GitHub has updated its Copilot usage metrics API to break out activity from third-party agent apps, giving enterprises and organizations visibility into which agents their teams are actually using. Until this change, agent activity landed in a single undifferentiated bucket, making it impossible to tell whether Copilot coding-agent work was done by GitHub’s own agent or through partners like Claude and Codex. The new field, totals_by_3rd_party_agent, appears in the enterprise, organization, enterprise-user, and organization-user 1-day and 28-day reports GitHub Changelog.
Each entry in the new array reports agent_name, agent_id, user_initiated_interaction_count, and session_count. GitHub explicitly recommends grouping by agent_id because display names can change. Per-user entries omit session_count, and activity from agents that cannot be identified is omitted entirely. Multiple apps belonging to the same agent are collapsed into one entry, so the report reflects distinct agents, not individual integrations GitHub Docs.
The update arrives as GitHub continues expanding Copilot beyond single-account scenarios. In related coverage, GitHub has also launched an internal Copilot-powered data analytics agent that translates plain-language queries into analysis, and Copilot in Visual Studio Adds Agent Preview and Built-In Skills shows how agent capabilities are spreading across the developer toolchain. For teams rolling out more than one agent, the per-agent breakout in usage metrics provides a clearer picture of adoption across those surfaces.
Enterprise owners, billing managers, organization owners, and anyone with a custom role granting the View Copilot Metrics permission can access these reports, provided the Copilot usage metrics policy is enabled GitHub Docs. The change is backward compatible. Existing fields retain their shape, and reports omit totals_by_3rd_party_agent entirely when there is no recognized agent app activity for the reporting period. Administrators should note that the nested user_initiated_interaction_count counts agent app job starts and is distinct from the top-level field of the same name, which counts explicit prompts from other supported telemetry; the two should not be summed.
Without this breakout, teams could not answer basic questions about which agents were gaining traction or whether a newly introduced agent was supplementing or displacing an existing one. The per-agent metrics turn usage data from a coarse aggregate into a dataset that supports rollout and licensing decisions.
