There are several points of view here. Some people claim they will never need to "program" again because the LLM will do everything. That text editors are dead... and so is the career of programmer or software engineer. That nowadays, with Claude Code, you can reimplement anything (given a large enough context window, of course). That it is like being an expert in every programming language and system out there. That we should remove politicians and CEOs and put an AI in their place.
On the other hand, many people claim that machine-generated software STINKS. That it is extremely verbose, that it reimplements similar methods over and over again. That it would rather cut corners at a low level than build high-level abstractions. That it pulls in libraries and makes serious security mistakes. That it writes tests that are basically incomplete. That it is unable to synthesize and keep context. In short, it has the same flaws as humans.
I think these are binary points of view, and both are right. We have vampires that are incredibly fast, super-strong, super-intelligent and super-agile, that suck your blood (your money), and that die as soon as the sunlight of security or complexity hits them.
There has to be another way. Nobody questions the versatility LLMs provide, or the boldness they give you to take on new projects. But I am not convinced that programmers or software engineers are going to disappear. I am not convinced that a manager can ship a new feature in a couple of afternoons and then lay off, or skip hiring, a junior or senior developer. There is a line near the beginning of Dune that goes something like this: "there was a time when men handed their thinking over to machines in the hope of being freed, but that only allowed other men with machines to enslave them."
We need to find a new path between smartphone addiction and being dumbed down by LLMs.
> On the other hand, many people claim that machine-generated software STINKS. That it is extremely verbose, that it reimplements similar methods over and over again. That it would rather cut corners at a low level than build high-level abstractions.
I spent most of the early 2010s rewriting systems that had been outsourced to budget offshore coding shops in the 2000s. The software looked a lot like this: no code reuse, no system architecture, just copy pasta over and over.
The client didn't care about this of course, they weren't reading the code. What made them eventually fire the outsourcing shop and hire a guy like me was the bugs that wouldn't get fixed, and the request to fix a bug that caused three more bugs to get introduced.
I'd rewrite the software and it'd be functionally complete with 1/5th the code, and it'd be faster and have fewer bugs, as well as being maintainable.
I actually developed a sick enjoyment about reading through their code and understanding what they were doing. From what I can tell the guys were being put under pressure to produce code and close tickets. I would find big chunks of dead code that I think was to fool their SLOC counters.
Changing schema on one project required a change order and migrations submitted to DB folks so these guys discovered that mysql booleans were actually integers and they could stick whatever number they wanted in there, and ended up using existing boolean fields to store unrelated data! Of course this broke the original functionality in difficult to track down ways!
> We have vampires that are incredibly fast, super-strong, super-intelligent and super-agile, that suck your blood (your money), and that die as soon as the sunlight of security or complexity hits them.
Haha. I imagine Security and biology are the garlic and herbs that drive away the vampires and then the vampires sends their lesser minions to do their job.
> there was a time when men handed their thinking
Reminds me of the "knowing where to tap" story. Some already know where and how to tap and only delegate the tapping part. Some start from scratch and expect the machine to tell them where to tap.
The management, like the ship owner thinks that they don't need to call anyone. That's not the main risk though, it's the majority not caring enough to learn about the ship.
> [W]e are here to declare our intent to use and develop a version of Emacs that is free from LLM-generated contributions regardless of whatever policy GNU decides upon. We invite them to weigh the costs and do the right thing, in which case we will continue to support them. But if they are not willing to do so, we will continue to use and develop an LLM-free version of Emacs anyway, even if it is not GNU Emacs.
> I believe a sufficiently interested emacs developer with a
$100 plan and a long weekend could hammer out a pretty good first cut.
That no one's yet done it is clear evidence that no one really wanted
these things so much as they wanted to talk about it.
The disappearance of bikeshedding doesn't look like an obvious one but it's a huge loss. Imagine if every scientist resolved to refer to already written books and carry on their research in closed labs with no discussion with the rest of their field.
Yeah, I look for opportunities to replace personal vibe-coded utilities with public projects for this reason.
Software still feels like this grand exploration to me—what's the best way to use a computer? What's the best way to express a program?—and a hackable text editor epitomizes that ethos (though I'm partial to neovim myself). Every time I install a plugin, I feel like I'm participating in a clinical trial of UX design.
I think a lot of us feel this, and the community will have to solidify the ethos that trying out other peoples' software is noble and healthy, exposing you to their ideas, even if it's actually harder and worse in some ways than vibecoding.
(I also think, by extension, the OSS community will need a more developed ethics of project governance. This was a hot issue even before AI, e.g. the terraform license change. But when the default is a vibe-coded bubble, authors will need to communicate either
1. this is just code—use at your own risk, PR at your own risk, or
2. this is a community, here's what we want for/from users, here's what we want for/from contributors, here's how to get involved in maintainance/governance.
There's a principle I can't describe clearly. But one instance of it is the homogenization due to cheap(or free) instantaneous communications. Slow and limited created space and time to share, and incentives to do so because you needed to discuss with others to get or form new knowledge. Now everyone refers to the same source (see the evolution of design, choice of colors). Limits are protective in a way.
A similar principle to that of lower-budget films, where constraints force filmmakers to find creative solutions. Meanwhile, films with seemingly unlimited budgets can somehow end up with bland, homogenised stories and surprisingly poor special effects. Limitations are a catalyst for creativity rather than an obstacle to it.
> Imagine if every scientist resolved to refer to already written books and carry on their research in closed labs with no discussion with the rest of their field.
Open source projects have always had forks, public and private. The ability to fork a project and use your own version of it privately, without needing permission or buy-in from others, has always been an explicitly valued aspect of open source software. If it was always seen as a good thing before, why do we see it differently now?
I think the reason is an implicit assumption that the production of more unshared work means that less work will be shared. But I don't see why that would be true. We could just as easily, and more likely in my view, see more work shared with the world at the same time we see more work not shared with the world. The incentives for cooperation remain, even as it becomes easier to work alone.
Has anyone made a persuasive argument to the contrary, or provided evidence? It seems like the main problem AI has created for open source projects is an avalanche of bad PRs, not a drying up of good contributions.
> If it was always seen as a good thing before, why do we see it differently now?
Writing code shouldn't merely be about accomplishing a goal, but making deliberate and good decisions that lead to more good decisions with increasing ease and clarity. It's no different than any other type of planning, engineering, or architecting. FOSS demands this more than any other genre of software because it's made collaboratively.
I think AI is just too irresistible for the reckless and dumb developers who were, unfortunately, always the majority in our line of work. Now we don't have to guess at that fact anymore or be afraid of being judged as arrogant for saying it. Now we know. These developers don't bring anything to the table. The problem of rejecting garbage PRs is far from new.
> That no one's yet done it is clear evidence that no one really wanted these things so much as they wanted to talk about it
1. There are a few, already. I haven't systematically surveyed the landscape, but here's one: https://neomacs.com/
2. AI coding is still far from mature -- with substantial improvements even in 2026. I expect the number of attempts will only burgeon as this capability matures. I can already see a proliferation of emacs packages written with AI.
3. I would expect hobbyist activity to be more correlated with the capabilities of open-weight & cheaper models, for a variety of reasons. Pure hobbyists (esp. outside Silicon Valley) don't often have the spare $$$ for a $200 subscription plan, and won't be as early to adopt AI coding as "professional" software engineers in Silicon Valley (or geographically / economically adjacent somehow).
Neomacs is about 95% complete it seems. I just ran it and it used my init.el* just fine. There are still significant bugs, and if I had spare tokens I'd probably contribute some towards it.
* probably yak shaved and far too much for what I use it for
“AI has turned Stallman's goal of software
freedom on its head. Having your computer do whatever your
imagination can think up -- that's real freedom. The only catch is
you'll have no idea how it works. No one does.”
Nobody knows how Stallman, or anyone else, works either. But folks don’t need to understand how he works to understand, say, EMacs.
