Harvard study: AI coding agents add 30% more code but review time jumps 49%

Harvard researchers find AI coding agents shifted the bottleneck to code review

Harvard study: AI coding agents add 30% more code but review time jumps 49%

A Harvard working paper by Ph.D. candidates Fiona Chen and James Stratton tracked 725,938 workers at 718 firms and found that AI coding agents raised lines of code per worker-month by about 30 percent, yet completed Jira issues stayed flat. The cost landed in review: pull request merge time rose 49 percent, formal change requests nearly doubled, and AI review tools barely dented the queue. The authors argue the bottleneck has moved to human sign-off.

After some outages that were caused by AI-generated commits, there were also new rules requiring senior engineer approval for AI-generated code, which created even more review bottlenecks.

More from this day

2026-11-17