Enterprise AI Measurement Guide
Team Performance
Output per Engineer
How do we measure whether AI coding tools are actually increasing the engineering output our team produces, rather than just increasing the number of commits or PRs?
What it shows
Output per Engineer is Larridin's primary engineering productivity metric, the weekly complexity-adjusted value each engineer produces, averaged across the team. It can be viewed as Raw Output (total), split between AI-assisted and human-only contributions, broken down by output category (features, bug fixes, KTLO), or drilled to individual performance. The weekly trend shows whether AI tool investment is translating into sustained productivity improvement over time.
Why it matters
Most AI productivity conversations rely on self-reported estimates or deploy counts, metrics that measure activity rather than output. Output per Engineer measures the complexity-adjusted value of what actually ships, which is what AI investment is supposed to improve. The AI vs Human split answers the question every CTO eventually faces: is AI-assisted output genuinely adding to total output, or is it replacing work the engineer would have done anyway?
The Larridin angle
Output per Engineer uses complexity adjustment, the Complexity-Adjusted Velocity (CAV) framework, which means a PR that ships a hard feature contributes more to the score than a PR that fixes a typo. This prevents the metric from being gamed by volume: a team that ships 200 trivial AI-generated PRs does not score better than a team that ships 50 hard PRs, which is the failure mode of raw PR count as a productivity metric.