Enterprise AI Measurement Guide
Agent Effectiveness
Outcome Success
What percentage of AI coding agent sessions in our engineering organization actually result in code that gets merged, and what's driving the sessions that don't?
What it shows
Outcome Success measures the share of AI coding agent sessions that produce a tangible, shipped output, specifically a merged pull request. It answers the question that adoption metrics cannot: of all the sessions engineers run with AI coding agents, how many of them result in code that actually makes it into the codebase? An Outcome Success rate of 10% means 9 out of 10 sessions produce nothing that ships.
Why it matters
Adoption metrics and session counts measure activity. Outcome Success measures results. An organization with 91% agent adoption and 10% Outcome Success is running many sessions and shipping from few of them, which means the cost per shipped outcome is dramatically higher than the cost per session would imply. For the CFO, this is the denominator problem: cost-per-seat and cost-per-session are the wrong denominators; cost-per-merged-PR is the right one. Outcome Success is what makes that calculation possible.
The Larridin angle
Larridin's Outcome Success metric is built on the connection between agent sessions and repository events, tracking each session through to its downstream result in the codebase, rather than treating the session as the end of the measurement chain.