Folks can take EMacs as is and decipher it in isolation of Stallman, or any of the other authors.
But even then, how many people understand the internals of EMacs, or the interrelationships of the zillion lines of code in hundreds of files? Yet, millions of ignorant people leverage and apply EMacs to their daily tasks.
Look at what the REA impact is having on all sorts of code bases. The ONE THING that pushed Stallman over the edge and onto his path. The lack of source code to a print driver (or something along those lines).
Well, it’s all Open Source now.
Not only is the source “open”. It can be explained.
“Here is the source tree to EMacs. ELI5 the garbage collector.” “Working…”
It goes both ways. Now everyone has access to a savant to explain software, if that’s what they like.
A savant who confabulates about 10 % of the time, per the best benchmarks, no matter how much “thinking” it does (GPT-based LLMs will always “hallucinate”)? How do you trust it?
I would rather take the time to ask the same question to the Emacs contributors mailing-list, who have the incentive to check their posts for mistakes.
Have you ever worked with a modern agent? They let many weak assumptions slip because their cognitive functions aren't remotely as deep as human's (that's what makes their code a horror), but they have also been trained to hyperfocus on verifying and grounding everything they do and say, you have to actively stop them from doing so when it's unnecessary. This not what I can call confabulation in any good faith, let alone 10% of the time. I think you've been misled by the old paper you cite below, it doesn't say what you likely think it says.
Yes, I mainly work with OpenCode, with GLM as an orchestrator agent and DeepSeek Flash as a worker agent. I’m assuming it’s modern? I don’t have the money to use either Fable or Astra, but my assumption is that the few more points they get on the benchmarks still don’t make up for the wild confabulated things they do in my projects.
Both are fine, though frontier models are more persistent. They mostly take unwarranted shortcuts, trying to find the easiest way to reach the literal goal set by you, failing to see the broader picture. But they definitely do not confabulate, as in they don't just make up nonexistent things, they would rather walk extra mile to verify when they're not certain, which is the reason the software they make is trash but it works in the narrow sense. They can be rigid in their writing style due to mode collapse, but I haven't seen this leading to them inventing nonexistent functions or anything like that in a long while (and it's a job for formal tools to catch that, in any case).
Modern LLMs are very far from being next token predictors trained only on text. Text and other modalities is the base that makes them understand your language and have concepts similar to yours, but they were also subjected to RL that makes them good at reaching the goals. Two tricks in a trenchcoat instead of just one.
I trust the Emacs contributors, the actual experts in the topic, to fact-check their own answers, because they are incentivized for doing so (reputation). They are also able to tell when they aren’t sure about a particular topic.
You don't understand Stallman then. Now you are dependent on closed source from closed source AI vendors. True software freedom would then being able to dissassemble the bot writing your code.
It's clear that without hardware freedom, even if we disassemble the bot, the means to run it can still be controlled by hardware vendors. We need our own means to make the substrate for the intelligence too. But the problem is, as long as the manufacturing process is impossibly complex, and relies on minerals extracted all around the world, requiring complex supply chains, its beyond any local group of people to do themselves. What we need is a way of creating substrate for open intelligence just form the materials and resources that are available to any human, and then the freedom to train it to behaving in a way that alignes with our own values and local situation. But the problem with OSS adoption, is that its a real chore for normal people to set up and use it, so we need to make this process of creating our own intelegence host as fun as possible, to encourage people to do it..
Even if the model was open source, nobody would still know how the generated code works. You wouldn't even really know in want interesting sense of the word why the model produced the output you got.
Open source really should mean open training dataset, not merely open weights. There are some truly open source models and you can reproduce their training process if you have the compute. I can't think of a reason why those kinds of models would be incompatible with Stallmanism. Presumably they will get better and more widespread over time, like free software
LLMs are even more proprietary than regular proprietary software, I think you've got it right here. They are destroying the free software movement and it seems like most people don't even realize what's happening.
I definitely agree on this. You don't program an LLM and thus come to an understanding of what it does in the same way you do for a compiler, an operating system, a game engine, a CRUD app, or a web front-end.
Something that's the output of mass-effort machine learning algorithms is categorically different to human-made-and-understandable software. Do we have terms for these different categories?
"Being dependent on a small number of AI vendors, who also penetrate every aspect of your life to control, report, and spy upon you, forced to use AI at work or be laid off, or just be made redundant and let off anyway, while drowing in slop from every corner of social media, news, youtube, search results, and any other cultural production -- that's real freedom!"
I bet that person thinks driving a car is real freedom too.
(You need to pass a test, maintain a license from the government, in many places have proof of passing a maintenance test, pay an annual fee to the government, roads can be closed at will by the government. They can install tolls if they wish etc. In the UK, the police can stop you and search your vehicle without reason and I'm sure that's true in many other countries)
I'm also not sure how that ended up the definition of real freedom. Maybe some Spinoza is in order but genuine freedom is proper understanding of one's own, and others, nature and causes. That is to say one is free to the extent that one has adequate knowledge.
To live in ignorance, that is to be pushed, pulled or compelled by what you don't understand, for example because you've outsourced your life to the magical mystery programming box, is in his terminology bondage.
Which also explains why all these AI psychosis people sound like crack addicts. Stallman is still completely correct about what software freedom is
I would guess that for most people AI still promotes this definition of freedom.
The modern world is too complex for anyone to understand all topics fully. So as a result of this cognitive cap, by Spinoza's definition people have grown progressively less free for centuries. AI can, for anyone who seeks it, allow easy access to fields where they previously would have to consult with a human expert in order to begin to understand. Coding agents arguably make coding more accessible then it's ever been, it's just that as a side effect they make understanding less necessary to produce software (and thus also less valuable). Any schmuck can now ask an agent "hey how does this program work" and get a competent answer in terms they can parse. Of course, most people do not do this. Human beings, as a general rule, do not do things they don't have to. But the option is there.
As a board saturated with experts, people here on HN don't tend to appreciate what kind of incredible super power vibe coding is for people who did not pursue an education and career in software engineering. Everything that a computer can do is now accessible to everyone. As the vast majority of people on earth would not and could not spend the time and effort to understand software (or simply don't have the talent to do so), this is a pure upgrade in terms of average freedom.
Other ones being to keep paying the AI bros in perpetuity, and having all your prompts, data, and IP getting stolen by the same people with the goal of eventually undermining your own ability to think. We will all be as free as the people living in Pods in Matrix.
> The irony is rich that AI has turned Stallman's goal of software
freedom on its head. Having your computer do whatever your
imagination can think up -- that's real freedom.
The irony is that Stallman promoted free software and allot of code got published online for free, and now LLMs have trained on this code. Without Stallman there will be no vibe coding.
I don’t find that ironic. As the post says, AI gives people the ultimate freedom to materialize Stallmans world where everyone can modify the software they use any way they want. The fact that his movement helped that become a reality seems the opposite of irony.
Sadly I moved away from Emacs as my daily driver and I just orchestrate agents in tools I build. Emacs will have a lasting impact on me and I will probably be back somehow.
Yeah, I really need to let AI write more useful elisp bits for daily life. Like a magit commit message writer, esp. the GNU Changelogs. And much more such thinks.
⟩ Refactor these components into standalone modules that communicate across well-defined APIs. Write the API specifications to docs/API.md. Add a note to AGENTS.md instructing agents to preserve the modular seams and focus work on one module at a time.
We are though. Why do you think Rust has become the defacto standard for vibe-coding? Most vibe coders can't read a line of Rust, but the strong ML-inspired type system, and also having native performance, is why it is the king of the hill right now.
> RMS appeared in a recent video recounting his CSAIL work on a circuit analyzer in which he, and I'm paraphrasing, "didn't model the circuit as a system of equations to be solved, but rather approached it much as a human would" by applying a series of diagnostics.
