankiness, I refer to identity in the technical sense, not personality etc. This doesn’t solve the challenge but it might point to a different foundation for a security model.
If the phone injures or defrauds you, does it have to make restitution? If so, how?
'Transacting' requires not only identity but accountability, a completely controlled-by-you phone cannot meaningfully transact separately from you.
Ownership does not imply co-identity. If my phone harms someone else, without me being responsible but e.g. the brand that updated its firmware? Your questions are good, and I don’t have the answers. I just think our current model doesn’t either.
> If a phone had its own identity, with which you transact as with any party then lots of the security concerns would dissolve.
I feel like a lot of new security concerns would be created.
Indeed. Some would go, some would come.
I mean, everything should eh sandboxed imo. Even OS included software.
Problem with sandboxes is that it doesn't solve the "user who doesn't know or care" problem.
If you have to give access to particular files to an app, your average user will just give access to all. Someone installing tiktok doesn't look at the permissions...
The best thing would be for apps to operate on their own copy of a file regardless. But that doesn't stop bad apps from leaking importsnt files. Unless files can be signed as being for x thing requiring y special permissions as granted by the originator and not the user.
The crowd on here we can't to be able to truly own our devices. But I think for the safety of the average person they should be locked down by default, same as dev mode on android devices where you accept liability for doing silly things.
Nope. There has never been a "good" post written in this genre ("I'm leaving my last team, here's my new project") and I'm not going to waste interesting bits on my pro-forma transition post.
I didn't submit this! Happy to chat about it, but I feel like I was pretty clear in the post that I wasn't bidding for the front page.
I think you got your comment replies mixed up? I don't see the relevance of this (https://news.ycombinator.com/item?id=49851040) to that (https://news.ycombinator.com/item?id=49850998).
Sorry, I'm responding to the (apt) complaint that there aren't a lot of details in the post. They're right, and I'm just adding why.
I doubt that we are entering a world where people will be using AI to write their own little apps, and that is coming from someone who does it.
All you need to do is look at history to realise that. We had things like BASIC, HyperCard, and the early web that attracted droves of people who were excited by the opportunities. While a lot of those people went into industry to write software for other people, a lot did not. Most people, I suspect, didn't even explore the potential because they simply don't think of computers as programmable machines -- even if they were aware that there are people who program computers for a living.
I think I agree for somewhat different reasons.
The barrier to entry is lower now because of AI. Even though I code, I ask the AI to write a lot of little helper apps for my wife who homeschools our kids, and I don't code review many of them. I only do enough user testing to ensure they function as requested.
Armed with this knowledge, I have tried in vain to get various family and friends who are not technically minded to do anything like this. In some cases, they just tell me that they can't think of anything the AI can help with.
Many things that seem simple to them are more complex, and they can't tell the difference between simple apps that the AI can knock out in a few minutes and larger projects that require internet hosting and/or vastly more AI credits than they would ever be willing to pay for.
By far the biggest problem is that many of them are not very good at describing what they want. For simple apps, you can go back and forth with the AI to ask for changes, but the more complex the solution, the more its becomes like a traditional engineering design project where carefully though out design of features and their interactions are critical for success. The new "code" may be more like human language, but the user facing details must still be specified.
The thing is the LLM interfaces encourage people to write their own apps.
Claude especially, where you might say “how can I do task xyz more quickly” and Claude responds with a little JS app. This creates an expectation that it’s possible to do anything if you ask Claude right.
I have seen this first hand where my partner - a deeply non technical person - came to ask what an API token is and how she can get one for Canva. Claude was busy automating her Canva -> Pinterest workflow.
She now has a half dozen of these things and hunts through her Claude chats to reuse them.
> The thing is the LLM interfaces encourage people to write their own apps.
LLMs are simply the latest incarnation of VB6 + Clippy.
I disagree. Basic etc still required effort.
In my opinion, an OS that hosts a local agent can produce things for you on the fly in response to a particular task in addition to the baseline set of features.
Too often have I ran into "you can't do that" in operating systems, especially Macos where you either have to buy a paid app to implement what they should have already, or what you want is niche enough that you have to build it yourself.
Plus and agent can operate across your hardware including the cloud. I could ask something like "graphics cards are expensive and always out of stock so when some pop up can you notify me and show me the price etc" and then receive a rich alert on whatever device I happen to be using at the time it happens, once agents are aware of where we are and what we're doing (yes, it will be pretty dystopian with current governments and the drooling animals who vote for tribalism).
