Static analysis as the reviewer
You will not read every line an agent produces. Pretending otherwise is how half-baked software ships.
Code analysis is the part of review that can run every time, on every diff, without getting tired:
- linters and formatters so style is not a debate
- complexity and duplication checks so the agent does not invent a fourth pattern
- security scanners for the boring, dangerous mistakes
- architecture tests that fail when a layer is skipped
The agent can write the code. The pipeline decides whether it is allowed to stay.
This also answers a quieter problem: if nobody on the team can still judge the output, analysis is the last honest signal you have. Raise the bar in tools, not in a pull-request comment that nobody has time to write.
If CI is green and the change still feels wrong, that is a missing rule — add it. Do not rely on remembering next time.
