Share with your CTO
GitHub is closing the visibility gap in its Copilot measurement stack. The expanded Copilot usage metrics API now attributes Copilot app activity to individual users, breaking out session counts, request counts, prompt counts, and token usage at the per-user level across enterprise and organization reports. Coding activity from the Copilot app also rolls into feature, model, and language breakdowns alongside IDE, chat, code review, and coding agent surfaces. Daily active user counts now include Copilot app-only users. The changes are backward compatible.
What this means for your business
Until this week, your Copilot app usage data told you the tool was being used but not by whom or to what effect. That’s not measurement, it’s attendance. Now you can see which engineers are adopting the Copilot app, what they’re generating, and how that output compares to what your IDE and coding agent users produce. That’s the difference between a license audit and an actual productivity signal.
The more interesting capability is the cross-surface comparison. When you can measure Copilot app code acceptance rates and lines added against IDE-generated code using identical fields, you get a controlled view of where autonomous app-level coding actually outperforms assisted in-editor coding. Most enterprise engineering orgs don’t have that comparison today. They have aggregate adoption numbers that obscure which surface is doing the real work and which is generating accepted-but-deleted noise.
The signal worth watching: GitHub is quietly building the measurement infrastructure that will let enterprises justify shifting budget from human-in-the-loop IDE assistance toward more autonomous coding workflows. Once your metrics show that Copilot app sessions generate code that gets accepted and retained at rates comparable to IDE completions, the argument for expanding autonomous surface access writes itself. That’s a procurement and staffing conversation, not just a tooling one.
Concept deep-dive: Token usage metrics
Token usage captures how much language model compute each interaction consumes, broken into prompt tokens (what goes in) and output tokens (what comes back), plus average tokens per request. It exists because model inference costs scale with token volume, not request count. A single Copilot app session generating a large scaffolded file costs far more than ten short autocomplete completions. Tracking this per user lets engineering leaders spot unusually expensive usage patterns before they surface as billing surprises, and benchmark the cost-per-line-of-code across different Copilot surfaces.
Based on reporting from GitHub Copilot app usage metrics now expand across report rollups, originally published 2026-07-28 19:35:00.
