The interesting part of agent adoption is rarely the demo.
It is the map after the demo.
GitHub’s July 17 Copilot changelog says the usage metrics REST API now reports repository-level activity. Enterprise and organization reports can show daily per-repository pull request activity for Copilot coding agent and Copilot code review, including pull requests created, merged, reviewed, and suggestion counts by comment type.
That sounds like dashboard plumbing. It is more important than that.
Once agents start touching real code, team-level adoption metrics are too blurry. A company does not only need to know that Copilot is being used. It needs to know which repositories are actually changing because of agent work, where reviews are happening, where merges are happening, and where the tool is mostly decoration.
AI readiness is local
A healthy repository and a messy repository experience AI very differently.
One codebase has tests, ownership, small pull requests, consistent review habits, and clear boundaries. Another has unclear architecture, weak tests, stale docs, and risky deployment habits. The same agent can look brilliant in the first environment and reckless in the second.
Repository-level metrics make that visible.
This is the bridge from agent adoption to AI readiness. If a repository gets frequent agent-created pull requests that merge cleanly, that tells one story. If another repository generates many reviews but few useful fixes, that tells another. If a third repository never sees successful agent work, the problem may not be the model. It may be the codebase, the tests, the ownership model, or the workflow around it.
That is the useful frame.
AI does not land on an abstract company. It lands on specific repositories, specific teams, specific conventions, and specific failure modes.
The next serious enterprise AI reports will not stop at seats, prompts, and token counts. They will ask where the work moved, where quality improved, where humans still had to rescue the agent, and which repositories need cleanup before more autonomy makes sense.
The agent era needs observability at the level where software actually breaks.
GitHub is moving the meter closer to that level.
Source: GitHub Changelog