This is longer than I've worked with copilot, but the manual fix versus prompt time feels pretty right on to me. I think that there's a sweet spot where continuing to prompt will take longer to get to the expected result than coding by hand from that point.

Being able to gauge when to dump the autopilot and take the stick is probably the development skill we all need to learn with LLMs.