When it consistently and regularly catches bugs that humans and other review agents miss, the benefit is quite clear and it becomes obvious there are more defects in your code than you realize, might be much more than that 1:90 rule of thumb. It's often less bugs and more about edge case handling, we all skimp on those exhaustive cases because it is exhausting while the agents never tire and are more than happy to work through them.

It's comment wall is uncovering enough that we are going to brave the burnout fodder while we work on a second agent to make the adjustments after we leave our own comments on the clanker comments. It's weird writing that sentence, I've generally been against the hype train, but they have become good enough there are real wins on the code review side. So yes, I will maintain you are doing a disservice to yourself, your code, and your users by not using a thing that catches bugs before you publish them. I still care about the quality when I'm putting it in other people's hands. Ai is now clearly facilitating improved code quality - when used on the review side.