We're going to the same direction in my company, AI produces and AI reviews it which I feel is kind of like shaking your own hand.
For now there's still human reviews but soon it will be all automated. I estimate that the AI review (latest Opus) is around 15% to 20% accurate
I'm still holding down keeping a human in the mix where I'm at now but it's a tough battle. Other problem with AI review is it will try to find issues that are not issues and you can get in this loop of an agent fixing something, triggering another review, fixing those findings and basically going in a circle. Has happened multiple times now which ends up with a massively overbuilt task that would have taken a human a few minutes to just do to begin with.