Recently [Faith Ekstrand] announced on Mastodon that Mesa was updating its contributor guide. This follows a recent AI slop incident where someone submitted a massive patch to the Mesa project with the claim that this would improve performance ‘by a few percent’. The catch? The entire patch was generated by ChatGPT, with the submitter becoming somewhat irate when the very patient Mesa developers tried to explain that they’d happily look at the issue after the submitter had condensed the purported ‘improvement’ into a bite-sized patch.
The entire saga is summarized in a recent video by [Brodie Robertson] which highlights both how incredibly friendly the Mesa developers are, and how the use of ChatGPT and kin has made some people with zero programming skills apparently believe that they can now contribute code to OSS projects. Unsurprisingly, the Mesa developers were unable to disabuse this particular individual from that notion, but the diff to the Mesa contributor guide by [Timur Kristóf] should make abundantly clear that someone playing Telephone between a chatbot and OSS project developers is neither desirable nor helpful.
That said, [Brodie] also highlights a recent post by [Daniel Stenberg] of Curl fame, who thanked [Joshua Rogers] for contributing a massive list of potential issues that were found using ‘AI-assisted tools’, as detailed in this blog post by [Joshua]. An important point here is that these ‘AI tools’ are not LLM-based chatbots, but rather tweaked existing tools like static code analyzers with more smarts bolted on. They’re purpose-made tools that still require you to know what you’re doing, but they can be a real asset to a developer, and a heck of a lot more useful to a project like Curl than getting sent fake bug reports by a confabulating chatbot as has happened previously.
I personally believe AI code should not be included in anything…. but I can possibly understand if someone writes a function, then asks AI to optimize said function. Reads the generated code, fully understands what it has changed and why, verifies it works, checks potential new edge cases, then modifies their original code accordingly.
Sadly that is NOT what happens, if the AI code even works at all.
I agree with a twist : That sound like asking a monkey to write code and going a complete code review + test which means basically working twice.
Welcome to the world of increased productivity by AI
The junior programmer’s productivity is greatly improved by AI. But the senior programmer’s productivity is greatly impacted by the junior programmer’s use of AI.
For businesses, the question becomes: who is more effective in keeping my software maintainable (the ease of safely adding new features without regressing the quality of existing features) for the next few years.
Surely, that is going to be the senior.
But the senior is expensive, so their time is expensive too.
So what is a quick way to cut costs? To decrease the amount of impediments that the senior has, so that they can be more productive for the same amount of pay. And juniors using AI and impacting the productivity of the seniors, are a big impediment to the productivity of the seniors. So there you have a quick cost cut and can expect an increase in productivity too.
The other potential product of this scenario is fewer qualified senior devs in a few years. The talent pipeline is being disrupted in ways that are hard to predict.
And bean counting financial management who doesn’t understand technology at all understands absolutely none of this other than the Mantra they’ve heard in CIO and CFO magazine that says that using AI can improve their bottom line and raise their stock price
writing good code to start with doesn’t avoid review + test. i never use AI but i absolutely work twice
No, but writing the code means you (at least think you) understand the code, and can work with a reviewer, rather than pure AI code which pushes all the work onto the reviewer
sounds like i’m fooling myself, in this example :)
imo the whole point of review is to assume a priori that i’m fooling myself
Nothing to complain about. Now you have twice as many jobs created. First people are worried about losing their job, now are complaining about having more work. Which is the problem? Fact is finally people who have ideas can implement them. I have been vibe coding since the beginning of chatgpt. Probably the first day I used it. Mostly failure, it’s like gambling if you have no clue what is going on. However I’m getting better and have made some working software for some devices. Not annoying anyone or asking for help. There is no going back. People talking about how poor it is is true. Just wait a few years.
If I specify to ChatGPT that it write me a function to, say, generate a series of datetimes segmented by a timespan, it will almost always write a perfectly good function in a few seconds, thereby saving me several minutes and some debugging. I gladly include such functions in my code. Giving ChatGPT an open-ended programming assignment, by contrast, is almost always a bad idea.
In my experience, it has never been able to optimise a function. Maybe because it always jumps to writing code, but never asks me to profile it.
Finding bugs, once I have been able to narrow it down to a specific function often works.