Or being able to be like "I want my photo gallery to show animated thumbnails when I've taken a series of photos of the same subject. Oh, little bit slower like one every second thanks" boom done.
And much UI will becomes defunct in the first place, UI only exists because of many annoying discrete steps we have to do to get something done. If I can just ask an agent to do the thing I want then I don't care about that.
It's like how I only need to make a PowerPoint now and then and every time I open it I've completely forgotten where all of the stupid fucking buttons are in the stupid fucking ribbon bullshit they made.
I don’t think the post has a lot to offer, but I will share thoughts on how a modern operating system can add _a lot of value_ both for non-agentic users and for agents.
Looking towards macOS and iOS for example, an underrated feature for the “it just runs your web browser” crowd is all of the system-provided databases. These turn rather specialized systems - photo libraries, music streaming, calendaring, contacts management, and mail into shared databases that link applications together.
On iOS it goes much further - because of HealthKit, my Withings scale, my Garmin (when I had one), my MyNetDiary nutrition tracker, my continuous Glucose Monitor, and my third-party analytics apps share the same access to my data. And it is done securely - I can choose exactly which part each sees. This is a tremendous resource for both developers and LLMs.
My calendar, hosted on a personal server but available through Calendar, is transparently available to my to-do list app - so I can see my appointments, my tasks, and even my workouts (thanks to health kit) in one view.
Weather is a system service on iOS. Calendar is a system service. Contacts is a system service. Health data is a system service. Even music - third-party apps can control and consume your music library on iOS - is a system service (if you use it).
My Mac knows how to develop RAW files out of the box, and it can connect with integrations to get them printed, and use my library metadata to sync my favorites to my phone. I have a third party app that syncs my Apple photos library to my NAS.
A good OS, like iOS or macOS, could be the fabric that holds all your vibe-coded and ephemeral LLM apps together. I see it making a whole new category of things possible.
For the record: I didn't suggest this post did have a lot to offer. I didn't submit or upvote it. I pretty conscientiously kept it simple and abstract, because (as I said in a different thread) this genre of posts ("I just left the company I'm highly associated with, to do this new thing") is deeply cursed.
You still gotta write the post, though, because otherwise what you get is a circulating Twitter post.
I'll write something more fulfilling about this next week, I think.
But so much of macOS is also uninstrumentable except via undocumented APIs that are unreliable at best. Why wouldn't Linux or even FreeBSD be even better? They're not limited to content but even the source that can be updated and rebuilt to do literally anything.
Honestly, as a long time Emacs user, agents are amazing because when all your functionality and data lives in open source code and text files, the world is literally your oyster.
Linux and the BSDs might be more instrumentable, but the rest of the experience is so fragmented in comparison. Regardless of category, there is no single service or data format that everybody has agreed to use and build on top of.
The advantage of the Apple approach is that the user’s data is instantly available across third party apps. For example, instead of needing to build (or import) a whole CalDAV apparatus to be able to build a little calendar-based todo app, you can just use the system calendar API and let the OS do the heavy lifting, making your app much more lightweight and instantly usable with almost no user configuration. This naturally applies for scripts and agent skills, too.
> Why wouldn't Linux or even FreeBSD be even better?
Because Linux doesn’t offer a rich set of system services built around the needs of desktop users and exposed through curated APIs.
Linux can’t be better at providing a shared calendar API or a shared photos library API, or one of dozens of other APIs because Linux just doesn’t have those features.
Some particular DE could offer those features, but then who is going to build the app ecosystem around that? And isn’t it going to suck when you want to make a Linux native photos app but it doesn’t work on Ubuntu because your LLM was trained on KDE?
My photos are not text files.
> Forget the map, just tell me if I should get off the Eisenhower Expressway and take Roosevelt, Madison, or Lake back home?
> Is there a meeting happening anywhere right this minute that I’m supposed to be in?
> You can go on and on like this. These all sound trivial. That’s the point. They’re normal-life problems nobody would have built serious programs around, because the audience sounds too small.
Waze and Calendars. Both have huge audiences.
Waze does not answer my commute question for me. Waze finds what it believes to be the quickest route from west loop to my house, but that route is virtually never reasonable. There's only 4 reasonable routes; I just need to know which one to take.
You're right, "tell me which of these four preconceived routes is fastest" is much nicher than "telly me the fastest route"
It’s not the fastest route, because the algorithm is stupid as a result of trying to be smart.
