Maybe you'll agree with another post I made about how UX/processes already fail us here (without LLMs) and they should be improved: https://news.ycombinator.com/item?id=47324816

I think that's the only shot at progress since it can address the general problem instead of trying to special-case unenforceable rules that you hope the lowest quality people follow.

For example, a 3000+ line PR with no communication beforehand is already a low quality PR before AI. And it's one of the most annoying contributions to deal with since you have to basically tell them "sorry but all that work you did isn't acceptable". Yet they probably did all of it in earnest.

Presumably you already have a policy where you accept random PRs for small tweaks like doc fixes, but you don't want unsolicited PRs that make substantial changes. So a rule against AI doesn't change anything there.

And if you saw an uptick in large unsolicited PRs, then surely the solution is to update the process like disallow PRs that don't link to an issue.