On July 30, 2026, GitHub published its July update for GitHub Copilot in Visual Studio. The important change is not a single model. It is the shape of the workflow around Copilot Chat: a new Agent (Preview) executes tasks, built-in skills add framework-specific context, code review checks the result, and organization-level instructions carry team preferences into daily use.
The new Agent appears in the Copilot Chat agent picker. GitHub says it is built on the Copilot SDK that also powers GitHub Copilot CLI. The company describes the preview as requiring less back-and-forth, getting more tasks right on the first attempt, and producing shorter, easier-to-scan responses. Those are GitHub’s product claims about a preview feature. Actual results will depend on repository structure, tool permissions, prompts, and verification steps.
Visual Studio also adds built-in skills authored by the .NET and Azure teams. When the relevant workloads are installed, developers can find them in the Built-in category of the tool picker. The skills are off by default, so users can inspect them and enable only the ones that fit a task. That default matters because a skill is more than a one-off prompt: it can influence how an agent understands a framework, uses tools, and interprets delivery standards.
Code review is moving closer to the editor. A developer can select a block of code, choose Review Selection from Copilot Actions, and receive inline comments. The sparkle control on a comment can apply a suggestion or ask Copilot to generate one. This moves review earlier than the final pull-request pass, while the output still needs human judgement and should not be treated as an automatic approval.
For enterprise teams, the update adds organization-level custom instructions. GitHub Copilot Business and Enterprise organization owners can define shared instructions that apply across repositories belonging to the organization. Developers can disable those instructions in Visual Studio. The control surface is useful for coding style, review conventions, and response preferences, but teams still need clear precedence rules so organization instructions do not conflict with repository instructions or task-specific constraints.
GitHub says availability varies by feature. The new agent, built-in skills, and selected-code review features are available across plans, while organization-level custom instructions require GitHub Copilot Business or Enterprise. The Agent is also a public preview. Teams should check their plan and Visual Studio channel before treating the announcement as universal availability.
The broader signal is that AI in the IDE is shifting from asking one question to configuring a repeatable work loop. The useful evaluation metrics are not only response speed. Teams should measure whether the agent stays within its permissions, whether skills reduce mistakes, how accurate review comments are, and whether shared instructions remain consistent across repositories. Moving a preview into production still requires diff review, tests, and explicit rollback points.