It sees a small side road and thinks “I will take the user down here because there’s no traffic” But it doesn’t consider that the side road is empty because you’ll have to turn across four lanes of traffic in rush hour. So you get stuck for twenty minutes.
You get that this is a commute, right? Not charting a course through unfamiliar territory. A drive I've done thousands of times. I don't care if Waze thinks I should be on Harrison or Race. Fuck you, Waze. There's 4 reasonable routes; I just want to know which one.
It's an app with an audience of just me. That's the point. You're not supposed to want it. You have your own problems.
Apple literally has automatic commute notifications that tell you the optimal route (15 minutes via Easy St.), in this hypothetical world where your agent knows enough about you to suggest today's route, why doesn't your digital panopticon purveyor (be it Apple, Google, your car manufacturer) automatically pull that info the second you get into your car?
we could spend time answering that question or, as tptacek proposes, diy the solution because the app making app is right there, and answers to your input not Apple/Google's
The idea of software provided by domain experts has to continue even if you can write software to do what you want by chatting to AI.
"Write an app to provide psychological advice to people with mild anxiety, then market it and sell it to people with mild anxiety."
If an entrepreneur did that with no domain expertise, imagine the potential liability.
How do you know the output is correct for anything that matters more than completing simple and repetitive tasks? You still have to know an expert in a domain has vetted the software. The result of applying AI to OS development is making itself fairly clear with MS I believe.
> If an entrepreneur did that with no domain expertise, imagine the potential liability.
It seems to me as if the current entrepreneur playbook largely is to run really fast and hope the judicial system doesn't catch up.
This is the reasonable stance. But both profit and convenience override reason.
> Is there a meeting happening anywhere right this minute that I’m supposed to be in?
Nothing about this is groundbreaking and could have been written in BASIC.
I grew up on BASIC using a BBC Micro in the 80s, using a QWERTY keyboard and a screen. Some decades later I'm using a QWERTY keyboard and a screen with a few more pixels. I'm betting nothing will change much. An OS will do OS things, just a lot faster and more securely.
Spreadsheets will still be useful for making deterministic calculations about finances. It will just no longer take 100 engineers to build one.
GIS software will continue to use deterministic algorithms for calculating areas because it's a useful function but, hopefully it will get more user friendly.
Most people don't have specialised computing / software needs or uses. In fact, they're generally Overwhelmed by the current software landscape. They want less, not more tools.
And to them software is generally a not much more than a home-screen bookmark to some client that could be a website.
> It’s not simply to “provide access to hardware resources”; plenty of embedded systems do that with runtimes that have no business calling themselves an “OS”. No, the core purpose of a modern operating system is to partition different applications off from each other, and carefully control how they can communicate.
First of all the initial definition is totally correct, and if your embedded function needs to call an intermediate layer to access the hardware that layer is an OS. Second of all, all those "purposes" are exactly achieved by providing _managed_ access to hardware resources.
And that is valuable in whatever world we end up living in. Sure, 10 years from now my bank app and my Hacker News browser may both be written by "my LLM" and tailored to my use cases and desires. Will I want them to share memory? I don't think so.
In recent times, I've palpably felt the excitement of being able to imagine like I was a kid again. BUT here's the reality I've simultaneously been hit with:
Most people, and I wouldn't be surprised if it was 90%, who've had any interaction with AI are still stuck on chatting with a chatbot. They've got no clue how far you can now go with a few well written markdown files and/or prompts, all in natural language!
So while I mostly agree on your outlook, and think it's super exciting, I've also felt firsthand that building with AI hasn't yet diffused as much I'd hoped it would. It's still concentrated in the hands of people who have an aptitude for building in some way, shape or form.
The question is: How do you get more people, who aren't power users or builders, to start building with AI? Maybe it starts with getting them to see that building to solve their own life's problems makes life more interesting?
I look forward to following your and Kurt's new adventure. I love Fly in no small part to your writing for its blog. Wild to see what feels like a complete changing of the guard at Fly, at least from the outside. Your post resonates; reminiscent of the Ink&Switch talk from local first conf that I watched today.
I'm tired of reading about the awesome future that could happen, but doesn't. If you think this future can happen, go and make it. Surely it's just a few Claude prompts away? Or surely you already live in this time you describe, so it's not the future, but the present?
Big tech will fight tooth and nail to prevent users from becoming power users.
They don't really have to fight, most people these days accept their role as a guest on their hand-held computer.
The hand-held computer tells them what they're permitted to do, and they conform.