> But there's one glaring headscratcher, and that's its full-blown,
self-contained implementation of a terminal emulator, in Rust no less. In the before times, a developer would undoubtedly seek such a heavy
lift from off-the-shelf kit, but the holmes replicated over a
coffee break what a billionaire with infinite resources had been
toiling over the last five years.
Is this really true, though? Did model improvement since Ghostty began simply eclipse Ghostty and anything achievable with that now "legacy" code base? Would that project be better off abandoned or started over from scratch with Opus 5.5? Or is there something more there?
There is, in terminal usage and programming lingo, "raw terminal mode" and "cooked mode" ; not seeing the original discussion linked, I do wonder if "cooked" was used in both meanings of the word and not just 1.
It's clearly in that sense that it's called "cooked". Which other meaning do you think applies?
The README of the project itself explains; "cooked" allocates the pty and interprets dynamically whether the other side is a command-line reader or full-screen program. See: https://github.com/vodik/cooked
"We may have felt safe bringing a gun to a knife fight, but now everyone's just sending a drone." Perfectly said. For those of us concerned about editor efficiency, it's hard to know where to go from here. For years I'd been working on a design tool that was inspired by Vim - entirely keyboard driven using mnemonic keychords. But LLMs really took the wind out of my sails, because eh, why bother now? Somewhere deep down I feel that it's still viable, because I want it to be true that perfectly deterministic tooling is still important. I just don't know if it will be considered important by everyone else.
> "I want it to be true that perfectly deterministic tooling is still important."
Of course it is. The AI coded apps only (debatably) work as end-user apps. They are thin little bumps on top of the software mountain. So much, including all the AI tools, depends on efficient, perfectly deterministic tooling.
It is almost hypocritical for AI enthusiasts to say we don't need to look at code anymore, while everything they do is build on piles and piles of infrastructure and tools that are carefully hard-coded, deterministic building blocks. The whole thing would come crashing down if everyone embraced vibe coding at every level.
I am just constantly baffled by people thinking that because we can make (crappy) end-user apps and one-off tools, no one needs to code anymore. Does anyone build solid tools and libraries? Does anyone still get pleasure from building a rock solid, snappy app?
I can feel all the AI coded apps and website turning to ever buggier sludge under my fingertips, and people are rejoicing that this is the way of the future.
Where were those apps even before AI came along? Every piece of software I use in my day across Windows and iOS is a slow, laggy, piece of resource hogging garbage, and that was true long before AI came along. I'm not convinced that AI makes this worse, it might make it much better.
i used a ton of good apps on windows xp and 7 (sumatrapdf, codeblocks, notepad++,…), macos (things, bear.app, cleanshot,…) and Linux. But that was a bit before SaaS and their electron clients. Now I still maintain a lean desktop, but it’s mostly standalone software with no reliance on the “cloud”.
We used to have them—in the 2000s and before. These days not so much, but there appears to be a correlation between the push to use LLMs and a hockey stick in the enshittification curve.
Same. That is why I am over in a corner happily coding away with Common Lisp and McCLIM to write myself an editor that mixes ACME UI with Lisp. Is it ultra-polished? Hell no. But it definitely keeps me sane and I know what I wrote.
I understand the sentiment. I do think it's important to experience it for yourself - not to convince you (I'm less and less convinced as the months go by) - but just to understand the visceral feeling that leads people into AI psychosis.
At times it feels literally magical. But more and more I feel like my brain has slumped onto the couch, lazily barking at a butler to do my bidding. Like a weird kind of professional brainrot.
I read about a guy who was doing a documentary about crack addicts. They told him he needed to try it once to understand. He did. He then got addicted, lost everything and spent years on the street. Years later he finally got sober and finished the documentary.
> But more and more I feel like my brain has slumped onto the couch, lazily
> barking at a butler to do my bidding. Like a weird kind of professional brainrot.
One metaphor could be a car. You have to motivate yourself to get out and walk now, since a car can take you almost everywhere.
Like the people who circle the parking lot trying to get the parking space closest to the store entrance, you can live your life around the "convenience" and not see the problems with what you're doing. Or you can be the person who intentionally parks far away from the entrance so you get a few more steps in. Or the person who cycles to the store. Many metaphors murdered malignantly.
But since this is mental rather than physical, we might end up being couch potatoes like you said, flipping through channels rather than reading, or doing a hobby. (is "doing" a hobby correct? It sounds like "doing a business")
> The whole thing would come crashing down if everyone embraced vibe coding at every level.
I think this is an interesting hypothesis and want to see what happens. I'm not really willing to go all the way to vibecoding, because I want to understand the system design. But I do have bots writing all the new code.
>Of course it is. The AI coded apps only (debatably) work as end-user apps. They are thin little bumps on top of the software mountain. So much, including all the AI tools, depends on efficient, perfectly deterministic tooling.
Most of that is also written via AI these days... Even the Linux kernel got that bug.
and we'll see how that plays out. The way I loom at the chains of "AI" adds feature and feature needs 5 extra PRs over a year to fix it, I think this will be fun!
I more or less agree with the article but we can move it one step further. Nowadays we don't even need text editors no matter how advanced they are. I am a VS Code user and I have not opened VS Code in more than 3 months because I don't need it anymore. Opening a text editor and editing code by hand seems like so ... yesterday.
This definitely is a new era and, personally, I love it! (Also I have broken my right arm so I can't write code even if I wanted to. I'm dictating this text.)
I was saying almost this same thing to a coworker. I'm building something that's like Emacs, but for agents, and it's mostly a Markdown-forward tool for managing all the agents and agent coding harnesses, and it's a vendor-agnostic things....If anything it's like a newer, faster, fancier org-mode, but knows how to inject all the right context into all the agentic projects.
This, after configuring Emacs to talk to LLMs. But that's not the right angle now for me....
LLM-assisted development really is like Literate Programming. I haven't used LP in a while, but I wonder how useful that workflow would be now. Almost like the reverse of LP, lots of agent work punctuated with smaller bursts of human-created text.
Emacs is not an advanced text editor, it's more like an OS. I run Claude Code inside emacs (with the claude-code-ide package). I seriously think LLMs will usher in a new golden era for emacs because you can now reap all the benefits without having to learn elisp.
I did, with another package, but running Claude inside a terminal has no glitches and I seldom have to edit code anymore. I'm using emacs almost only to edit md files in which I make Claude write analysis and plans. I make changes, add clarifications, we iterate. Most of the times the chat is enough. I remember that I discovered that I broke Ruby autocompletion only weeks after I did (I investigated the cause.) Maybe notepad would be enough but I stick to emacs. I am about to get to 40 years of marriage with it.
If everyone is just vibe coding elisp, who is left to learn it deeply and push it forward as a language and ecosystem? Or is AI just going to do that too. Very unfortunate world we live in
I spend a lot of time thinking about AI’s effect on me, as a person. TLDR: it’s a mixed bag: for non-coding things, AI has helped me learn and understand the world better. For coding, using AI coding harnesses and agents feels like an addiction.
I will say that I have used AI coding tools to make my Emacs setup the best it has ever been in 40+ years using Emacs.
It is weird to be using tools that sometimes make me think more about myself and their effect on me than the actual technical work I am doing.
> sealed up transistors can reproduce the same outputs given the same inputs
with unerring precision. The power to unambiguously differentiate an
on- versus an off-signal -- the binary "digital" -- might seem
trivial, but is 100% the reason why we're so comfortable handing over
the reins to computers.
Is the author not familiar with the ECC memory business?
I'm glad I'm not the only one who see this comment a lot, and hates it. Computers are only "deterministic" because they have about 47 layers of error correction between you and reality.
> Whither now, we ask -- now that we have, even to my
jaded eye, a god machine that makes short work of all the stuff we
emacsers believed we were faster at than people using lesser tools.
Editing speed, customization, the expressiveness of LISP: every
advantage we claimed vanished the moment we stopped typing and started
telling the AI what we wanted.
