I actually think that auto-completing pull requests with any merge strategy other than fast-forward is completely insane. WTF do you mean the final version of the code is produced with no humans present?

Humans, schumans. Just assign it to copilot, he'll take care of everything.

@donkulator I bet one of the few niches where LLMs will stick around after the hype dies down and everyone reckons with the cost of maintaining LLM-generated code is merging. It's not a terribly intellectually demanding task, it benefits from non-spatial thinking (you can look at previous nearby merges for precedent), it's massively quantity over quality since by that point your approach should be correct and you should have tests, and it demands no original thought.

The only hitch, but I'm sure that'll be fixed eventually, is that it's crucial for merge automation to be able to declare that it isn't sure how to resolve something, and an LLM in its current state will just bullshit something if a problem is beyond its complexity.

Squash. Who cares.