Their simply mad because now people who didn’t waste 20 years learning to code can develop software as good as them. LLMs are here to stay and people critisizing them are just as stupid as those who opposed IntelliSense in early 2000s.
“they’re
Also: code should be readable, understandable and maintainable.
I deal with enough shitty code written by humans.
You clearly didn’t read the thread. I did, and, as a developer myself, I have to say that the MESA people has been extremely polite. The user who sent the patch is just a lawyer that, in his own words, knows zero about coding and refuses to learn. He just wants to send patches generated by, again his own words, “carefully constructed prompts” and let the true developers untangle them, discover what does work and what is just cosmetic, and cleanup the patches. And all this after the developers already went through that patch and discovered that half of the code did, in fact, make worse the performance because just duplicated calls already done.
Developing is much more than writing code.
Dude…
Don’t feed the obvious trolls…
But they’re not developing software as good as them…
They’re developing software faster than them.
But either you hire some seniors to clean up their code to make it not just work, but also production-ready and maintainable, or you just hire seniors to make software that works, is production-ready and maintainable.
In the first instance, you have to pay two people. In the second instance, you have to pay only one.
In the first instance, it will take extra time to fixup all the generated code from a new feature. In the second instance, it will take extra time to develop the new feature.
So time-wise, there is not really a win or loss between the two. But paying the salary of two people instead of one, is of course a big loss.
Unless of course you don’t care about production-readiness and maintainability of the code of your application.
Then I’d even say: don’t even hire a junior developer, hire your neighbor’s kid to do the job for some pocket money. Anyone can use an AI. Maybe you could ask your office manager to do your software development. Or the office cleaner. Or your mom. :P
https://arxiv.org/abs/2507.09089
People think they’re coding 20% faster. They’re actually coding 20% slower.
Arguments from people who don’t know the difference between “their” and -“they’re” somehow never impress me much.
i oppose intellisense today
*criticising
I just have to say that Joe Kim, once again, nailed the cover art. That is a really amazing title picture.
That’s an oldie but goodie!
This is simply a skill issue.
Using vibe coding capabilities efficiently, understanding the changes, and minimizing hallucinations/bugs on the first pass (not fixing them after they happened) requires knowledge and skills.
Plenty of experienced programmers are used to a certain kind of thinking, and find it difficult to adopt the new tools effectively.
This is not how to use LLMs people.
If the devs working with the code decide to use AI code review to try and catch anything they’ve missed or could improve that’s one thing but dropping the results wholesale as a bug report with no filtering or understanding of if it’s even valid is not cool.
If you enjoy prompt engineering then go work on a project more like the Animal Crossing LLM. I’m sure there would be a lot of good work that could be done in that area to develop characters and help integrate interactions with the larger gameworlds of various games. You don’t need to understand code to get a character behaving correctly.
Yes actually you do please stop trying to bash the game design community which spends 98% of its time elbow deep in code more complext than most people dream of then get bitched at because a butterfly didn’t flap a wing 1 time during an animation loop.
Don’t be daft, that’s not what I meant by behave correctly.
I meant the basic integration with whatever API the designers use is called correctly on the AI side and that they respond as they should. Basically something more in the writing track than programming and I doubt you think that every writer needs intimate knowledge of the game engine.
Also you defense of game developers is unnecessary and the example you give is overly extreme.
Nothing to complain about. Now you have twice as many jobs created. First people are worried about losing their job, now are complaining about having more work. Cleaning slop. Which is the problem? Fact is finally people who have ideas can implement them.
I have been vibe coding since the beginning of chatgpt. Probably the first day I used it. Mostly failure, it’s like gambling if you have no clue what is going on. However I’m getting better and have made some working software for some devices.
Not annoying anyone or asking for help. There is no going back. People talking about how poor it is, is true. Just wait a few years.
I’ve been a professional coder since 1986, not sure if I’m senior yet. I use AI a lot. It’s like when people started using stack overflow, or previously Experts Exchange. It’s a great tool as long as you understand every line that in generates. i like it, it does stuff that I haven’t thought of sometimes. I don’t think it saves me much time but I love honing the queries until it gets it right.
Isn’t that Locutus of Borg (illustration)? I thought so :]