Speeding up code review with `small-change`
AI Summary
As engineering velocity at Imprint increased dramatically, code review bottlenecks grew proportionally, with average PR review time jumping from 33 to 77 minutes. To address this without sacrificing quality, the team introduced the "small-change" framework. By combining strict automated criteria (such as lines of code limits, AI code review bot verification, author experience thresholds, and service impact limits) with specialized AI review skills that exclusively approve or ignore without leaving line-by-line comments, Imprint bypasses traditional manual bottlenecks for routine changes while treating any resulting errors as systematic failures to be fixed via automated infrastructure improvements.
Key Takeaways
- Small-change criteria include being under 1,000 lines of code, passing AI code review with no p1/p2 findings, and requiring author experience of 10+ prior merged PRs in the targeted service.
- The automated review skill has only two actions: approve or ignore, deliberately avoiding nitpicks and "meat proxy review" bottlenecks.
- Systemic failures from small changes are addressed by improving CICD, automated checks, and harnesses rather than blaming individual engineers.
Why it matters
Scaling engineering velocity often breaks traditional human-centric code review loops. This approach offers a concrete model for safely automating routine code reviews using clear guardrails, metrics, and automated tools, shifting quality assurance from manual bottlenecking to systematic infrastructure enhancement.
Conversation 0
Be the first to share your perspective.