You might call me a luddite for this article, but I actually have used AI to do some (not so very vibe though) coding. But I always check and double-check what the clanker has done before I use it in production. If it doesn’t look like it will run satisfactory, it probably won’t.

That may be an approach that hinders the full effect of using AI for your code, but it at least makes me feel like I’m somewhat in control.

However, there’s a trend to just let Claude (btw. a hyped up piece of code itself) “co-author” your more or less dormant open source project. First it was Tridge giving Claude more or less free access to rsync (which in some cases broke horribly in the process, AI does seem to have that going “for” it), now it’s Dustin Kirkland doing it to byobu (a tmux wrapper, because basic tmux is… cumbersome).

I mean, I do get why you may want an AI to take at least some of the workload of maintaining an open source package, but the results in my honest (and maybe hurtful) opinion speak for themself - clankers produce sub-par code and break working stuff. If you don’t have time to maintain your project, you don’t have time to watch an AI do it either.

So, what I do instead is letting AI do its thing totally separate from the main project, let it play around in a branch maybe. Then transfer the working parts manually to main development, test and test again (and not with the tests AI provides, they tend to want to please you by creating working tests, so in many cases you end up with placebo tests that always succeed even if the program is broken beyond repair). Then maybe you learn a new way to look at a problem, but normally you get much better code out of vetting the AI rigorously.

Don’t do what many like to do, just give Claude full access to your git repo and let it push its abysmal hallucinations into production. Remember, the main goal of the AI is not the same as your main goal, you want to enhance or fix your program, the AI just wants to see you happy, and it will happily tell your whatever is necessary to make you happy.

Also see AI is not lying about the most common misconception about AI.