The number of bugs tends to be linear to lines of code written meaning fewer lines of code for the same functionality will have fewer bugs.

So I’m pretty skeptical that reviewing 2000 lines of code won’t take any more time than reviewing 200 lines of code.

Furthermore how do you know the AI generated lines are the open highway lines of code and not the mountain road ones? There might be hallucinations that pattern match as perfectly reasonable with a hard to spot flaw.