What happened
GitHub’s August 28 changelog describes new GitHub Copilot features in Visual Studio. Organization and enterprise owners can publish custom agents that Visual Studio detects and displays with descriptions and organization source in the agent picker. The capability requires a GitHub organization.
GitHub says August 2026 brought several Copilot changes to Visual Studio, centered on user control, organizational sharing and pre-pull-request review. Organization-level custom agents can now be published by GitHub organization and enterprise owners for use across repositories in their organization. Visual Studio is described as automatically detecting those agents and showing their descriptions and organizational source in the agent picker. The source specifies that this functionality requires a GitHub organization, so it is not presented as an organization-free feature.
The update also changes how users manage Copilot models and usage. From the Copilot prompt box, users can open the context window and select View all Copilot usage to see plan details. GitHub says revised notifications clarify when users are approaching a limit and what options are available. Supported models receive Low, Medium and High thinking-effort controls, allowing users to select lower effort for straightforward work or higher effort for complex debugging, algorithms and architecture decisions. GitHub frames this as a way to balance reasoning depth with token usage, but does not identify every supported model or quantify the tradeoff.
Visual Studio users can pin favorite models, collapse less-used models and open Manage models for a detailed view. GitHub says that view includes model capabilities, context-window sizes, cost information and controls for Copilot and custom models. Separately, the Git agent can review uncommitted changes or commits before a pull request is opened. Findings appear inline in the editor and in a navigable list in Git Changes; users can continue the Copilot Chat conversation to understand or address suggestions. GitHub says these reviews work with both GitHub and Azure DevOps repositories and are available across Copilot Free, Student, Pro, Pro+, Max, Business and Enterprise plans.
Why it matters
The update gives developers and organizations more control over how Copilot is configured and used. Model-effort settings can trade reasoning depth against token usage, while custom agents could help teams distribute specialized workflows. Git-based reviews also move Copilot beyond code generation toward checking changes before a pull request.
The most consequential change is the combination of model selection and thinking-effort controls. Developers often face a choice between a faster, lower-cost response and a more extensive analysis. GitHub’s controls make that choice explicit inside Visual Studio, at least for supported models. This could make Copilot easier to adapt to different tasks, but the source does not establish that higher effort produces more accurate code or that lower effort materially reduces cost in every plan or workflow.
Organization-published custom agents could make specialized instructions and workflows easier to share across repositories. The organization source shown in the agent picker may also help users distinguish centrally maintained agents from personal or other custom configurations. However, the source does not explain the permission model, review process, update behavior, isolation between repositories, or safeguards against an incorrectly configured agent affecting many projects. Those details will matter for teams using agents in production development environments.
The Git agent feature places AI review earlier in the software-development process. Reviewing uncommitted changes and commits can give developers feedback before a pull request is created, potentially catching issues while the relevant context is still fresh. Inline findings and a navigable Git Changes list may reduce the friction of inspecting suggestions. Still, GitHub provides no accuracy rate, defect-detection comparison, false-positive rate or evidence that the feature improves software quality. The feature should therefore be understood as an additional review aid rather than a substitute for testing, human review or repository-specific safeguards.
What to watch next
The source does not provide independent testing, performance measurements, pricing changes, adoption data or details about which models support each thinking-effort level. Users should watch how reliably custom agents are surfaced, how model costs and context limits affect workflows, and whether Git-agent findings reduce defects without creating additional review noise.
A key unknown is the exact scope of model support. GitHub says thinking-effort controls are available for supported models, but the changelog does not list them. The model-management view is said to show capabilities, context-window sizes and cost information, yet the source does not state whether those costs are estimates, plan-dependent charges or usage indicators. Users evaluating the feature will need to inspect the controls available in their own Visual Studio and Copilot plan.
Organizations should watch how custom-agent publishing interacts with governance. The source confirms that organization and enterprise owners can publish agents across repositories, but it does not describe approval workflows, audit logs, version controls or how agents are removed. Teams may need to determine who can create, modify and invoke these agents, what repository context they receive, and how their outputs are reviewed before being used in code or development operations.
The Git agent’s practical value will depend on the quality and scope of its findings. GitHub says it reviews uncommitted changes and commits and can work with GitHub and Azure DevOps repositories, but does not disclose supported languages, repository-size limits, latency, data-handling terms or how it treats generated code and sensitive material. Future release notes, documentation and user testing may clarify whether the feature complements established static analysis and human review or mainly adds another layer of suggestions.
The update is dated August 28, 2026, and GitHub points users to the Visual Studio blog and release notes for more information. The source does not provide rollout timing beyond saying the update is available to users on all listed Copilot plans. It also does not report independent reactions or testing. Those omissions leave availability details, operational limits and real-world effectiveness unresolved.