Is that a common thought? I learned vim, then emacs, and various other tooling not because I wanted to increase my speed competitively, but just to reduce my frustration with the gap between thinking and doing. Using something like VSCode to code is like playing on a laggy piano (if that exists). You think about I need to quickly update the signature of this function and it turns into a mouse/keyboard chaotic dance with your eyes focusing all over the place. It's like doing a three legged race.
With emacs and vim, I think about the code, notice that I want to change to change something, and can do the necessary actions (almost) subconsciously. Which means, I never stop thinking about the code. It may be longer or faster than someone using another software, but the advantage for me is to rely mostly on muscle memory while I'm thinking deeply about some problems.
Another thing is multiple buffer/views at once. This is why I don't like editors that have vim mode, because it's a pain to browse code with them, especially when you want a global view of something. With both emacs and vim, I can have multiple related files/sections of files/output of commands visible at once. Like how you would want physical papers and lay them on a desk.
Yeah, I think framing in terms of 'effort' or 'flow' might make more sense than 'speed' here.
If you expect that you can complete some task just from the keyboard, then the idea of having to use the mouse interrupts what you're doing.
The original post gives "stage and commit in git" as an example. Magit is much nicer than CLI, since you can quickly invoke the actions you want; it's less effort than typing out the full command on the CLI. -- With LLM coding agents, getting an LLM agent to make a commit is slower than doing it yourself in magit (if all you're going to do is stage and commit), but it's much less effort.
> The irony is rich that AI has turned Stallman's goal of software freedom on its head. Having your computer do whatever your imagination can think up -- that's real freedom. The only catch is
you'll have no idea how it works. No one does.
Why are people so sure this will keep scaling? We're still so far away from me being able to get anything even remotely close to what I can imagine using a free model running on hardware I own.
Then you need to remember there are a lot of very powerful people currently invested in not letting us have any of this, so I really don't think it's time to call free software irrelevant yet.
Oh they __WANT__ you to have it... but you gotta $$$ for it... every month. Anything your imagination can think up... just put your card # in... that's American freedom!
>ut at the time, rules-based
programming was still novel enough to be considered a viable approach
to AI.
Pick MLite from https://t3x.org and you'll get it pretty fast. Or with invariants under Lisp. Or its weird cousing, CLIPS, born from Common Lisp clauses.
On code quality, Neomacs (an Emacs clone in Rust, it has Elisp support) crashes a lot. And by a lot, I mean once you hit the menu with the mouse, it segfaults.
>he only catch is
you'll have no idea how it works. No one does.
Is not about the rules, but the relations. It's a network of clauses, and not something akin to if-then-else with flow controlling. There's no actual "flow".
I didn't got a crash on Linux/Wayland. And I didn't tested on MacOS or Windows.
What's your device envionment?
Could you please submit a GitHub issue to neomacs? (better with a minimal reproducible script and log `RUST_LOG=INFO ./neomacs` can output the debug log)
It's a highly experimental thing someone decided to mess around with, without any prior design discussion with any Emacs core devs. It's like a forty-minute-old baby and you're asking it to parallel park without crashing. What's your point here, I don't get it?
GNU/Emacs is incredibly stable. It's one of the most reliable pieces I've ever had; I sometimes run it for months without ever restarting. Which I can't say about pretty much any app on my system. Including the system itself.
It's a txt file. If you don't know what that is, it's bit like a markdown file, but without the emojis and bullet points. If you can't consume a txt file, that's a failing on your part.
I use it mostly as a personal StackOverflow. I describe my issue as best as I can and get answers/suggestions, sometimes a code snippet that I check and paste over, but often still writing my own code or rewriting the snippet so much that it's as if I wrote it in the first place.
The friction makes it much slower than agentic coding but it helps me stay in control and it's still faster than before.
Here's a fun question - if AI is trained on OSS, particularly with licenses that are copy left... wouldn't that make all the code it generates copy left? wouldn't that mean the weights themselves should all be open sourced.
Threatening to whom? Politicians are in the pockets of the same billionares controlling AI. And even if they weren't, AI is useful to them for controlling everything.
Like demanding commented machine code between 1985 and 2025.
Sure you can get it, but that's not where the software lives even if it's how the software runs.
Reverse engineering closed source software is the simplest thing in the world now. The only reason why more of it isn't done is because it's so boring even minding the AI isn't worth most people's time.
Fable 5.2 is designing entire devices now, from PCB layout to case assembly.
If you're in a STEM program at university, quit. By the time you graduate, you will have negative value proposition compared to Claude. Get into business. The only way you're going to make it is to join the ownership class. Or pick up a torch or pitchfork and join the revolution.
This is such a wrong way of looking at it. The fact that AI is getting better and giving us better toolings is more of a reason to go and get an education.
Tools are an extension of your mind. AI is becoming like unto a mind of its own. It's taking the power to make decisions away, rather than speeding your execution on such decisions. Note that by "decisions" I don't mean just higher-level decisions, but the micro-decisions about details that go into crafting a thing also. And we're being told oh, just accept this new layer of abstraction (on top of all the others). And don't worry about the details anymore, the machine does them so much better. I mean, sure you can work at a detailed level with an LLM, but on the job they'll PIP you for not just trusting the dark factory. Judgement is the new moat. Wait, Opus 5.5 is pretty good at that too; business context is the new new moat. It's just an upward retreat into fewer and fewer, less and less concrete decisions made by people, until engineering simply isn't done by people anymore; there's only a need for human business people, product managers, and QA testers. And then AI starts coming for their jobs too, especially since the CEO has been using ChatGPT to do product-market fit analysis for what potential things we might ship to maximize profitability over the next FY...
As for university, I actually don't really know what to tell kids these days. Many of them will succumb to pressure to keep their GPA up, and use AI to cheat to ensure that they do. So whatever they choose, unless they are super disciplined, super honest, and super incentivized to come by their degree without cheating, they won't get a whole lot of learning out of it. Especially not for the money they're spending now.
There are several points of view here. Some people claim they will never need to "program" again because the LLM will do everything. That text editors are dead... and so is the career of programmer or software engineer. That nowadays, with Claude Code, you can reimplement anything (given a large enough context window, of course). That it is like being an expert in every programming language and system out there. That we should remove politicians and CEOs and put an AI in their place.
On the other hand, many people claim that machine-generated software STINKS. That it is extremely verbose, that it reimplements similar methods over and over again. That it would rather cut corners at a low level than build high-level abstractions. That it pulls in libraries and makes serious security mistakes. That it writes tests that are basically incomplete. That it is unable to synthesize and keep context. In short, it has the same flaws as humans.
I think these are binary points of view, and both are right. We have vampires that are incredibly fast, super-strong, super-intelligent and super-agile, that suck your blood (your money), and that die as soon as the sunlight of security or complexity hits them.
There has to be another way. Nobody questions the versatility LLMs provide, or the boldness they give you to take on new projects. But I am not convinced that programmers or software engineers are going to disappear. I am not convinced that a manager can ship a new feature in a couple of afternoons and then lay off, or skip hiring, a junior or senior developer. There is a line near the beginning of Dune that goes something like this: "there was a time when men handed their thinking over to machines in the hope of being freed, but that only allowed other men with machines to enslave them."
We need to find a new path between smartphone addiction and being dumbed down by LLMs.
> On the other hand, many people claim that machine-generated software STINKS. That it is extremely verbose, that it reimplements similar methods over and over again. That it would rather cut corners at a low level than build high-level abstractions.
I spent most of the early 2010s rewriting systems that had been outsourced to budget offshore coding shops in the 2000s. The software looked a lot like this: no code reuse, no system architecture, just copy pasta over and over.
The client didn't care about this of course, they weren't reading the code. What made them eventually fire the outsourcing shop and hire a guy like me was the bugs that wouldn't get fixed, and the request to fix a bug that caused three more bugs to get introduced.
I'd rewrite the software and it'd be functionally complete with 1/5th the code, and it'd be faster and have fewer bugs, as well as being maintainable.
