Yeah, like the rest of LLM output, I’d say 80% of an average code review is worth considering - and if that includes anything you might have otherwise missed, that’s a win compared to learning about the problem post-launch.