Posted by ggcr 16 hours ago
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.
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.
Shills and luddites say AI is capable of replacing SWE
only the clueless, stupid ones
Have you ever written parsers for complex languages? It is a cumbersome slog.
The way we are accepting this is extremely unsettling...
Ah right, it’s free software. There is no milkshake, only people with similar ideology contributing to a common cause.
> I believe a sufficiently interested emacs developer with aThe 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.
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.)
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.
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.
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).
* probably yak shaved and far too much for what I use it for
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.
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.
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've definitely seen both Labs' "best" stuff confabulate to this day.
GPT-based LLMs, by design, cannot. https://arxiv.org/abs/2409.05746
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?
Asbestos is an amazing fire retardant. That's why nearly every building built before we knew how cancerous it was contains it.
(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)
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
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.
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.
We need to be thinking about modularity with well-defined APIs now, to isolate the chunks created by AI coding. Maintainabilty requires firewalls.
⟩ 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.
> 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.
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