I actually developed a sick enjoyment about reading through their code and understanding what they were doing. From what I can tell the guys were being put under pressure to produce code and close tickets. I would find big chunks of dead code that I think was to fool their SLOC counters.
Changing schema on one project required a change order and migrations submitted to DB folks so these guys discovered that mysql booleans were actually integers and they could stick whatever number they wanted in there, and ended up using existing boolean fields to store unrelated data! Of course this broke the original functionality in difficult to track down ways!
It's like a tractor or a chainsaw. There is still just as much need for good engineering, if not more, than before.
> We have vampires that are incredibly fast, super-strong, super-intelligent and super-agile, that suck your blood (your money), and that die as soon as the sunlight of security or complexity hits them.
Haha. I imagine Security and biology are the garlic and herbs that drive away the vampires and then the vampires sends their lesser minions to do their job.
> there was a time when men handed their thinking
Reminds me of the "knowing where to tap" story. Some already know where and how to tap and only delegate the tapping part. Some start from scratch and expect the machine to tell them where to tap.
The management, like the ship owner thinks that they don't need to call anyone. That's not the main risk though, it's the majority not caring enough to learn about the ship.
> In short, it has the same flaws as humans.
only the clueless, stupid ones
I'm wondering- did you read the article? Because you seem to be answering conclusions that it didn't reach.
Software engineers know that AI code is kind of bad and leads to issues down the road.
Shills and luddites say AI is capable of replacing SWE
Yes, great argumentation. Anyone who doesn't agree with me is either a shill or a luddite.
> [W]e are here to declare our intent to use and develop a version of Emacs that is free from LLM-generated contributions regardless of whatever policy GNU decides upon. We invite them to weigh the costs and do the right thing, in which case we will continue to support them. But if they are not willing to do so, we will continue to use and develop an LLM-free version of Emacs anyway, even if it is not GNU Emacs.
https://human-emacs.org/
good to see treesitter is now vibecoded too...
Woah since when, that one hurts
That's obviously the best use of an AI's time.
Have you ever written parsers for complex languages? It is a cumbersome slog.
Yeah it's also super easy to automate testing.
it's on the responsible side of vibecoding. Effectively since a person took over who decided that Claude helps them go brrrr...
The way we are accepting this is extremely unsettling...
Gonna get its milkshake drunk if GNU decides to allow AI.
How could they survive with fewer users and lower revenue?
Ah right, it’s free software. There is no milkshake, only people with similar ideology contributing to a common cause.
The disappearance of bikeshedding doesn't look like an obvious one but it's a huge loss. Imagine if every scientist resolved to refer to already written books and carry on their research in closed labs with no discussion with the rest of their field.
Yeah, I look for opportunities to replace personal vibe-coded utilities with public projects for this reason.
Software still feels like this grand exploration to me—what's the best way to use a computer? What's the best way to express a program?—and a hackable text editor epitomizes that ethos (though I'm partial to neovim myself). Every time I install a plugin, I feel like I'm participating in a clinical trial of UX design.
I think a lot of us feel this, and the community will have to solidify the ethos that trying out other peoples' software is noble and healthy, exposing you to their ideas, even if it's actually harder and worse in some ways than vibecoding.
(I also think, by extension, the OSS community will need a more developed ethics of project governance. This was a hot issue even before AI, e.g. the terraform license change. But when the default is a vibe-coded bubble, authors will need to communicate either
1. this is just code—use at your own risk, PR at your own risk, or
2. this is a community, here's what we want for/from users, here's what we want for/from contributors, here's how to get involved in maintainance/governance.
You can be either but you have to say which.)
There's a principle I can't describe clearly. But one instance of it is the homogenization due to cheap(or free) instantaneous communications. Slow and limited created space and time to share, and incentives to do so because you needed to discuss with others to get or form new knowledge. Now everyone refers to the same source (see the evolution of design, choice of colors). Limits are protective in a way.
A similar principle to that of lower-budget films, where constraints force filmmakers to find creative solutions. Meanwhile, films with seemingly unlimited budgets can somehow end up with bland, homogenised stories and surprisingly poor special effects. Limitations are a catalyst for creativity rather than an obstacle to it.
> Imagine if every scientist resolved to refer to already written books and carry on their research in closed labs with no discussion with the rest of their field.
Open source projects have always had forks, public and private. The ability to fork a project and use your own version of it privately, without needing permission or buy-in from others, has always been an explicitly valued aspect of open source software. If it was always seen as a good thing before, why do we see it differently now?
I think the reason is an implicit assumption that the production of more unshared work means that less work will be shared. But I don't see why that would be true. We could just as easily, and more likely in my view, see more work shared with the world at the same time we see more work not shared with the world. The incentives for cooperation remain, even as it becomes easier to work alone.
Has anyone made a persuasive argument to the contrary, or provided evidence? It seems like the main problem AI has created for open source projects is an avalanche of bad PRs, not a drying up of good contributions.
> If it was always seen as a good thing before, why do we see it differently now?
Writing code shouldn't merely be about accomplishing a goal, but making deliberate and good decisions that lead to more good decisions with increasing ease and clarity. It's no different than any other type of planning, engineering, or architecting. FOSS demands this more than any other genre of software because it's made collaboratively.
I think AI is just too irresistible for the reckless and dumb developers who were, unfortunately, always the majority in our line of work. Now we don't have to guess at that fact anymore or be afraid of being judged as arrogant for saying it. Now we know. These developers don't bring anything to the table. The problem of rejecting garbage PRs is far from new.
> That no one's yet done it is clear evidence that no one really wanted these things so much as they wanted to talk about it
1. There are a few, already. I haven't systematically surveyed the landscape, but here's one: https://neomacs.com/
2. AI coding is still far from mature -- with substantial improvements even in 2026. I expect the number of attempts will only burgeon as this capability matures. I can already see a proliferation of emacs packages written with AI.
3. I would expect hobbyist activity to be more correlated with the capabilities of open-weight & cheaper models, for a variety of reasons. Pure hobbyists (esp. outside Silicon Valley) don't often have the spare $$$ for a $200 subscription plan, and won't be as early to adopt AI coding as "professional" software engineers in Silicon Valley (or geographically / economically adjacent somehow).
Neomacs is about 95% complete it seems. I just ran it and it used my init.el* just fine. There are still significant bugs, and if I had spare tokens I'd probably contribute some towards it.
* probably yak shaved and far too much for what I use it for
“AI has turned Stallman's goal of software freedom on its head. Having your computer do whatever your imagination can think up -- that's real freedom. The only catch is you'll have no idea how it works. No one does.”
Nobody knows how Stallman, or anyone else, works either. But folks don’t need to understand how he works to understand, say, EMacs.
Folks can take EMacs as is and decipher it in isolation of Stallman, or any of the other authors.
But even then, how many people understand the internals of EMacs, or the interrelationships of the zillion lines of code in hundreds of files? Yet, millions of ignorant people leverage and apply EMacs to their daily tasks.
Look at what the REA impact is having on all sorts of code bases. The ONE THING that pushed Stallman over the edge and onto his path. The lack of source code to a print driver (or something along those lines).
Well, it’s all Open Source now.
Not only is the source “open”. It can be explained.
“Here is the source tree to EMacs. ELI5 the garbage collector.” “Working…”
It goes both ways. Now everyone has access to a savant to explain software, if that’s what they like.
A savant who confabulates about 10 % of the time, per the best benchmarks, no matter how much “thinking” it does (GPT-based LLMs will always “hallucinate”)? How do you trust it?
I would rather take the time to ask the same question to the Emacs contributors mailing-list, who have the incentive to check their posts for mistakes.
Have you ever worked with a modern agent? They let many weak assumptions slip because their cognitive functions aren't remotely as deep as human's (that's what makes their code a horror), but they have also been trained to hyperfocus on verifying and grounding everything they do and say, you have to actively stop them from doing so when it's unnecessary. This not what I can call confabulation in any good faith, let alone 10% of the time. I think you've been misled by the old paper you cite below, it doesn't say what you likely think it says.