Sometimes, one of the users has a non-conforming idea and says "I'd like to be a superuser on my computer" and they're shouted down with a chorus of "you'll be hacked" and "nobody will be able to trust your computer and you'll be excluded from online services". Sometimes the organisations saying this even use the excuse that "restricting the user is essential to maintaining the user's privacy".
That's the bet.
An OS is what launches the web browser. Duh.
>What I did on my computer, any power user can do on theirs. Will do on theirs. It’s fucking inevitable.
Hubris detected.
In all seriousness though, I'm not so sure.
Power users are lazy users.
And we're not all serial entrepreneurs with 400k karma points on HN, remember.
The OS is the distilled wisdom of other experts. Wisdom that neither I, nor the current generation of next-word-guessers, could possibly keep pace with; no matter how many reddit bots the frontier lab operates in its attempt to extract complex concepts into simpler vectors. It will be the collection of things that are so unbelievably expensive to reinvent and re-learn for each machine and user, that they are not subject to to the market forces that come with making the slop cheaper. Any well-maintained OS will maintain that role. And be ever more important in the future, as apps diverge into even lower quality and even less shared benefit from improving them (for the other approx. 2 users).
The OS is a giant collection of legacy compatibility shims which many people agree isn’t how you would design things if you were starting from a clean sheet of paper today.
If you look at software like sophisticated database kernels they bypass the OS almost completely because they can reimplement a better OS in userspace, using the kernel as little more than a device driver. Over time this has become more common, not less. OS like Linux have actively embraced this role.
Being able to effectively specialize the OS for specific applications has large benefits. It is a specialized skill but not unusually expensive to implement.
I agree with the first part of the thesis (AI means unlimited custom software for power users who are able to describe exactly what they want) and strongly disagree with the second part (this means we don’t need any of the technology we’ve developed for software development, or distribution, resource management, etc.)
I didn't say we don't need operating systems anymore. That's not what I believe.
Adult Roblox
I love the idea that normal people building software to solve real problems in their life is, necessarily, a toy; the only real software is blessed by our sacred priesthood of PLT experts.
> Don't be discouraged if what you produce initially is something other people dismiss as a toy. In fact, that's a good sign. That's probably why everyone else has been overlooking the idea. The first microcomputers were dismissed as toys
https://www.paulgraham.com/organic.html
This is why he has a bajillion dollars and I have more HN karma than him. I'd rather yell about it!
Didn’t mean any judgement just the metaphor that came to mind
Roblox is a game where you make the game. This sounds like an OS where you make the OS.
Probably could have spelled it out more, being too pithy perhaps
No, you're fine, and I might be a little hair trigger; sorry about that.
LMAO that's horrifying.
There won't be an OS anymore. All you need will be a network enabled graphics driver (Hello X11) for the AI to connect to and draw on.
Then the AI will just create the image that resembles a spreadsheet or whatever on the fly and write straight to the framebuffer.
All other tasks that go with the spreadsheet will also be done on the fly.
But I could imagine our language will change and certain words will basically act like macros and create a set of assumptions and specifications for the prompt that you're using.
It's going to come full circle and the model's output layer is going to be the grid of pixels. Nvidia will have returned home.
The same as it has always been, per Computer Science definitions.
Why does this article have this title? I could see nothing in it at all that relates to the question posed by the title.
> Why does this article have this title?
Uncharitably... because eventually they hope to produce a phone that isn't running iOS or Android?
More charitably because the title is actually an important question right now in the abstract for almost everything, not just operating systems.
But the forces that shape the question are so large and hard to get a handle on when we are living through the moment that you have to start thinking about it somewhere, why not OSes?
Sure, but then think about OSes ... not just "software".
I think people will still buy templated software - applications that come with source your AI modifies or plugins your AI writes. Most people need Excel, not a badly written clone, but they may want a bunch of custom plugins that can deeply modify the UI for their use case.
Is this how the internet finally dies? Not with a bang but with the whimper of thousands of AI slop apps scrapping websites into oblivion?
Most of these problems the slop apps can solve are dependent on gathering information to answer questions. If no one ever visits weather.com to get weather information, ad revenue will completely dry up and the site will go under. Then the slop stops slopping and I can’t get my hyper local weather app to work.
I’m not going to go as far as to say this article is wrong, in fact in this cursed timeline it’s probably right. I just don’t see how it’s good.
Ironically this may finally force the fulfillment of the Web 2.0 paid API service dream. Remember when people were pushing companies that just provided a paid API?
