There’s a hidden tax on every AI-generated merge request
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
AI coding tools are shifting bottlenecks from code generation to code review, overloading senior engineers with review queues that grow faster than review capacity. DORA 2025 data shows key delivery metrics haven't improved with AI adoption. The real cost is attention fragmentation: senior engineers spend more time reviewing and less time on design and architecture. AI-assisted review can help when teams already have solid CI, clear ownership, and written review standards — enabling pre-triage, risk routing, and pattern matching. The recommended approach is workflow redesign: risk-tiered review paths, WIP limits for reviewers, author-provided risk declarations, and measuring reviewer load rather than MR volume. The key leadership signal is whether senior engineers are regaining design time, not whether MR counts are rising.
Table of contents
The review queue became the sprint planPassing CI does not mean it is cheap to reviewGeneration scales, judgment doesn’tWhen AI review actually closed the gapWhere the tax compounds: seams, exceptions, and risk ownershipMeasuring the constraint instead of the outputAdding structure through workflow designThe decision rule for expanding AI usageSort: