One email each morning — yesterday's AI, sortedGet it in your inbox

METAL LAB

Copilot API now breaks down usage by individual agent

The GitHub Copilot usage metrics API has added the ability to track agent app activity per individual agent. The change applies to both daily and 28-day reports at the enterprise and organization levels.

이미지: 깃허브 코파일럿

Summary

  • The GitHub Copilot usage metrics API now breaks down third-party agent app activity by individual agent instead of lumping it together.
  • The new `totals_by_3rd_party_agent` field includes the agent name, a stable identifier, the number of user-initiated interactions, and session counts.
  • The change is backward compatible, and the field is simply omitted from reports for periods with no recognized agent activity.

Agent activity is no longer lumped into one bucket

On August 7, 2026, the GitHub Copilot usage metrics API released a feature that reports third-party agent app activity broken down by individual agent. Previously, activity from partner agents such as Claude and Codex was aggregated into a single bucket, leaving no way to distinguish how much each agent was actually being used. As teams began adopting two or more agents, this limitation became a real problem, making it difficult to compare adoption rates or make licensing decisions.

The update applies across daily and 28-day reports at the enterprise, organization, enterprise-user, and organization-user levels. Being able to track activity per agent independently now makes it possible to compare the adoption of a newly introduced agent against existing agents using actual figures.

Structure and usage of the new field

The new optional array added to the response, totals_by_3rd_party_agent, contains one entry per recognized agent app. Each entry includes agent_name (the display name), agent_id (a stable identifier), user_initiated_interaction_count (the number of agent app job runs initiated by users), and session_count (the number of agent app sessions).

GitHub notes that because display names can change, agent_id — not agent_name — should be used as the key when linking data across reporting periods. session_count is only included in aggregated enterprise and organization reports; it is absent from per-user entries.

A number that's easy to confuse

There is one important caveat. The newly added nested field user_initiated_interaction_count counts the number of agent app job starts, and is a completely different value from the existing top-level field of the same name. That existing top-level field aggregates explicit prompts from a different telemetry source. The two values should not be added together or treated as equivalent.

Additionally, multiple apps belonging to the same agent are combined into a single entry, and activity from unidentifiable agents is excluded from the report. This metric is only available to enterprise owners, billing managers, organization owners, and holders of custom roles granted "View Copilot metrics" permission. The Copilot usage metrics policy must be enabled to use it.

Backward compatibility and current limitations

The change is backward compatible. Existing fields retain their original shape, and the totals_by_3rd_party_agent field itself is simply omitted from a report if there was no recognized agent app activity during that period. Existing integration code does not need to be modified and will continue to function.

However, since activity from unidentified agents is excluded from the aggregation, there is no guarantee that all agent usage is fully captured. Detailed criteria for which agents are included as "recognized" have not been disclosed in the source.