Yes, I mainly work with OpenCode, with GLM as an orchestrator agent and DeepSeek Flash as a worker agent. I’m assuming it’s modern? I don’t have the money to use either Fable or Astra, but my assumption is that the few more points they get on the benchmarks still don’t make up for the wild confabulated things they do in my projects.
Both are fine, though frontier models are more persistent. They mostly take unwarranted shortcuts, trying to find the easiest way to reach the literal goal set by you, failing to see the broader picture. But they definitely do not confabulate, as in they don't just make up nonexistent things, they would rather walk extra mile to verify when they're not certain, which is the reason the software they make is trash but it works in the narrow sense. They can be rigid in their writing style due to mode collapse, but I haven't seen this leading to them inventing nonexistent functions or anything like that in a long while (and it's a job for formal tools to catch that, in any case).
Modern LLMs are very far from being next token predictors trained only on text. Text and other modalities is the base that makes them understand your language and have concepts similar to yours, but they were also subjected to RL that makes them good at reaching the goals. Two tricks in a trenchcoat instead of just one.
> But they definitely do not confabulate, as in they don't just make up nonexistent things
I've definitely seen both Labs' "best" stuff confabulate to this day.
I can't speak for the other providers but Gemini carbon feels like the difference between gpt3 and gpt5 when compared to Chinese open models.
How do you trust a human? Are humans infallible, or at least better than AI at explaining things? I known my answer.
I trust the Emacs contributors, the actual experts in the topic, to fact-check their own answers, because they are incentivized for doing so (reputation). They are also able to tell when they aren’t sure about a particular topic.
GPT-based LLMs, by design, cannot. https://arxiv.org/abs/2409.05746
You don't understand Stallman then. Now you are dependent on closed source from closed source AI vendors. True software freedom would then being able to dissassemble the bot writing your code.
It's clear that without hardware freedom, even if we disassemble the bot, the means to run it can still be controlled by hardware vendors. We need our own means to make the substrate for the intelligence too. But the problem is, as long as the manufacturing process is impossibly complex, and relies on minerals extracted all around the world, requiring complex supply chains, its beyond any local group of people to do themselves. What we need is a way of creating substrate for open intelligence just form the materials and resources that are available to any human, and then the freedom to train it to behaving in a way that alignes with our own values and local situation. But the problem with OSS adoption, is that its a real chore for normal people to set up and use it, so we need to make this process of creating our own intelegence host as fun as possible, to encourage people to do it..
This. Nailed it. This right here is the core issue. Thanks for writing.
No one knows how the open source models work either.
Even if the model was open source, nobody would still know how the generated code works. You wouldn't even really know in want interesting sense of the word why the model produced the output you got.
Open source really should mean open training dataset, not merely open weights. There are some truly open source models and you can reproduce their training process if you have the compute. I can't think of a reason why those kinds of models would be incompatible with Stallmanism. Presumably they will get better and more widespread over time, like free software
LLMs are even more proprietary than regular proprietary software, I think you've got it right here. They are destroying the free software movement and it seems like most people don't even realize what's happening.
LLMs are a fundamentally different category of thing to software. Developing them is closer to breeding cows than writing code.
I definitely agree on this. You don't program an LLM and thus come to an understanding of what it does in the same way you do for a compiler, an operating system, a game engine, a CRUD app, or a web front-end.
Something that's the output of mass-effort machine learning algorithms is categorically different to human-made-and-understandable software. Do we have terms for these different categories?
The damage isn't restricted to software. AI is the intellectual equivalent of asbestos.
Asbestos is an amazing fire retardant. That's why nearly every building built before we knew how cancerous it was contains it.
"Being dependent on a small number of AI vendors, who also penetrate every aspect of your life to control, report, and spy upon you, forced to use AI at work or be laid off, or just be made redundant and let off anyway, while drowing in slop from every corner of social media, news, youtube, search results, and any other cultural production -- that's real freedom!"
A freedom that only works as long there is an AI model one is able to reach for, with stolen data from the whole known universe.
I bet that person thinks driving a car is real freedom too.
(You need to pass a test, maintain a license from the government, in many places have proof of passing a maintenance test, pay an annual fee to the government, roads can be closed at will by the government. They can install tolls if they wish etc. In the UK, the police can stop you and search your vehicle without reason and I'm sure that's true in many other countries)
I'm also not sure how that ended up the definition of real freedom. Maybe some Spinoza is in order but genuine freedom is proper understanding of one's own, and others, nature and causes. That is to say one is free to the extent that one has adequate knowledge.
To live in ignorance, that is to be pushed, pulled or compelled by what you don't understand, for example because you've outsourced your life to the magical mystery programming box, is in his terminology bondage.
Which also explains why all these AI psychosis people sound like crack addicts. Stallman is still completely correct about what software freedom is
I would guess that for most people AI still promotes this definition of freedom.
The modern world is too complex for anyone to understand all topics fully. So as a result of this cognitive cap, by Spinoza's definition people have grown progressively less free for centuries. AI can, for anyone who seeks it, allow easy access to fields where they previously would have to consult with a human expert in order to begin to understand. Coding agents arguably make coding more accessible then it's ever been, it's just that as a side effect they make understanding less necessary to produce software (and thus also less valuable). Any schmuck can now ask an agent "hey how does this program work" and get a competent answer in terms they can parse. Of course, most people do not do this. Human beings, as a general rule, do not do things they don't have to. But the option is there.
As a board saturated with experts, people here on HN don't tend to appreciate what kind of incredible super power vibe coding is for people who did not pursue an education and career in software engineering. Everything that a computer can do is now accessible to everyone. As the vast majority of people on earth would not and could not spend the time and effort to understand software (or simply don't have the talent to do so), this is a pure upgrade in terms of average freedom.
Other ones being to keep paying the AI bros in perpetuity, and having all your prompts, data, and IP getting stolen by the same people with the goal of eventually undermining your own ability to think. We will all be as free as the people living in Pods in Matrix.
> The irony is rich that AI has turned Stallman's goal of software freedom on its head. Having your computer do whatever your imagination can think up -- that's real freedom.
The irony is that Stallman promoted free software and allot of code got published online for free, and now LLMs have trained on this code. Without Stallman there will be no vibe coding.
I don’t find that ironic. As the post says, AI gives people the ultimate freedom to materialize Stallmans world where everyone can modify the software they use any way they want. The fact that his movement helped that become a reality seems the opposite of irony.
Sadly I moved away from Emacs as my daily driver and I just orchestrate agents in tools I build. Emacs will have a lasting impact on me and I will probably be back somehow.
Yeah, I really need to let AI write more useful elisp bits for daily life. Like a magit commit message writer, esp. the GNU Changelogs. And much more such thinks.
It's a whole new world.
We need to be thinking about modularity with well-defined APIs now, to isolate the chunks created by AI coding. Maintainabilty requires firewalls.
Strong isolation where it matters, completely free otherwise.
Isn't that just... software engineering?
No, it's:
⟩ Refactor these components into standalone modules that communicate across well-defined APIs. Write the API specifications to docs/API.md. Add a note to AGENTS.md instructing agents to preserve the modular seams and focus work on one module at a time.
It's a weird world where we're refactoring everything, but not just using a type system that can enforce those seams and limitations at compile time.
We are though. Why do you think Rust has become the defacto standard for vibe-coding? Most vibe coders can't read a line of Rust, but the strong ML-inspired type system, and also having native performance, is why it is the king of the hill right now.
"Just"?
> RMS appeared in a recent video recounting his CSAIL work on a circuit analyzer in which he, and I'm paraphrasing, "didn't model the circuit as a system of equations to be solved, but rather approached it much as a human would" by applying a series of diagnostics.
bitter-lesson-pilled RMS before GTA VI
Loved it.
