GitHub announced on September 11, 2026 that Copilot usage metrics are generally available for the dedicated VS Code Agents window. The update turns AI-agent adoption from a product idea into queryable operational data. Organizations can see whether teams are actually using agents and which agent types account for that activity.
Aggregate enterprise and organization reports add optional fields for daily_active_vscode_agent_users and totals_by_vscode_agent. The latter includes session_count and total_user_messages. Together, those fields answer more than how many active users exist: they add the number of sessions and the volume of interaction by VS Code agent.
Enterprise-user and organization-user reports add used_vscode_agent and per-user totals_by_vscode_agent. Those fields can show adoption distribution and usage depth. They remain telemetry, though, not direct measurements of code quality, delivery speed, or business outcomes.
GitHub defines the scope narrowly. The metrics cover the dedicated VS Code Agents window, not Agent Mode in the editor window and not generic Copilot rollups. The fields are optional, and historical data may be absent or null. Consumers should not interpret missing values as zero or add figures from different interfaces without checking the definitions.
Access is available to enterprise owners, billing managers, organization owners, and custom roles with View Copilot Metrics, subject to the relevant policy being enabled. Teams connecting the data to an internal dashboard should confirm the report level, role permissions, field version, and export schedule so that a permission or schema change does not silently create a false trend.
Adoption numbers alone cannot prove that an agent works. More sessions might mean more tasks or more retries. More messages might mean productive collaboration or unclear prompts and unreliable output. A stronger evaluation combines usage data with pull request outcomes, sampled reviews, task time, cost, defects, and human intervention, with an explicit denominator for each metric.
Governance also needs an answer for who can see user-level data, how long it is retained, which uses are approved, and how management analysis is separated from employee surveillance. Aggregation reduces unnecessary individual tracking, but organizations still need internal definitions, retention rules, and access auditing. One daily-active-user number should never be presented as proof of productivity improvement.
The significance of the update is that an agent rollout now needs instrumentation like any other production system. Mature measurement should cover adoption, workflow outcomes, quality, and risk, with comparable baselines when the agent, model, or permission set changes. GitHub supplies useful measurement primitives; turning them into credible engineering and management decisions remains an organization-specific design task.
