Assigning different agents to have different focuses has worked for me. Especially when you task a code reviewer agent with the goal of critically examining the code. The results will normally be much better than asking the coder agent who will assure you it's "fully tested and production ready"