> But there's one glaring headscratcher, and that's its full-blown, self-contained implementation of a terminal emulator, in Rust no less. In the before times, a developer would undoubtedly seek such a heavy lift from off-the-shelf kit, but the holmes replicated over a coffee break what a billionaire with infinite resources had been toiling over the last five years.
Is this really true, though? Did model improvement since Ghostty began simply eclipse Ghostty and anything achievable with that now "legacy" code base? Would that project be better off abandoned or started over from scratch with Opus 5.5? Or is there something more there?
These are honest questions.
There is, in terminal usage and programming lingo, "raw terminal mode" and "cooked mode" ; not seeing the original discussion linked, I do wonder if "cooked" was used in both meanings of the word and not just 1.
It's clearly in that sense that it's called "cooked". Which other meaning do you think applies?
The README of the project itself explains; "cooked" allocates the pty and interprets dynamically whether the other side is a command-line reader or full-screen program. See: https://github.com/vodik/cooked
> Which other meaning do you think applies?
https://www.merriam-webster.com/slang/cooked
https://www.urbandictionary.com/define.php?term=cooked
Given that we're talking about terminal emulation, I'd be more surprised if "cooked" wasn't a reference to cooked mode
"We may have felt safe bringing a gun to a knife fight, but now everyone's just sending a drone." Perfectly said. For those of us concerned about editor efficiency, it's hard to know where to go from here. For years I'd been working on a design tool that was inspired by Vim - entirely keyboard driven using mnemonic keychords. But LLMs really took the wind out of my sails, because eh, why bother now? Somewhere deep down I feel that it's still viable, because I want it to be true that perfectly deterministic tooling is still important. I just don't know if it will be considered important by everyone else.
> "I want it to be true that perfectly deterministic tooling is still important."
Of course it is. The AI coded apps only (debatably) work as end-user apps. They are thin little bumps on top of the software mountain. So much, including all the AI tools, depends on efficient, perfectly deterministic tooling.
It is almost hypocritical for AI enthusiasts to say we don't need to look at code anymore, while everything they do is build on piles and piles of infrastructure and tools that are carefully hard-coded, deterministic building blocks. The whole thing would come crashing down if everyone embraced vibe coding at every level.
I am just constantly baffled by people thinking that because we can make (crappy) end-user apps and one-off tools, no one needs to code anymore. Does anyone build solid tools and libraries? Does anyone still get pleasure from building a rock solid, snappy app?
I can feel all the AI coded apps and website turning to ever buggier sludge under my fingertips, and people are rejoicing that this is the way of the future.
I refuse. I will not touch Claude.
"rock solid, snappy apps"
Where were those apps even before AI came along? Every piece of software I use in my day across Windows and iOS is a slow, laggy, piece of resource hogging garbage, and that was true long before AI came along. I'm not convinced that AI makes this worse, it might make it much better.
i used a ton of good apps on windows xp and 7 (sumatrapdf, codeblocks, notepad++,…), macos (things, bear.app, cleanshot,…) and Linux. But that was a bit before SaaS and their electron clients. Now I still maintain a lean desktop, but it’s mostly standalone software with no reliance on the “cloud”.
We used to have them—in the 2000s and before. These days not so much, but there appears to be a correlation between the push to use LLMs and a hockey stick in the enshittification curve.
Same. That is why I am over in a corner happily coding away with Common Lisp and McCLIM to write myself an editor that mixes ACME UI with Lisp. Is it ultra-polished? Hell no. But it definitely keeps me sane and I know what I wrote.
I understand the sentiment. I do think it's important to experience it for yourself - not to convince you (I'm less and less convinced as the months go by) - but just to understand the visceral feeling that leads people into AI psychosis.
At times it feels literally magical. But more and more I feel like my brain has slumped onto the couch, lazily barking at a butler to do my bidding. Like a weird kind of professional brainrot.
I read about a guy who was doing a documentary about crack addicts. They told him he needed to try it once to understand. He did. He then got addicted, lost everything and spent years on the street. Years later he finally got sober and finished the documentary.
> But more and more I feel like my brain has slumped onto the couch, lazily > barking at a butler to do my bidding. Like a weird kind of professional brainrot.
One metaphor could be a car. You have to motivate yourself to get out and walk now, since a car can take you almost everywhere.
Like the people who circle the parking lot trying to get the parking space closest to the store entrance, you can live your life around the "convenience" and not see the problems with what you're doing. Or you can be the person who intentionally parks far away from the entrance so you get a few more steps in. Or the person who cycles to the store. Many metaphors murdered malignantly.
But since this is mental rather than physical, we might end up being couch potatoes like you said, flipping through channels rather than reading, or doing a hobby. (is "doing" a hobby correct? It sounds like "doing a business")
> The whole thing would come crashing down if everyone embraced vibe coding at every level.
I think this is an interesting hypothesis and want to see what happens. I'm not really willing to go all the way to vibecoding, because I want to understand the system design. But I do have bots writing all the new code.
>Of course it is. The AI coded apps only (debatably) work as end-user apps. They are thin little bumps on top of the software mountain. So much, including all the AI tools, depends on efficient, perfectly deterministic tooling.
Most of that is also written via AI these days... Even the Linux kernel got that bug.
and we'll see how that plays out. The way I loom at the chains of "AI" adds feature and feature needs 5 extra PRs over a year to fix it, I think this will be fun!
I more or less agree with the article but we can move it one step further. Nowadays we don't even need text editors no matter how advanced they are. I am a VS Code user and I have not opened VS Code in more than 3 months because I don't need it anymore. Opening a text editor and editing code by hand seems like so ... yesterday.
This definitely is a new era and, personally, I love it! (Also I have broken my right arm so I can't write code even if I wanted to. I'm dictating this text.)
I was saying almost this same thing to a coworker. I'm building something that's like Emacs, but for agents, and it's mostly a Markdown-forward tool for managing all the agents and agent coding harnesses, and it's a vendor-agnostic things....If anything it's like a newer, faster, fancier org-mode, but knows how to inject all the right context into all the agentic projects.
This, after configuring Emacs to talk to LLMs. But that's not the right angle now for me....
LLM-assisted development really is like Literate Programming. I haven't used LP in a while, but I wonder how useful that workflow would be now. Almost like the reverse of LP, lots of agent work punctuated with smaller bursts of human-created text.
Emacs is not an advanced text editor, it's more like an OS. I run Claude Code inside emacs (with the claude-code-ide package). I seriously think LLMs will usher in a new golden era for emacs because you can now reap all the benefits without having to learn elisp.
I did, with another package, but running Claude inside a terminal has no glitches and I seldom have to edit code anymore. I'm using emacs almost only to edit md files in which I make Claude write analysis and plans. I make changes, add clarifications, we iterate. Most of the times the chat is enough. I remember that I discovered that I broke Ruby autocompletion only weeks after I did (I investigated the cause.) Maybe notepad would be enough but I stick to emacs. I am about to get to 40 years of marriage with it.
Yeah. AIs let you exist without learning. A great advance for the brainless.
If everyone is just vibe coding elisp, who is left to learn it deeply and push it forward as a language and ecosystem? Or is AI just going to do that too. Very unfortunate world we live in
I mostly use Emacs not to write code, though. In fact, getting Claude to do explorations for me and output its findings as org docs is fantastic.
Great read, thanks.
I spend a lot of time thinking about AI’s effect on me, as a person. TLDR: it’s a mixed bag: for non-coding things, AI has helped me learn and understand the world better. For coding, using AI coding harnesses and agents feels like an addiction.
I will say that I have used AI coding tools to make my Emacs setup the best it has ever been in 40+ years using Emacs.
It is weird to be using tools that sometimes make me think more about myself and their effect on me than the actual technical work I am doing.
Navel gazing.
