I think one of the things that will need to be embraced is carefully curating .md context files to give the prompts better/shared direction to contributors. Things like any new feature or fix should include a test case (in the right place), functions should re-use existing library code wherever possible, function signatures should never change in a backwards-incompatible way, any code changes should pass the linter, etc etc. And ideally ensure the agent writes code that's going to be to the maintainer's "taste".

I haven't worked out how to do this for my own projects.

Once you've set it up it's not too hard to imagine an AI giving an initial PR assessment... to discard the worst AI slop, offer some stylistic feedback, or suggest performance concerns.