The AI Code Review Bottleneck: Why Our Merge Time Tripled
The pull request was 1,140 lines. The description was four bullet points. It had been opened eleven minutes after the ticket was assigned. I looked at it for forty minutes, approved it, and merged something I did not fully understand. That was the week I started keeping numbers, and the numbers showed an AI code review bottleneck sitting in the middle of my team like a parked truck. We were…
The AI code review bottleneck refers to the gap between how quickly code can be generated and how slowly it can be approved. As AI-generated code became faster and cheaper to produce, human attention remained unchanged, leading to longer queues and slower merges. This bottleneck was exacerbated by larger PR sizes and a lack of effective review strategies.
The bottleneck arose due to three main factors: a doubling of PR arrival rates, the increasing size of PRs, and superlinear review effort with growing diff size. AI-generated code proved particularly challenging to review due to its uniform plausibility, lack of intent, and redundant code. Responding to these issues, the author implemented a series of measures, including a hard diff-size cap, a "what I actually verified" section in PR templates, and an "explain-back" rule requiring reviewers to understand the code in one sentence.
These changes significantly improved review efficiency and reduced the number of PRs stuck in the queue.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
