New controls for automatic model selection

GitHub Copilot users can now choose how automatic model selection weighs cost, quality and response time. The new efficiency, balance and intelligence tiers are designed to make those trade-offs explicit when Copilot chooses a model.

GitHub says all three tiers select from the same set of available models, so the update changes the selection preference rather than introducing a separate model catalog. The feature is rolling out across VS Code, Copilot CLI and the Copilot app.

Code review gains more precise feedback handling

Copilot code review now resolves comments that have been addressed when a subsequent review is run, while leaving outstanding feedback open. This is intended to keep later review results focused on issues that still require attention.

The reviewer can also suggest commit messages when users apply its code suggestions. Reviews can use shell tools to validate changes, adding a way to check proposed modifications through project tooling. GitHub also says Lite reviews now combine findings from multiple agents.

More visibility and control for organizations

Usage metrics for Copilot agents are now generally available in the dedicated Agents window in VS Code. Enterprise and organization reports include daily active users, session counts and user message totals, along with user-level activity data. GitHub notes that these figures are separate from metrics for the editor window.

Copilot can also suggest values for repository custom properties, a feature now in public preview. For example, when an internet-facing property is created, Copilot can suggest yes and no as allowed values. Organization and enterprise owners can enable or disable these suggestions through a Copilot policy.

Another generally available feature lets users request a higher Copilot budget from their organization or enterprise after reaching an AI credit limit. Owners and billing managers can approve, adjust or deny requests in settings, and approval restores access to AI credits. The capability applies to Copilot Business and Copilot Enterprise plans with usage-based billing, but GitHub says it is unavailable for enterprises with managed users.

Sentry reports become a starting point for fixes

The GitHub Copilot app now includes a Sentry canvas intended to help developers move from a production crash report toward a code fix. Users can review errors, stack traces and related context, then work with Copilot to investigate the cause, validate a fix and prepare a pull request.

The integration places the investigation and proposed remediation in a single Copilot workflow, according to GitHub. The release announcement describes the capability as a way to work from Sentry reports to a pull request, without adding details about supported Sentry configurations or deployment requirements.

Agent workflows expand in VS Code

Agents can now run with a project’s tools and dependencies in local Dev Containers through the Agents window. The support is rolling out gradually and requires Docker together with a supported Dev Container configuration.

Agent Host sessions can also create pull requests directly from the Agents window. Developers can review the generated title and description, choose whether the pull request should be a draft, and create it directly or ask the agent to do so.

For session management, inactive sessions can be automatically marked as Done once all of their pull requests have merged. Optional deletion can occur after a separate grace period. Automatic cleanup is opt-in and is currently in preview, giving teams control over whether completed sessions should be removed.

Sources

• GitHub ChangelogGitHub Copilot weekly releases — September 14