Share with your CTO
GitHub is pushing Copilot measurement past vanity metrics. The new total pull requests merged field in Copilot’s usage metrics API lets enterprise and org admins see not just how productive individual developers are within each AI adoption phase, but how much of the organization’s total delivery throughput each phase actually produces. Previously, the API reported per-user averages only. Now totals are available in both 1-day and 28-day reports, giving engineering leaders a population-level view of where AI adoption is actually moving code to production.
What this means for your business
The average-versus-total distinction sounds like a minor API detail. It isn’t. Per-user averages can look strong even when the high-adoption cohort is tiny, meaning a handful of power users flatter your AI ROI story while the bulk of the engineering org contributes little. Totals expose that gap immediately. If 80 percent of your merged pull requests still come from developers in the earliest adoption phase, no amount of per-user productivity gains at the edges changes your actual delivery velocity.
The deeper analytical move here is that GitHub is building the instrumentation layer for what you might call adoption-weighted throughput: the ability to correlate organizational output not just with headcount or tool licenses, but with where developers actually sit on the AI adoption curve. That reframes the CTO’s job from “did we buy Copilot” to “are we systematically moving developers through adoption phases fast enough to compound the throughput gains.” The measurement and the management objective are now the same variable.
The signal worth watching: how quickly enterprises start gating engineering team health reviews on phase distribution rather than raw license utilization. License counts are a procurement metric. Phase-weighted throughput is an engineering strategy metric. The two are not the same, and GitHub just made it harder to confuse them. I’d revise this view if adoption phase classification turns out to be too coarse to reflect real skill variation within large engineering orgs, which is a real risk in shops with heterogeneous codebases.
Concept deep-dive: AI adoption phase cohorts
GitHub segments Copilot users into cohorts based on how deeply and consistently they use AI assistance in their workflow, from occasional accepters of suggestions to developers who rely on AI for the majority of their code authoring. The cohorts exist because aggregate utilization data obscures the difference between broad shallow usage and narrow deep usage. Think of it like distinguishing between a gym where everyone shows up once a month versus one where a core group trains daily. For engineering leaders, the cohort breakdown determines whether AI investment is building a durable productivity baseline or just inflating headline adoption numbers.
Based on reporting from Track total merges by adoption phase in enterprise and organization reports, originally published 2026-06-26 16:55:00.
