Technology

Code is code. "Hello, my name is Fred." is an equally valid sentence whether a human wrote it or an AI wrote it. It doesn't contain some magical AI pollution that makes it different from the normal human-written sentence "Hello, my name is Fred."

AI written code, especially in largely "vibe coded" projects are written by people with zero experience in maintaining or developing software. They don't understand what bad code is. They have no way of recognizing it when they see it. If AI writes bad code that works, it is indistinguishable from good code that works, from the perspective of somebody who knows nothing about coding, which an increasing amount of code and an increasing number of projects are submitted to and written by.

The problem with AI is that it cannot be trusted to write good code. This is a problem when people who don't know any better start trusting it to write code. It sometimes, often, writes REALLY BAD CODE. Code with GIANT security flaws. Code that is unmaintainable, that does not fit with the rest of the code or the goals of the project.

The Linux kernel famously does not trust anyone except Linus. It has a very thorough process for reviewing all code that is suggested for inclusion, and all code is reviewed in extensive and sometimes expletive-laden detail, by Linus himself if it gets that far.

Bad AI code will be caught by the same process that catches bad human written code (which there is also a fair bit of).

Good AI code will be fine, because it gets reviewed by the same process that reviews good human written code, to ensure that it is, in fact, good code. Who actually wrote it is largely irrelevant, as long as the code is high quality and written in a safe, reliable way, because good code is still good code.

Vibe code, on the other hand, is not always good code. Sometimes it is good enough code. Often it is atrocious code. It is not a substitute for experienced software developers, if anything, it necessitates them more. Reviewing complex AI generated code quickly and accurately enough to keep up with the relentless onslaught of even more generated AI code is no minor feat, and skilled, experienced software developers are required to accomplish this.

Either it will work, or it won't and it will get removed.

I'm not really sure what's interesting about this.

AI code is like alternative medicine, it's called that when it's bad and doesn't work. If it does, it's just called code. And the issue isn't using code made by AI, it's when people who don't know how to code think the AI does, and blindly do without checking. That's very unlikely to happen with the Linux kernel, as the entire project is basically just one constant code review where it really doesn't matter if bad code was written by a human or an AI.

Even Torvalds has used AI to help with his projects, because it would be kinda silly not to.

Nothing? I mean an if/else works the same way, no matter if it's written by a human or an AI or a cat or whatever...

The Linux kernel developers are opinionated, though. Everything gets quite an amount of scrutiny. There will be several people having their eyes on submissions. They're looking for security vulnerabilities. They're adamant on maintainability. Have a standard on how to phrase things, indent lines... Send in the patches... They generally have high standards. I mean if someone submits some AI slop, there's a high chance it just gets declined and they're getting scolded for doing it.

There's of course always the chance someone tries to sneak something in. Or it creeps in on its own. But it's the same for bugs or security attacks. And maybe some of the devs work for companies who push AI and they'll do silly things. But the Linux community is pretty strong. They'll find a way to handle it. And maybe in the far future, AI will get as good as human programmers and there won't be an issue accepting AI code, because it has the same quality as human code. But that's science fiction as of now.

That will means all good developers are dead and world is on death spiral

Considering the level of oversight and quality control, probably not much. The only way the general public would know is that people would complain very loudly.

Do we know it doesn't?

The code reviews will get tougher.

Probably get forked.

It won’t.

The assumption here is that the AI-generated code wasn't reviewed and polished before submission. I've written stuff with AI and sometimes it does a fantastic job. Other times it generates code that's so bad it's a horror show.

Over time, it's getting a little bit better. Years from now it'll be at that 99% "good enough" threshold and no one will care that code was AI-generated anymore.

The key is that code is code: As long as someone is manually reviewing and testing it, you can save a great deal of time and produce good results. It's useful.