1. Developers that create the mess and don't have to deal with the consequences
2. Developers that fix the mess and have to deal with the consequences
I've noticed that the former category is significantly more pro AI than the latter
Gradually I moved to asking questions about the code instead, something like "if X and Y, will Z still hold? did we not forget to check this?" I realized that this is what I am doing in my head when looking at the code. And Claude understands well enough what I mean and checks it.
What I found mind-blowing though is that surprisingly often it says me something like "while looking this up for you I think found a potential bug, would you like me to quickly check it?" or "I noticed that actually when X and Y true, Z holds indeed, but I believe there is a rare situation (...) when we don't want Z because it makes zero sense, what do you think?"
and claude opens up a new emacs frame (aka "window" if you're not an emacs doc writer) with a magit diff buffer of whatever we've been working on. This happens instantly because the emacs server is already running since startup and this is just popping up a little client window
I mean, it's totally possible they just aren't doing anything complex.
That being said, for even the simplest stuff I do I benefit from looking at the code, making changes etc.