There might very well be more of that. In a world where everyone has agents filtering data on their behalf, it seems more complicated to figure out a way to force a user to view certain content along with their desired data than to simply refuse to supply data until a small transaction completes.
On the other hand, that’s just traditional banner/video style advertising where the goal is to force specific content on the user. If someone wants to spread the word about their product or service, I’m sure they’ll pay money to advertising services that use agents to inject their message into the global context. This is happening already, we just talk about it in the frame of “dead internet theory” and AI influencers, etc. Anywhere that content is being produced and consumed is an opportunity for the message to be nudged in favor of the person paying money.
So we go from the days when we have to pay our ISP so that we could go to weather.com to see it if it's raining outside, to a future where we have to pay for our ISP, so that we can connect to the AI company who we also have to pay, so that we can have their AI connect to the API at weather.com which we also have to pay for, so that we can see if it's raining outside? I'm not convinced that'd be making our lives better through technological progress.
Not better, just more profitable (for a small handful)
> Is this how the internet finally dies?
Like many major cultural phenomenon, expect a pendulum effect, more than a beginning or end of a trend.
Surely it's simply software that has its own kernel, that could be booted to from BIOS/uEFI or any other low level hardware initialisation process.
What it has always been. Nothing has changed in this regard.
Oh, ok. I guess computer science is finished, then. I should have known. It's not like there's any evidence that it's been blown wide open recently.
(The irony is that all the reading that's gone into our design is like, old CSAIL papers. Maybe it's true! Maybe everything we're doing is just crate-digging through old settled computer science. That would be on brand. I should find a way to work Athena into it. Maybe we'll use Zephyr for messaging.)
Are you reducing computer science to… just OSs?
I guess it really is over if this is what professionals (supposedly) seem to think.
OP is right though, an OS remains an OS, same as it’s always been. What’s changed is how we interact with it and what we lay on top of it. Nothing magical or philosophical about it.
I didn't say there was no such thing as an OS.
I think operating systems have changed. They don't interact directly with hardware much anymore, and the hardware is filled with SoCs all running systems of their own, sometimes with entire network stacks, communicating with devices and each other, collecting and processing data, often totally invisible to the operating system the user gets to work with. The operating systems most people use have gone from giving users control over their computer to being a sandboxed interface to manage files and some of what their computer is doing.
Could you elaborate on what you mean by "don't interact with hardware much anymore"? If you compare modern drivers with similar ones from 20 years ago, for example, you'll find the modern ones tend to be quite a bit more complex.
No directly anyway, for the OS it's all abstractions and isolation. There was a time when the OS could basically see and control everything, but now there are subsystems all over the place with their own memory, their own software, all outside of the control of the OS. CSME runs while the computer is powered off.
I've heard very competent people claiming similar things. I think there's an abstraction I'm missing. I know they don't mean it literally.
The most recent was Loris Degioanni yesterday in a conference talk.
I think saying " This dumb dumb doesn't know what a device driver is" is a very incorrect read since you know, he's written a bunch of them
I mean it literally.
This is by and for people who do not distinguish between the UI and the OS, like how "MSIE" is "The Internet" for them.
There could be some interesting actual OS design innovations for new permissions, access control, and sandboxing systems that are designed from the ground up for these "conjurable" experiences (or even OS components themself).
Finding some balance between access and safety for vibe-conjuring parts of the OS in a way that's still powerful but won't accidentally nuke or corrupt itself.
The historical baggage is so heavy I don't think anyone can realistically throw it away, so what we'll end up getting are almost certainly incremental improvements to existing foundations.
I do think there's some interesting space to be explored in capability-based systems, but even then I think it would mostly end up looking similar to how existing systems are, just with one more element added
The more AI is wideapread and is used by power users to build apps, the more devices will be locked from running any custom ("non-certified / non-licensed") software. For your safety and security, of course.
If you give an ape a slingshot he might think he can take over the entire world. But he doesn't even know what rubber bands are, or what it takes to make them.
What I did on my computer, any power user can do on theirs.
Back when Google was the newest thing, I would Google all the obscure things that I could think of, trying to fill in gaps in my knowledge. Trying to verify urban myths but usually debunking them, etc.
Any power user or even any user, could have done that. But most people who considered themselves critically thinkers actually did the opposite, digging deeper into all the harmful rabbit holes we see today.
A lot of things can done "with just a few steps" with AI but I think it's underestimated how hard it is to come up with those steps.
Not that the article isn't interesting otherwise.
I kind of agree with the premises of the author.
