AI Coding Agents Quadruple Output but Stretch Code Review Time 441.5%
Updated
Updated · O'Reilly Media · Jun 26
AI Coding Agents Quadruple Output but Stretch Code Review Time 441.5%
3 articles · Updated · O'Reilly Media · Jun 26
Summary
Faros AI data on 22,000 developers across 4,000 teams found higher AI adoption lifted throughput but drove median review time up 441.5%, code churn up 861% and zero-review merges up 31.3%.
GitClear’s 2025 analysis put the trade-off starkly: daily AI users generated about 4x more code for only roughly 12% more delivered value, leaving human verification as the new bottleneck.
CodeRabbit’s study of 470 open-source PRs found AI-coauthored changes carried about 1.7x more issues, with logic problems up 75%, security issues 1.5 to 2x higher and readability problems more than tripling.
The report argues teams should shift from reviewing every change equally to risk-tiered review—small PRs, strict CI, required test evidence and multiple AI reviewers—while keeping humans on high-blast-radius merges.
For software teams, the implication is that AI has made code generation cheap, not understanding: review capacity, not coding speed, now determines whether productivity gains turn into reliable software.