GitHub has launched Copilot Projects in public preview in the GitHub Copilot app, letting developers describe a goal spanning several repositories and have Copilot run as many agents as needed to deliver pull requests for review. The release, part of GitHub’s Universe announcements, also adds computer use in public preview on macOS and Windows, so Copilot can work inside desktop apps that have no programming interface.
From Single Sessions To Whole Projects
Work does not always fit in a single repository or session: a feature can span several repositories, and a launch plan can pull in specifications, designs and assets without touching a repository. Projects give that work one home, holding its context, agents and results together, with Copilot coordinating in the background and checking in when it needs the developer.
To start a project, a developer describes the goal and adds the context Copilot needs, such as repositories, design mocks, Slack threads or notes from a design meeting. They choose the models and tools for the job, and Copilot runs as many agents as the work needs in fully managed parallel sessions, bringing back results such as pull requests for review.
Project automations handle repeatable tasks, such as creating a promotional video whenever a feature lands, and Copilot can use context from connected tools such as Slack and Figma to research a request, draft a specification or work through a design alongside code. For a focused task in a single repository, developers can still start a session directly from that repository.
Computer Use Comes To The Desktop
Computer use, also in public preview on macOS and Windows, lets Copilot read app content, click controls, enter or edit text and navigate across apps, including those without an API, command-line interface or MCP integration. A developer can ask Copilot to work through tasks in desktop apps, such as an expense-report workflow, without building a custom integration first.
GitHub says Copilot asks for approval before controlling an app, and organisations can disable the feature through managed settings. On macOS, users must grant the Accessibility and Screen Recording permissions, a reminder that computer use is powerful precisely because it sees and touches the same interface a human would.
The release is distinct from GitHub’s other Universe announcement, the Copilot Fusion model router, which picks the right model for each coding task. Projects organise multi-repo work and agents, while Fusion routes individual tasks to models; the two are complementary pieces of the same agent push, not the same feature twice.
What It Means For Development Teams
For teams, the practical change is orchestration: instead of managing several agent sessions across repositories by hand, a developer states a goal and reviews the pull requests that come back. That could speed up cross-repo features, migrations and launch checklists, but it also shifts the developer’s job toward reviewing agent output carefully, since errors can now arrive at scale.
Computer use extends that shift beyond code, into the everyday desktop tasks that surround a release, from expense reports to checking tools that were never built for automation. The approval step and the organisational off-switch are GitHub’s main safeguards, and how strictly teams use them will determine whether the feature feels like assistance or intrusion.
The move also fits Microsoft’s wider agent strategy, including the Copilot harness for builders that opened agent services to every developer. GitHub is positioning the Copilot app as the place where those agents are actually managed, reviewed and shipped in everyday work.
Why It Matters
Coding assistants began as autocomplete inside one file, then grew into agents inside one repository; Projects is the next step, treating a whole programme of work as one job. If the model works, small teams can take on larger, messier projects, and the bottleneck moves from writing code to specifying goals and checking results.
The risk is equally clear: parallel agents can produce plausible-looking work across many repositories at once, and a review process built for human pull requests may struggle to keep up. Teams that adopt Projects will need review habits to match the new pace, or the productivity gain will be paid for in bugs.
Background
GitHub has been expanding Copilot from a coding assistant into an agent platform, and the Universe releases mark its most ambitious step yet toward multi-agent, multi-repository development.
Source: GitHub changelog
0 Comments