Yep pretty much same, although if I’m lax at any point of the reviewing (in-progress or final), I’d say the quality quickly drops to below my average manual eff

Yep pretty much same, although if I’m lax at any point of the reviewing (in-progress or final), I’d say the quality quickly drops to below my average manual effort, and then I don’t even have the benefit of thinking it all through as directly. I think getting really quality results out of LLM code generation for non-trivial projects still needs quite a bit of discipline and work.