I "don't" use AI, I mean like most devs nowadays, I have a Copilot suggesting "smart" autocomplete, that's all. I wouldn't let an AI pilot my codebase. If you have a bit of probability knowledge, you may be aware of what happens when you're trying to get specific outputs from generic inputs.

And if you don't have a bit of probability knowledge: generic input just means high variance in the output, so "it looks right" isn't the same as "it is right." In other words, you'll just get a noisy sample out of many possible ones... and not what you really want.

I know you're being sarcastic but I think some people may find this interesting:

https://minimal-agent.com/

I made one based on this tutorial (about 50 lines), which is about 8x faster and cheaper for small tasks in my experience. I highly recommend the learning experience.