GitHub Copilot can now control desktop apps and run workflows defined in code

GitHub published three Copilot changelog entries on October 1, 2026. Two of them change what the agent can do. Computer use lets Copilot operate desktop applications, and dynamic workflows run a process that you define in code. The third entry sums up September in VS Code, versions 1.136 through 1.140.

What computer use does

Computer use is in public preview in Copilot CLI and the GitHub Copilot app, on macOS and Windows. Copilot reads what an application shows, through the operating system's accessibility tree or through screenshots. It then clicks controls, enters and edits text, presses keys, scrolls, and drags. GitHub says the feature is meant for software that offers no API, no command-line interface, and no Model Context Protocol (MCP) integration. Its documentation adds that a more direct tool, where one exists, typically gives more predictable results.

The feature is disabled by default. In Copilot CLI you turn it on with /computer on, check its status with /computer show, and turn it off with /computer off. In the app, the switch is in Settings, in the Computer Use section. On macOS, Copilot guides you through granting the Accessibility and Screen Recording permissions. The announcement does not say which Copilot plans include computer use.

Who approves what

Copilot asks for approval before it controls an application. You can grant access for the current session, save the approval for future sessions, or deny access. The saved approval, called Always allow, is stored locally and applies to both Copilot CLI and the Copilot app on the same computer. Removing it deletes the approval for future sessions, but it does not revoke access already granted in a running session.

Organizations can disable the feature through managed settings, and enabling it locally does not override an enterprise policy. GitHub's documentation warns that computer use can select the wrong control or enter text in the wrong location. It also warns that ambiguous instructions or unexpected on-screen content may cause unintended actions that affect your device, data, or connected accounts. To interrupt a running operation, press Esc twice in Copilot CLI or click Stop in the app.

Dynamic workflows put the process in code

Dynamic workflows are in public preview in Copilot CLI, the GitHub Copilot app, and the GitHub Copilot SDK, on all Copilot plans. A dynamic workflow is a program inside a Copilot extension. The code defines the steps, the point where agents get involved, and the way their results are used. Steps can run one after another, in parallel, or both, and agents handle the parts that need analysis or judgment.

That is the difference from the /fleet command, where Copilot itself decides how to divide the work among subagents. A workflow can pause at a checkpoint so that you can review the results, and subagents can verify each other's findings. GitHub's documentation says that reusing a workflow means reusing its steps and rules, while agents can still give different answers on different runs.

The app needs no setup. In Copilot CLI you enable experimental features, with the --experimental option or with /experimental on. You can write a workflow yourself or ask Copilot to write it. According to the documentation, the chat agent creates or runs a workflow only when you explicitly ask for it, or when a skill or slash command tells it to. An existing workflow also runs directly from the terminal with copilot workflow run, so it can be used from a script or from a CI/CD pipeline.

September in VS Code

The monthly summary covers VS Code 1.136 through 1.140. Two preview features stand out. Automations run a task on an hourly, daily, or weekly schedule, or on demand. Agent merge lets the agent in an active session handle review feedback, failed checks, merge conflicts, and workflow reruns until a pull request is ready to merge. You can also continue a Codex conversation from the ChatGPT app in VS Code.