Measuring AI Adoption in Three Phases

▶ Watch (00:34)

GitHub’s Copilot metrics program has three phases: adoption and usage first, then proficiency and efficiency, then business ROI. Today’s dashboard sits in phase one. The refreshed Copilot Metrics dashboard and API launched in public preview at the event, with GA scheduled for February 2026. Enterprise admins can opt in now through Agent Control plane policies.

The sequence matters. Adoption data gives teams a concrete baseline instead of anecdotal feedback. Without it, claims about AI impact stay opinion. Phase three, the ROI layer, is where GitHub says it wants to take customers, but phase one has to come first.

Same Dashboard, Four Different Questions

▶ Watch (03:39)

Four roles read the same Copilot dashboard and ask different questions. A CTO at Mona Inc. looks at a 72% agent adoption rate as a strategic baseline. The VP of Engineering checks an 85% active user trend to confirm Copilot has moved past early-adopter enthusiasm into daily delivery. The CIO maps billing against lines of code accepted to justify spend. The engineering manager downloads the user-level report to find who needs coaching and who has become a power user.

The dashboard serves all four: aggregate views for executives, per-user API exports, and 28 days of rolling data for team-level decisions.

Reading the Live Dashboard

▶ Watch (07:01)

Shruti Corbett walked through the Copilot Metrics dashboard live demo (07:01), showing Mona Inc.’s acceptance rate at 25-30%, the range GitHub calls healthy. Claude Sonnet accounts for nearly 80% of model activity. GPT-4.1 and GPT-5 add about 15%, mostly on back-end Go work. Over 70% of usage spans languages outside the top four, confirming org-wide adoption.

“Accepted completions stay strong, even while suggestions fluctuate.” Shruti Corbett

For the CIO use case, the demo converted lines-of-code data from a Jupyter notebook into estimated hours and dollar savings.

“We’re freeing engineers to focus on what moves the business forward.” Shruti Corbett

From Usage Data to Business Impact

▶ Watch (21:06)

Lines-of-code metrics move from the API into the dashboard before end of 2025. Fine-grained role permissions and data residency support arrive in January 2026. The dashboard reaches GA in February 2026. After that, the data window extends beyond 28 days, non-IDE sources (Copilot Coding Agent, Code Review, next-edit suggestions) get their own metrics, and data warehouse connectors let teams pipe data into their own analytics stack.

The longer-term goal is SDLC-wide coverage: PR velocity, time to production, and a survival metric tracking how much AI-generated code stays in the codebase. That last metric is new ground.

Notable Quotes

Accepted completions stay strong, even while suggestions fluctuate. Shruti Corbett · ▶ 11:13

We’re freeing engineers to focus on what moves the business forward. Shruti Corbett · ▶ 15:08

And we understand what gets measured, gets better, and what gets better. Sharanya Doddapaneni · ▶ 24:24

Key Takeaways

  • The Copilot Metrics dashboard and API are in public preview now, with GA in February 2026.
  • A 25-30% code acceptance rate at Mona Inc. is the benchmark GitHub calls healthy and sustainable.
  • GitHub plans SDLC-wide metrics covering PR velocity, AI code survival rate, and data warehouse connectors.