AI code output has tripled in volume and quality at the delivery level, but production incidents and bugs are accelerating faster—revealing a system overwhelmed by the speed of AI-generated code.

What happened: AI has become the primary code author in 80% of teams studied, with acceptance rates rising from 20% to 60%. Epics completed per developer are up 66%, task throughput per developer is up 33.7%, and PR merge rate per developer is up 16.2%. However, code churn has increased 861%, incidents-to-PR ratio is up 242.7%, bugs per developer are up 54%, and pull requests merged without any review are up 31.3%.

Why it matters: The productivity gains are real at the business level, but they mask growing reliability and quality costs downstream. For every PR merged, production incidents are now occurring at more than three times the prior rate. Senior engineers are spending 156.6% more time on first review and 441.5% longer in review overall—reviewing superficially convincing code that contains structural and logical failures. The data suggests that organizations with strong engineering practices are experiencing the same downstream deterioration as everyone else.

What to watch: Code churn of 861% warrants investigation—whether it reflects rework of insufficient AI-generated code or productive refactoring. Monthly incidents are up 57.9%. The report cautions that organizations cutting engineering headcount on the basis of AI output gains should examine this data, which shows output is up but also reveals the hidden costs accumulating at every stage downstream.