Software in the future will have to be building blocks for agents to extend. For my projects I decided to generate LLM targeted guides in the docs/ folder, to teach agents quickly "how to" use the project as a framework.
How to implement an MVC view, where to place the files, a guide how to write a good backend route, how to extend the schemas, and other articles like that.
In my opinion we've already reached "Star Trek like computers". Our OS will have to accommodate for better sandboxing and better permission management and better role separation policies, because currently that whole security sector assumes that programs will be static and won't change before/after an approval.
We need to figure out how to better sandbox programs in terms of not only control flow, but also in terms of access to _data_ on a user's behalf.
> What even is an OS now?
It's the thing that runs your AI agent, that runs your AI pocket bot (AIBI looks cute btw), that runs the software doing the training of AI models, that runs the datacenters doing the inference, etc.
There's still data. There are still bits moving. Filesystems and networking aren't going anywhere.
In a charitable reading the title is something like "what is even a desktop OS now?" or "what is even a consumer OS now?".
As for the new project, I've got one bet: there's Linux involved to host it.
Linux is an OS.
P.S: as a sidenote the entire "users don't install apps anymore" was already true to some extent. A shitload of users are totally fine with an OS that runs a browser and that's it. And they're fully operational. And we still all know what an OS is, even for, say, a Chromebook. Heck, I'm not even sure "installing apps" as ever been, except during the first install of the OS, what an OS was ever about.
Not really related except the title: Its time for operating systems to rediscover hardware - USENIX 21: https://youtube.com/watch?v=36myc8wQhLo
This reminds me of WWDC 2010 when Steve Jobs said developers using private APIs 'know exactly what they're doing'. Good luck with your project, but I think you're vastly underestimating the problem. And many OSes are already integrating user generated AI apps.
A solution in search of a problem already solved that no longer exists.
This seems to be only thought out after a day of vibe coding apps that no-one uses.
Taking it at face value, it seems to me this will also collapse and get folded into the model.
Unclear what will be left next to it. It will just live on your phone and take care of everything. The apps will be markdown files describing what they do, probably compiled ahead of time for latency.
But the model will be in control, so what else is left to implement besides giving it bash, a compiler and a display?
I guess there is value in the year before that happens.
I agree with the gist of how weird everything is going to get, and how we are living in a time where people haven't come close to internalizing the shift.
IMO even this AI phone is doomed in the slightly longer run.
One more step into the weirdness and people will just be working with AI to build their own bespoke phone with the specs sent over to a manufacturer that is "like PCBWay, but spits out whole devices".
I have no idea what this does to society, the economy, etc when we reach a point where nobody really "needs" any kind of middleman for almost anything (well, except for things they actually really need, like food and shelter), but it looks like I'll see it happen for a bit before I die.
Hardware is still constrained by compromise. If I want a screen for movies with a battery that lasts and doesn't burn my leg when it's in my pocket, I give up a physical keyboard and nice speakers and a bigass physical camera lens.
The physical device that is best for me is probably the best car compromise for several million other people.
if my ex calls and I want my glass brick to flash the "help me" in SOS on the screen, play Hector Berlioz, and dim all the lights in the room – I can have that. I don't have to want the same thing a million other people do.
A bunch of dot files, obviously.
> Consider: what does computing look like in a world where many (maybe most) applications have an audience of just 1-2 people? How is software distributed? What does it run on?
The same. The slop apps remain just that. They may do something useful for you, but that's it. A bit like a branch in a repository.
Have you thought carefully about this? What does it mean to call a piece of software that solves a real problem you have in your life "slop"?
It's the same thing as looking down your nose at people who use Excel as a database because it's not a "real" database.
Yes, I'm sure we all have horror stories about spreadsheets that got out of hand, but that ignores the thousands (or more) spreadsheets that are silently productive everyday.
That's a pretty good simile in fact. Excel programs are very useful to the people who write them (including myself sometimes); and yet - almost none of them are widely shared, becoming fundamental tools.
Kernel + Shell + Slop
Don't forget systemd :-\
he did mention slop
So - yet another way to turn a computer into a subscription service (to pay for the AI providers). No, thanks.
>yet another startup nerd talking about “the future of programming” before announcing that I’ve joined, like, Sri Lanka’s answer to Uber or something.
not just startup nerd, he also make gas town and drunken rant
steve yegge is not tptacek (thanks for cryptopals, tptacek!)
You've misunderstood; the quoted line from the post is most likely subtweeting Steve Yegge (in particular https://steve-yegge.medium.com/why-i-left-google-to-join-gra...).