Login
You're viewing the functional.cafe public feed.
  • Oct 21, 2026, 11:44 PM

    You may be surprised to learn I hold this belief/train of thought:

    - llm technology is fundamentally interesting despite big tech's best efforts to bring us the worst possible version of it
    - what makes dev tooling good for humans is also what makes it good for llms. it's the same problem
    - programming languages are not interchangeable; they have development time costs ("ergonomics") and runtime costs ("performance")
    - by merit alone, zig will eventually be the preferred lang for agentic coding

    💬 13🔄 9⭐ 50

Replies

  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 8
  • Oct 22, 2026, 12:01 AM

    @andrewrk if LLMs were, on average, producing higher quality code than humans, do you think it would be acceptable to replace humans who write code with LLMs?

    💬 2🔄 0⭐ 1
  • Oct 22, 2026, 1:15 AM

    @wolf480pl @andrewrk

    that question's for Andrew of course, but I'd like to offer up the relevant socioeconomic point on which it's based, which is that income should not be fundamentally dependent on labour. such a system is antisocial for many reasons, primarily because those who can't work still deserve an income, and also primarily because it precludes life quality for all.

    a person's views within a wrong system do not reflect their vision for a functioning world.

    💬 2🔄 0⭐ 1
  • Oct 22, 2026, 1:38 AM

    @falcennial
    Thing is, even in a world where income is not dependent on labour, I still would prefer if software that affects my life was written and understood by humans.
    @andrewrk

    💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • Oct 22, 2026, 6:48 AM

    @wolf480pl no, I find it to be entirely unacceptable. I think it's wild how eagerly and thoughtlessly software developers gave up our collective labor power in an instant. what a bunch of idiots we are

    💬 0🔄 3⭐ 22
  • Marisawitchgirls@woof.tech
    Oct 22, 2026, 12:05 AM

    @andrewrk LLMs are destroying the planet, zig being good for agentic coding doesn't really seem like it'll ever be a good thing

    💬 0🔄 0⭐ 5
  • Oct 22, 2026, 12:30 AM

    @andrewrk most people have a hard time understanding that not wanting to take contributions or directly supporting the major LLM products do not mean not understanding the technology. at the same time, most LLM supporters always push for "skill issues" when someone mention its problems, then you go look at the code they are producing with LLMs just to find out that it is worse than what you could produce yourself with the same tool and they say you have skill issues 🤡🙃

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • Oct 22, 2026, 12:58 AM

    @andrewrk I love the idea of zig, but man I feel like the only way I can write it is with llms. Syntax is so counterintuitive for my human brain for some reason.

    💬 0🔄 1⭐ 1
  • Oct 22, 2026, 1:34 AM

    @andrewrk - what makes dev tooling good for humans is also what makes it good for llms. it's the same problem

    This is a very strong claim. I don't think good visual debugging is useful to LLMs for example. There may be some stuff that is directionally useful for both in a very bad dev tooling environment but there will be divergence at some point.

    💬 1🔄 0⭐ 5
  • Oct 22, 2026, 6:54 AM

    @DerivativeThoughts @andrewrk

    Visuals are interfaces. The way LLMs and we interact with the computer/world is different, so interfaces are almost the only exception, I'd say.

    However, I've heard of some research suggests that images could help them perform better.

    💬 1🔄 0⭐ 0
  • Oct 22, 2026, 2:04 PM

    @koichi @andrewrk

    1. I'm unfamiliar with any devtool that isn't an interface to some otherwise opaque part of the system or an additional abstraction layer. Maybe devtools are broader than that it's not my exact area.

    2. LLMs don't interact with the world. They predict next tokens stochastically. Harnesses interact with the world in predefined ways based on token output. LLMs have no world model and no internal ideas. We need to be epistemologically precise about real technology.

    💬 0🔄 0⭐ 2
  • Oct 22, 2026, 2:02 AM

    @andrewrk I'm not sure I agree that Zig will be adopted by LLM's strictly for merit reasons.

    I read an article a while back that a company that was using Haskell ended up moving to python because their LLM tooling kept using python.

    Not because python was better, but because python was much more common.

    In that case I'd say whatever language is most popular will be the language of choice simply from momentum. So, either python or JS.

    💬 0🔄 0⭐ 3
  • Oct 22, 2026, 7:27 AM

    @andrewrk "I wanna really, really, really wanna zig a zig ah" — The Spice Girls

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Oct 23, 2026, 6:35 AM

    @andrewrk How do you feel about a possible future where the majority of Zig code is generated by LLMs? Do you feel incentivised or pressured to make it easier for LLMs to generate Zig?

    💬 1🔄 0⭐ 0
  • Oct 23, 2026, 4:24 PM

    @annika I don't like it, I think code should stay readable and editable by humans because that makes software more free. I think LLMs are completely unnecessary and adding them to the equation causes communities to decay and detracts from code malleability while adding an expensive dependency to the dev environment

    💬 0🔄 2⭐ 7
  • Oct 24, 2026, 7:08 PM

    @andrewrk personally I prefer zig over other languages such as rust for agentic coding, because I can review the code easier.

    💬 0🔄 0⭐ 0