Multi agent QA splits finding bugs from fixing them, because an agent that implements its own findings rationalises them. Here is the workflow we run.