TL;DR
- AI made refactoring cheap, so I stopped flagging minor issues in code reviews.
- Review back-and-forth with our offshore team dropped from about 10 rounds to 4.
- Every comment is tagged Critical, High, Medium, or Low. Only Critical and High must be fixed.
The problem: too many review rounds
I'm a PM, and I also review code. Our development team is offshore, and review threads were getting far too long.
So I stopped flagging small things. If something needs fixing later, AI can fix it quickly.
What I no longer flag
- Personal preferences in variable and function names
- How finely functions are split
- Small duplicated code that could be shared
Minor style improvements with little business impact are optional. Delivery speed comes first.
Test code is fine as long as it covers what needs to be tested. We generally don't write TODO comments either.
These exclusions are written into the review criteria I give Claude.
What I still always flag
- Anything that affects testing effort
- Deviations from existing conventions without a reason
- Security issues
- Code whose intent isn't documented
Intent goes in a code comment or the PR description.
Four severity levels
| Severity | Action |
|---|---|
| Critical | Must fix |
| High | Must fix |
| Medium | Optional |
| Low | Optional |
Until Critical and High issues are fixed, the code can't even go to the test environment. I don't want engineers spending time on Medium and Low. The levels also make fix priorities clear.
The result
Review rounds dropped from about 10 to 4. In practice, leaving the skipped items unfixed turned out to be fine.
Takeaway
AI has lowered the cost of refactoring, so you can let small things go in code review. Focus on four things: testing effort, existing conventions, security, and intent.
Top comments (0)