> sealed up transistors can reproduce the same outputs given the same inputs with unerring precision. The power to unambiguously differentiate an on- versus an off-signal -- the binary "digital" -- might seem trivial, but is 100% the reason why we're so comfortable handing over the reins to computers.
Is the author not familiar with the ECC memory business?
I'm glad I'm not the only one who see this comment a lot, and hates it. Computers are only "deterministic" because they have about 47 layers of error correction between you and reality.
"We may have felt safe bringing a gun to a knife fight,
but now everyone's just sending a drone."
> Whither now, we ask -- now that we have, even to my jaded eye, a god machine that makes short work of all the stuff we emacsers believed we were faster at than people using lesser tools. Editing speed, customization, the expressiveness of LISP: every advantage we claimed vanished the moment we stopped typing and started telling the AI what we wanted.
Is that a common thought? I learned vim, then emacs, and various other tooling not because I wanted to increase my speed competitively, but just to reduce my frustration with the gap between thinking and doing. Using something like VSCode to code is like playing on a laggy piano (if that exists). You think about I need to quickly update the signature of this function and it turns into a mouse/keyboard chaotic dance with your eyes focusing all over the place. It's like doing a three legged race.
With emacs and vim, I think about the code, notice that I want to change to change something, and can do the necessary actions (almost) subconsciously. Which means, I never stop thinking about the code. It may be longer or faster than someone using another software, but the advantage for me is to rely mostly on muscle memory while I'm thinking deeply about some problems.
Another thing is multiple buffer/views at once. This is why I don't like editors that have vim mode, because it's a pain to browse code with them, especially when you want a global view of something. With both emacs and vim, I can have multiple related files/sections of files/output of commands visible at once. Like how you would want physical papers and lay them on a desk.
Yeah, I think framing in terms of 'effort' or 'flow' might make more sense than 'speed' here.
If you expect that you can complete some task just from the keyboard, then the idea of having to use the mouse interrupts what you're doing.
The original post gives "stage and commit in git" as an example. Magit is much nicer than CLI, since you can quickly invoke the actions you want; it's less effort than typing out the full command on the CLI. -- With LLM coding agents, getting an LLM agent to make a commit is slower than doing it yourself in magit (if all you're going to do is stage and commit), but it's much less effort.
This reads like someone getting AI psychosis from building one-shot prototypes. Many such cases.
> The irony is rich that AI has turned Stallman's goal of software freedom on its head. Having your computer do whatever your imagination can think up -- that's real freedom. The only catch is you'll have no idea how it works. No one does.
Why are people so sure this will keep scaling? We're still so far away from me being able to get anything even remotely close to what I can imagine using a free model running on hardware I own.
Then you need to remember there are a lot of very powerful people currently invested in not letting us have any of this, so I really don't think it's time to call free software irrelevant yet.
Oh they __WANT__ you to have it... but you gotta $$$ for it... every month. Anything your imagination can think up... just put your card # in... that's American freedom!
Yes, exactly, renting isn't freedom.
>ut at the time, rules-based programming was still novel enough to be considered a viable approach to AI.
Pick MLite from https://t3x.org and you'll get it pretty fast. Or with invariants under Lisp. Or its weird cousing, CLIPS, born from Common Lisp clauses.
On code quality, Neomacs (an Emacs clone in Rust, it has Elisp support) crashes a lot. And by a lot, I mean once you hit the menu with the mouse, it segfaults.
>he only catch is you'll have no idea how it works. No one does.
Is not about the rules, but the relations. It's a network of clauses, and not something akin to if-then-else with flow controlling. There's no actual "flow".
> hit menu with the mouse
I didn't got a crash on Linux/Wayland. And I didn't tested on MacOS or Windows.
What's your device envionment?
Could you please submit a GitHub issue to neomacs? (better with a minimal reproducible script and log `RUST_LOG=INFO ./neomacs` can output the debug log)
> Neomacs crashes a lot
It's a highly experimental thing someone decided to mess around with, without any prior design discussion with any Emacs core devs. It's like a forty-minute-old baby and you're asking it to parallel park without crashing. What's your point here, I don't get it?
GNU/Emacs is incredibly stable. It's one of the most reliable pieces I've ever had; I sometimes run it for months without ever restarting. Which I can't say about pretty much any app on my system. Including the system itself.
Obvious. Clueless vibecoding creates disasters.
It should have be written in a memory safe language instead.
Is it too much to ask to format your text in the middle aligned?
Reach out to Qwen, he can help you.
HN avant-garde article formatting to show they are cool.
Read the article in emacs. Looks great as it’s plain text.
It's a txt file. If you don't know what that is, it's bit like a markdown file, but without the emojis and bullet points. If you can't consume a txt file, that's a failing on your part.
Well browsers can't format plaintext. Better question why is he serving plaintext documents on the web in the 21st century?
So text files are considered avant-garde now?
I'm still not using AI to do more than get me in the door with finding new features, like: "tell me how to get my matplotlib output into tiff".
The AI tooling seems too unstable still.
I use it mostly as a personal StackOverflow. I describe my issue as best as I can and get answers/suggestions, sometimes a code snippet that I check and paste over, but often still writing my own code or rewriting the snippet so much that it's as if I wrote it in the first place.
The friction makes it much slower than agentic coding but it helps me stay in control and it's still faster than before.
Here's a fun question - if AI is trained on OSS, particularly with licenses that are copy left... wouldn't that make all the code it generates copy left? wouldn't that mean the weights themselves should all be open sourced.
Only if the lawmakers and courts side with the way less powerful FOSS side in interpreting the impact of those licenses on the derivative LLM code.
So, no.
Well, politically, if AI is seen as threatening, they'll see siding with FOSS as beneficial.
Threatening to whom? Politicians are in the pockets of the same billionares controlling AI. And even if they weren't, AI is useful to them for controlling everything.
The whole idea of free software seems quaint.
Like demanding commented machine code between 1985 and 2025.
Sure you can get it, but that's not where the software lives even if it's how the software runs.
Reverse engineering closed source software is the simplest thing in the world now. The only reason why more of it isn't done is because it's so boring even minding the AI isn't worth most people's time.
Fable 5.2 is designing entire devices now, from PCB layout to case assembly.
If you're in a STEM program at university, quit. By the time you graduate, you will have negative value proposition compared to Claude. Get into business. The only way you're going to make it is to join the ownership class. Or pick up a torch or pitchfork and join the revolution.
> join the ownership class
...which is rapidly running away from you. Ignore this nonsense and do useful work. It will be okay.
This is such a wrong way of looking at it. The fact that AI is getting better and giving us better toolings is more of a reason to go and get an education.
AI is not a tool. It's a service.
Tools are an extension of your mind. AI is becoming like unto a mind of its own. It's taking the power to make decisions away, rather than speeding your execution on such decisions. Note that by "decisions" I don't mean just higher-level decisions, but the micro-decisions about details that go into crafting a thing also. And we're being told oh, just accept this new layer of abstraction (on top of all the others). And don't worry about the details anymore, the machine does them so much better. I mean, sure you can work at a detailed level with an LLM, but on the job they'll PIP you for not just trusting the dark factory. Judgement is the new moat. Wait, Opus 5.5 is pretty good at that too; business context is the new new moat. It's just an upward retreat into fewer and fewer, less and less concrete decisions made by people, until engineering simply isn't done by people anymore; there's only a need for human business people, product managers, and QA testers. And then AI starts coming for their jobs too, especially since the CEO has been using ChatGPT to do product-market fit analysis for what potential things we might ship to maximize profitability over the next FY...
As for university, I actually don't really know what to tell kids these days. Many of them will succumb to pressure to keep their GPA up, and use AI to cheat to ensure that they do. So whatever they choose, unless they are super disciplined, super honest, and super incentivized to come by their degree without cheating, they won't get a whole lot of learning out of it. Especially not for the money they're spending now.
> If you're in a STEM program at university, quit. By the time you graduate, you will have negative value proposition compared to Claude
I hope you are not advising any young people out there.