Enterprise AI Measurement Guide
Team Performance
AI Output Share / AI PR Share
Is AI coding producing a proportionate share of our engineering team's most complex and valuable work, or is it handling the simpler tasks while humans carry the heavier PRs?
What it shows
AI Output Share measures the percentage of total engineering output value (complexity-adjusted) that came from AI-assisted work. AI PR Share measures the percentage of total merged pull requests that involved AI assistance. These two percentages can, and often do, diverge significantly. An engineering org with 66% AI PR Share and 56% AI Output Share is generating more AI-assisted PRs than it is AI-assisted value, meaning the average AI-assisted PR carries less complexity-adjusted weight than the average human-authored PR.
Why it matters
PR count is the most commonly cited measure of AI coding productivity. Output Share is more honest. If AI PR Share is growing faster than AI Output Share, the team is shipping more AI-assisted PRs but each PR is carrying less weight, which may indicate AI is being used for simpler tasks while engineers handle complex work themselves, or that AI-generated code is winning the volume race while losing the value race. The gap between the two percentages is the diagnostic.
The Larridin angle
This two-metric comparison is unique to Larridin's complexity-adjusted framework. Without CAV weighting, PR count and output would be the same number. The divergence between AI PR Share and AI Output Share is only visible when output is adjusted for complexity.