Well, it was inevitable.

Yeah at this point trying to ban ai code is like trying to ban sunlight in the desert.

For 12 hours out of every day? :D

Doesn’t mean you shouldn’t try.

Maintainers are busy enough as is without having to take on the additional role of AI police

Doesn't it? Trying to ban sunlight in the desert means you will waste a ton of resources, fail, and look foolish in the process. Maybe you shouldn't do that.

Obviously I’m not talking about the metaphor.

It applies to the root too. It's a good metaphor.

It was a poor metaphor. You absolutely can ban AI software in software you control. If someone then managed to lie and sneak some in and gets found out, ban them and purge their contributions.

Ok. Then you have an extremely broken piece of software. Congrats. Destroying a community project for ideology seems counterproductive.

I guess every community project was extremely broken until now that slop can save us. What a ridiculous take.

I am very anti slop. But ripping out code that was good enough to pass review and was merged is just dumb. Ripping all the code a contributor contributed just because they where caught using AI is a recipe for destroying a code base.

Open source maintainers have enough trouble without trying to ban all AI generated code just to comply with a distributor.

AI code is inevitable. Even if you outright ban it, people will try to sneak it in. Lets say the change is literally just changing the order of calls to fix a bug - but the person who found and fixed the bug used AI to do so. Do you ban that? If it's not disclosed, how would you know? After all the commit is literally just a line order swap. What if the committer actually hand wrote that fix but used AI for the PR body because they don't speak English well and don't want to appear dumb?

The best practice is to allow AI-generated code as long as the person providing the commit can properly reason about its function - meaning they understand what the code does and have done their due diligence. Good quality AI generated code isn't the issue, it's the skiddies who think prompting Claude with the most basic queries and blindly accepting the output until the one thing they wanted to achieve works, results in usable code. It results in unmaintainable slop. You need an actual engineer with experience to review it, but maintainers don't have that capacity, so banning slop and requiring contributors to understand what they're trying to get merged is the best way forward.

There are serious legal concerns about what integration of AI code could mean for open source licenses as it’s already been decided that AI generated content cannot be copyrighted.

No there aren’t

Well good news then. Fuck copyright as it currently is. It needs a god damn massive overhaul.

This might actually get some work put towards defucking it.

Yeah, the powerful open source software political lobby will certainly ensure positive change is made.

The problem is never the tool. It’s the idiot using it.

Give an idiot a table saw and you will have a finger on the floor by end of shift. Give it to a master carpenter and you might still have a finger on the floor by end of shift but you will also have a well cut piece of wood either way.

Llms can cause harm, but it’s only idiots who only cause harm with them

Makes sense. AI can accelerate experienced coders. I’m ok to see people use AI for helping but NOT for Vibe Coding.

That’s like saying social media can make people more social, when we know that for most people it just leads to addiction.

Human brains are designed to find shortcuts, vibe coding is one of them. I’ve seen plenty of senior devs fall into this trap.

no it cant. every time someone does a study they find it slows people down and degrades quality. youre just parroting an ad basically.

I can only talk from personal experience, but it absolutely does speed up things. I know because I went from working away at one project to progressing many projects at the same time. Starting up new projects I never would have bothered to do in the first place just because the work would have been so tedious and I wouldn’t have enough time in the day. Mind you, I don’t vibe code things for public consumption. These are all for my personal use, but the fact that they exist at all is evidence contrary to your points.

(Important note: how I use AI in important software is very different from how I use AI for slop projects; the speedups are not as dramatic when you aren’t vibrcoding and you actually need good quality; I suspect most AI usage would actually be a slight downgrade for speed unless I delegated them strictly to review and menial implementation; but that actually assumes that I know the language I’m working in very well, which isn’t always the case for vibecoded projects: In those cases, the AI slop is probably equal to or better than what I would write by hand because I’m not as competent in elixir or javascript as I am in C or C++)

I feel like this is dishwasher discourse… Washing by hand is faster than the dishwasher by the wallclock; but you don’t actually need to rinse as much as you probably do, and you can walk away and do other stuff while the dishwasher is working. If you just sit there and stare at the dishwasher, yes, it’s slower, but that’s not how people should use it, and that’s what most of those studies are showing. Again, that level of sloppiness isn’t applicable for all use cases, but for where it is useful, it is very useful.

“Degrades quality” assumes developers are already writing good code to start. Unfortunately that just isn’t true. For people who know what they are doing and can properly guide it, it absolutely speeds up the menial tasks that plague software engineering. I’m not even a fan of the direction AI is going yet I absolutely see the value here.

For the AI haters, we really need to attack it from a better angle than “it doesn’t work” because it absolutely does.

My coworker made a tool(because where we work doesn’t allow plugins to it ide yet) that will load our code for context and we can write prompts against it. After we found the source of a big he used his tool and it found the problem in less than a min after asking why we were seeing the behavior in the application( tied up application pools)

That’s exactly what I’m talking about. It absolutely does help and works well for very specific tasks. That’s a perfect use case.

Listen, I’m a VERY healthy skeptic of AI on a number of fronts, but saying it can’t speed things up is just not true.

I feel like I understand where you’re trying to come from with the argument about skill degradation, but the flat statement you say here is undeniably false.

except its not. maybe a fat wad of code happens faster, but then you have to go read it and understand it and edit it and resolve all the stupid shit it did, and all of that takes longer than just writing it yourself in the first place. if all you care about is lines of code, sure yeah it speeds up the creation of garbage, but if you care at all about having good maintainable code at the end it absolutely does not. but what does what i think matter, nuance on the internet is dead.

I’ve run this experiment on myself a great number of times.

Your story is correct, sometimes. It’s not correct all of the time. Sometimes AI is slower and the output is better (think: pair programming). Sometimes AI gets me the first draft and the cognitive understanding takes time (overall, for me, usually faster, but sometimes slower I will readily admit). I’m really good at reading code, I guess? And still other times it will one shot near perfection (in true slot machine fashion). It’ll also produce absolute shit, as well, frequently.

My firm point to you is it’s not a blanket truth.

Boo

That’s okay, I guess. Disclosure is important.

i mean am i surprised? its just not worth it at this point