The News, Explained
On October 22, 2026, GitHub added industry comparisons to the Copilot impact dashboard. Copilot is an AI tool for development work, and this dashboard connects organizational usage with code-change output. The new feature uses aggregated results from participating businesses as a benchmark: a reference for comparison. Participation is off by default. Source
Eligible participants are enterprise accounts and organizations outside an enterprise. Enterprise-owned organizations participate through their enterprise; enterprises used only to assign Copilot Business licenses cannot participate. Enterprise or independent-organization owners enable participation, while people with dashboard viewing access can see results. The existing Copilot usage metrics policy must also be enabled. Source Source
Owners can opt in through the widget or central policy controls. Both change the same Copilot metrics industry comparison setting. Industry and company size are both required. The documented enterprise dashboard route is Insights, then Copilot impact, where the comparison widget appears. Source
The selected 28-day reporting period shows agentic adoption and leverage. An agent is an AI capability that executes multiple stages of a task. Adoption is the share of active users GitHub classifies as agent-first or multi-agent. Leverage compares their pull request merge rate per user per month with that of nonagentic users. A pull request proposes code changes for review; merging incorporates accepted changes into the project. Source Source
These phases depend on activity surfaces and qualifying days. At least two active days in the trailing 28 days are required. One GitHub-based agent surface qualifies as agent-first; two or more surfaces, or the GitHub Copilot app, qualify as multi-agent. Examples include cloud agent, code review and the command-line tool Copilot CLI. Requested and automatically assigned code reviews count as the same surface. Classification is recalculated daily. Source
If the selected industry has too few participants, the comparison expands to all industries at the same company size, with the actual cohort labeled. If that broader group is also insufficient, no ranking appears. Median comparisons appear first; a top-decile reference is added as participation grows. The median describes the middle of a distribution, while the top decile marks the upper 10% level. GitHub gives no numerical participant threshold in this announcement. Source Source
Rankings are also absent when an organization’s own score for a metric cannot be calculated. Examples include no active users or every active user being agentic, leaving no nonagentic baseline for leverage. Reading a rank therefore requires checking both the displayed peer cohort and the existence of a usable comparison group within the organization. Source
Comparisons refresh daily. Opting out removes the display immediately and excludes data from peer cohorts at the next daily update. GitHub says previously calculated aggregates contain no identifying organizational information and are deleted after 30 days. Enterprise Cloud with data residency compares participants only within the same region. Industry comparisons are available in the dashboard, outside the usage metrics API, NDJSON exports and downloads. Source Source
OYOPICK’s Take
We think recording the actual peer cohort alongside a ranking will determine how useful this feature becomes. Selecting one industry can still yield an all-industry comparison, and an organization where everyone uses agents can lose its nonagentic baseline. Reports that retain company size, actual industry scope and the 28-day period could reduce comparisons between scores produced under different conditions.
A difference in pull request merge rates can be a starting point for asking about adoption. If task size or review practices differ within an organization, however, output differences between its user groups do not by themselves establish a causal effect of AI adoption. We would welcome using the comparison to examine which work is adopting these tools and how code quality and review workload change alongside it.