GitHub lets companies rank their Copilot agent use against industry peers

GitHub, owned by Microsoft, has added an “Industry comparison” section to the Copilot impact dashboard, according to an October 22 changelog post. Enterprises can now measure their agentic Copilot adoption against aggregated peers, but only if they opt in.

Peer benchmarks arrive with strict opt-in rules

The new section covers two measures. The first is the percentage of a company’s users who are agent-first or multi-agent. The second is how the pull request merge rate of agentic users compares with that of nonagentic users. Once enough businesses have joined, a score is placed relative to the typical peer over the 28-day reporting window, and as the peer group grows, relative to the best-performing tenth.

Users pick a company size band (the examples given are 1-999, 1,000-4,999 and 5,000+ employees) and one of ten industries, and can switch between their industry and all industries. If the chosen industry group is too small, the dashboard widens the comparison to all industries for that size and labels the group it used.

Participation is disabled by default. It is controlled by a “Copilot metrics industry comparison policy”, which can be switched on from the dashboard or from enterprise or organization Copilot policies. Both places share one setting, so turning it off in either stops a company from contributing data and from seeing comparisons.

Who can use it is narrower than the headline suggests. Enterprise owners, and owners of organizations not owned by an enterprise, can opt in and set the peer group. Organizations inside an enterprise participate through that enterprise, and enterprises used only to assign Copilot Business licenses cannot take part. The Copilot usage metrics policy must be enabled. Everyone with access to the dashboard can view the results.

GitHub warns that early rankings may be thin. Soon after launch, users might see only the median, a broader peer group than they selected, or no ranking at all. A ranking is also withheld when a company’s own score cannot be calculated, for instance when every user is already agent-first or multi-agent, which leaves no nonagentic baseline.

What the data handling says about privacy

Feeding usage data into a pool that competitors can compare against raises obvious questions, and the post answers some of them. Comparisons refresh daily. Opting out removes the comparison view immediately, and the company’s data is excluded from the next daily update. GitHub says the pooled benchmark figures carry nothing that identifies a company and are erased after 30 days.

Customers on the data-residency flavor of GitHub Enterprise Cloud see peers from their own region only, and figures from different regions are never pooled. That matters for regulated customers who keep data in a specific jurisdiction. The comparisons appear only in the dashboard and are not in the Copilot usage metrics API, NDJSON exports or downloads, so a company cannot pull peer figures into its own reporting systems.

The post does not say how many companies must opt in before a peer group is shown, how the ten industries are defined, or whether GitHub applies any minimum group size to protect small participants. Those are the details a compliance team would want before switching it on, and the claims about non-identifying data are GitHub’s own and not independently verified.

The categories come from a dashboard only three months old

The agent-first and multi-agent labels are not new. GitHub’s July 2026 impact dashboard introduced cohort cards for code-first, agent-first and multi-agent users, plus a passive segment, with average pull requests merged per user per month and median merge velocity for each. The classification matches the one in the Copilot usage metrics API, based on a rolling 28-day window. In the API, added in May 2026, the multi-agent phase means a user engaged with two or more GitHub-based agent surfaces or with the Copilot app.

So the October update adds an outside reference point to numbers companies could already see internally. Until now, an enterprise could tell that, say, a given share of its developers were agent-first, but not whether that was ahead of or behind similar firms. That is the gap GitHub says the feature fills, by helping leaders set realistic goals and communicate progress.

There is a caveat in the metric itself. Merge rate of agentic versus nonagentic users is a throughput measure, and the source does not claim it reflects code quality or business results. It also depends on who opts in: a peer group made only of enthusiastic adopters would make an average company look like a laggard. GitHub’s own note that early rankings may be absent or broad suggests the benchmarks will be most useful once participation grows.

GitHub has also been giving administrators more control elsewhere in Copilot. Last week it added a way for admins to script which AI models each team gets. The industry comparison follows the same pattern of tooling aimed at enterprise owners rather than individual developers. Documentation for setup is on the Copilot impact dashboard docs page linked from the announcement.