AI agents are rats in a maze: they reach for what they know. And unless you show them better, what they know is slop.
MOST software is garbage, and they'll happily imitate it.
You can't just let AI run loose if you give a damn about the craft. So we babysit. I constantly yell at them to stop writing terrible code. It works, but takes forever, and had become a big bottleneck in my work.
For code to be maintainable, it needs to be simple, boring, even "dumb". You shouldn't need a PhD to read production software.
So i built stupify - it reviews my agents against code I actually respect, and yells so i don't have to.
The quality of review is anecdotally much better than BugBot and CodeRabbit, for pennies on the dollar.
Give it a shot, curious what HN thinks.