AI coding agents generate more code, but not more software
Study finds coding efficiency gains get “absorbed” by human review “bottleneck.”
TL;DR
- AI coding agents increase total lines of code generated by 30% and commits by 20%.
- Despite increased code generation, software output and the resolution rate of issues and epics do not significantly change.
- The code review process becomes a bottleneck, with review times increasing by 49% on average.
- Pull requests submitted after AI agent introduction are nearly twice as likely to require revisions, with 35% more comments.
- The share of workers performing code reviews increased by 14% after AI agents were introduced.
- AI has had a marginal impact on the code review process itself, with humans still responsible for the vast majority of reviews and comments.