# Unreviewed AI code anti-pattern
Filing pull requests containing AI-generated code you haven't personally verified is the most common anti-pattern in [[Agentic Engineering]]. It amounts to delegating your work to the code reviewer: they're the ones who now have to determine whether the code works, is correct, and meets quality standards. They could have used an agent themselves.
The bar is simple: the code works, and you are confident that it works. That means you've run it, tested it, and reviewed it. Not just checked that CI is green.
Mitigations:
- **Run tests locally** before opening the PR
- **Keep PRs small**: split large changes into multiple commits rather than one massive diff
- **Provide context**: link issues, explain goals, reference specifications
- **Show your work**: test notes, implementation comments, screenshots, or demo videos that demonstrate the feature works
- **Review generated PR descriptions**: agents write plausible-sounding descriptions that may not reflect what the code actually does. Validate them
This is a form of [[Cognitive debt]]: if you don't understand the code well enough to verify it, you've lost comprehension of your own system. The antidote is [[Agentic TDD]] (tests that objectively verify behavior) combined with actual code review.
## References
- https://simonwillison.net/guides/agentic-engineering-patterns/anti-patterns/
## Related
- [[Agentic Engineering]]
- [[Cognitive debt]]
- [[Agentic TDD]]
- [[Code is cheap, quality is not]]
- [[Technical debt]]
- [[Simon Willison]]