How do you review a pull request that was largely written by AI?

Summary

The same way as any PR, with extra attention to what assistants get wrong: run the code and tests yourself, read every line, test edge cases the generated tests skip, verify every new dependency exists and is the intended package, check for injection, weak defaults and secrets, and make sure the author can explain the code.

Longer Answer

Generated tests often mirror the implementation and cover only the happy path, so they are weak evidence. Large AI-generated PRs should be split; reviewers can't meaningfully review 900 lines. The author owns the code regardless of who typed it.

Key Points

• Run, read, test edge cases.
• Distrust tests generated with the code.
• Check dependencies and security.
• Author must understand it; keep PRs small.

Deep dive: Reviewing AI-Written Code