Linear B data shows triple delivery rates for AI users
Linear B found that heavy AI adopters achieved nearly triple the delivery rates of other cohorts after a capability inflection point. The company argues traditional metrics fail to capture this impact due to human review bottlenecks.
Linear B’s analysis of industry data, covering millions of pull requests since January, identified a significant inflection point in December of the previous year when AI models became sufficiently capable. At this point, early adopters who used AI the most achieved delivery rates more than double, and almost triple, those of other cohorts. Ofer Afias, from Linear B, argues that measuring AI impact requires differentiating between user cohorts and tracking new key performance indicators like merge frequency.
The company noted that traditional DORA metrics, such as cycle time, failed to capture the full impact because AI accelerates coding but does not eliminate human bottlenecks like code review. Consequently, Linear B shifted its measurement strategy to track merge frequency, which spiked among heavy AI users. This data supported the internal transition to spec-driven development by demonstrating that high-skilled developers compensated for faster coding by handling more concurrent deliveries.
Afias contends that traditional end-to-end delivery metrics are capped by human review processes, necessitating a new approach to evaluating productivity. The findings suggest that while AI tools significantly accelerate code generation, the overall benefit is realized only when development workflows are adapted to manage the increased volume of concurrent deliveries.