I see this pattern constantly: developer prompts an LLM to build a feature, the LLM writes clean-looking code, the developer asks the same LLM to review it, and the LLM says it looks great. Of course it does. It wrote it. This is confirmation bias baked into your workflow. The model that generated the code has the same blind spots when reviewing it. It won't catch the assumptions it made during generation. Hostile Review exists because your code needs adversarial eyes — agents that assume the code is wrong and try to prove it. That's fundamentally different from asking your copilot "does this look ok?" Anyone else running into this pattern? How do you handle AI-generated code review in your team?