A fresh review still needs judgment

If you ask an agent to find something wrong, it will find something wrong. That doesn't mean the problem is real.

I still find adversarial review with a fresh context window useful. Asking the same agent to keep checking its work can reinforce the assumptions already in its context. A separate review gives it another starting point.

You still need enough understanding of the work to judge what comes back.

Originally posted on X. Lightly edited for reading here.