A 10-point checklist for reviewing AI-written pull requests

AI-generated pull requests look polished, with tidy code, sensible names, confident descriptions and passing tests, which can make them easier to approve without careful scrutiny. The underlying mistakes still exist, but they tend to appear in different places than in hastily written human changes. The recommended review starts by confirming the diff matches the original request and contains no unrequested refactors or extra files. Reviewers should also watch for removed validation, error handling or tests, verify that tests actually exercise the change, and check that new inputs are validated and new dependencies are real and necessary. Finally, the guide advises checking error handling, scanning for hard-coded secrets and sensitive logging, and confirming that the approved commit is the one that tests ran against.