GitHub Copilot code review and cloud agent metrics

Can GitHub Copilot code review and cloud agent be measured separately in usage metrics

Yes. GitHub’s usage metrics separate Copilot cloud agent from Copilot code review at the metric level, so you can measure adoption and activity for each surface on its own. The part people miss is that billing and usage are not the same thing, so one shared credit pool does not mean one shared metric.

GitHub documents distinct usage fields for used_copilot_cloud_agent, daily_active_copilot_cloud_agent_users, weekly_active_copilot_cloud_agent_users, and monthly_active_copilot_cloud_agent_users. In the same metrics set, it also documents daily_active_copilot_code_review_users, weekly_active_copilot_code_review_users, and related code review indicators. That is the cleanest answer: the surfaces are tracked separately in usage telemetry.

The inconvenient part is that the separation is not uniform across every report. GitHub says repository-level reports include pull requests created by Copilot cloud agent and reviewed by Copilot code review, which helps at the repo level. But the aggregated reports do not expose the same user-level detail, so if you want a precise split by team or person, you need the right report shape and sometimes a join step.

Another part people get wrong is assuming there is one “agent usage” number that covers everything. GitHub’s metrics page distinguishes chat, agent, CLI, cloud agent, and code review. It also notes that used_agent does not include Copilot code review activity, because code review is captured separately. That makes it possible to compare the two surfaces without mixing them together.

If you are looking at billing, the answer changes. GitHub says Copilot usage is measured in AI credits for billing, and Copilot features that use AI models consume those credits. That billing layer includes Copilot Chat, Copilot CLI, Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents, while code completions and next edit suggestions are not billed in AI credits. Billing is therefore not the same thing as the usage split you get from metrics.

For measurement work, the practical question is what you want to prove. If you want to know whether developers tried each surface, use the active-user fields and repository-level pull request records. If you want to estimate output, GitHub’s usage metrics also include lines-of-code measures across completions, chat, and agent features, but those are directional and should not be treated as a precise productivity score for code review versus cloud agent.

A concrete way to separate them is to use the metrics API or export, then filter by the relevant fields. For cloud agent, look for the cloud-agent usage indicators. For code review, look for the code-review usage indicators and review-related repository records. GitHub also says team-level metrics are not pre-aggregated, so teams are constructed by joining user-teams data with per-user usage data. That is the step that usually gets skipped and then causes bad charts.

The short version is simple: yes, they can be measured separately in usage metrics, and GitHub’s docs name separate fields for each. The catch is that some dashboards are broader than the raw reports, and billing credits do not give you the same separation. If you need exact attribution, use the usage metrics API or export, not just the billing view.

If you are deciding how to track this inside a team workflow, DevConnect keeps the testing side free and simple, which matters when you are comparing tool usage across people and repos. The platform itself is not the source of truth for Copilot metrics, but it can fit into a broader testing workflow without adding cost. https://devconnectplatform.com

What metric should I use for cloud agent versus code review

Use the cloud-agent active-user fields when you want adoption of the agent surface, and use the code-review active-user fields when you want adoption of review activity. GitHub’s documentation explicitly separates these counters, and repository-level reports also record pull requests created by cloud agent and reviewed by code review.

Can I compare cloud agent and code review inside the same dashboard

Yes, but only if the dashboard is built from the underlying metrics rather than from one rolled-up field. GitHub exposes multiple report shapes, and the aggregated dashboard data is not the same as the per-user or repository-level export. If you need a side-by-side comparison, build it from the API or export.

Does billing tell me the same thing as usage metrics

No. Billing tracks AI credit consumption, while usage metrics track how people used Copilot surfaces. GitHub says cloud agent is billed in AI credits, and it says code review has its own usage signals. Those are related, but they answer different questions.

What usually goes wrong in reporting

Teams often mix aggregated billing data with feature usage data and expect them to line up exactly. They do not. Another common mistake is treating used_agent as a catch-all for every agent-like action, when GitHub states that code review is tracked separately. That is where false totals usually come from.

Is there a single source of truth for this split

Yes, GitHub’s Copilot usage metrics documentation is the source of truth for the usage split, and the billing documentation is the source of truth for AI credit consumption. Use the first for adoption and activity, the second for spend.