How to safely integrate AI-generated code: a review process that works

A team shares their process for reviewing AI-generated code after adopting tools that write a third of their diffs. They enforce human ownership of every merged line and require automated checks before human review. Key failure modes include hallucinated APIs and security anti-patterns, which they address with targeted checklists.