My Checklist for Reviewing AI-Generated Code

Short summary

AI-generated code fails differently than human code: it's uniformly fluent, so surface-level review instincts misfire. The author presents a cost-ordered checklist for reviewing AI code, starting with hallucinated APIs (invented methods, wrong argument orders) that look correct but don't exist. The fix is mechanical—run type checkers and tests rather than relying on reading—because you cannot read your way to catching a hallucinated API.

• •AI code is uniformly fluent, making normal review instincts fail
• •Checklist ordered by cost of error, not frequency of defect
• •Hallucinated APIs are the most AI-specific and cheapest-to-catch defect via type checking

Generated with AI, which can make mistakes.