Copilot code review resolves its own comments, and Lite reviews now run an ensemble

A review comment that no longer applies is noise, and until this change someone had to clear it by hand. GitHub changed that on September 11, 2026. When you push a commit that addresses a comment from Copilot code review, Copilot resolves that comment itself during the rereview. Feedback that is still outstanding stays open, so the open threads on a pull request describe what is left rather than everything that was ever said. Applying one of Copilot's suggestions now also fills in a commit message written from the change, instead of the standard one.

What the review agent is now allowed to run

Copilot code review already read the files in a pull request. It now uses the full set of shell tools from the Copilot SDK, running behind the Copilot agent firewall. The review agent can run build commands, execute tests, run a targeted script, and pull information from the tools and APIs available to it. The practical difference is that a review no longer stops at reading the code; it can also run it.

GitHub says that in its own experiments developers left more positive feedback on Copilot's comments, and that Copilot surfaced more high severity findings and fewer nits. No figure was published with that claim, which sets it apart from the measurement in the next section.

The Lite level now works as a committee

Until now a review at the Lite effort level was the work of a single agent. It is now an ensemble: each agent contributes its own reading of the code, and Copilot merges their findings into one review. GitHub reports that the ensemble raised the average number of addressed comments per review by 47% for high severity findings, 31% for medium and 11% for low, while cutting the cost of a review by about 8%.

The figure counts addressed comments, not findings produced, and the difference matters. One reading is that it measures how much of the feedback a developer accepted and acted on, which is a better signal than raw volume but says nothing about what the review missed. The measurement is the vendor's own, published without a description of the sample, and nobody has repeated it independently.

Usage metrics now count the VS Code Agents window

On the same day, GitHub added generally available metrics for the dedicated VS Code Agents window to Copilot usage metrics reports. Aggregate enterprise and organization reports, over both 1-day and 28-day periods, gain a daily count of unique users active in that window and an aggregate of session counts and total user messages. Per-user reports gain an indicator of whether that person used the window, plus the same totals for them.

• The metrics cover only the dedicated VS Code Agents window, and stay separate from editor-window Agent Mode and from generic usage rollups.
• The fields are optional: where the data is unavailable they stay absent or null, so existing reports keep parsing.
• Access goes to enterprise owners and billing managers, organization owners, and anyone with a custom role that grants View Copilot Metrics, and the Copilot usage metrics policy has to be enabled.