AI coding agents increase code output but fail to boost software delivery

Recent analyses of AI adoption in software development reveal a discrepancy between increased coding activity and actual project completion. Data covering approximately 725,938 workers across 718 firms shows that while AI coding agents increased lines of code by 30%, commits by 20%, and pull requests by 23%, there was no statistically significant rise in resolved Jira issues or epics.

This gap suggests that increased output may be creating bottlenecks elsewhere. Specifically, code review demands have risen, with average pull-request review times increasing by 49% and the share of pull requests requesting changes nearly doubling.

Telemetry from 10,000 developers indicates that while teams with high AI adoption merged 98% more pull requests, their overall delivery metrics remained stagnant. Furthermore, a controlled trial by METR found that experienced open-source developers actually took 19% longer to finish tasks when using AI tools, despite their own perception of being 20% faster. Experts suggest that activity metrics like pull request counts can be easily inflated by AI, whereas quality and delivery metrics are more reliable indicators of true productivity gains.