What happened
GitHub says work does not always fit within one repository or session: a feature may span repositories, while a launch plan may involve specifications, designs, and assets that do not belong in a repository. Copilot Projects, now in public preview, give this work a shared home for its context, agents, and results. A user starts a project, describes a goal, and adds relevant context such as repositories, design mocks, Slack threads, or meeting notes. They can choose models and tools for the job. Copilot then coordinates as many agents as the work needs and returns results, including pull requests, for review. For a focused task in a single repository, GitHub says users can still start a session directly from that repository.
The changelog also describes project automations for repeatable work. Its example is creating a new promotional video whenever a feature lands. The post does not detail how automations are configured or what triggers are supported, but frames them as a way to reuse work rather than start each recurring task from scratch.
Projects can draw on connected tools such as Slack and Figma. GitHub says this context can help Copilot research a request, draft a specification, or work through a design alongside code. The app can also generate and edit images using natural language in chat. These are described as capabilities in the Copilot app; the announcement does not state particular plan requirements or limits for the integrations.
Computer use is in public preview on macOS and Windows. GitHub says Copilot can read app content, click controls, enter or edit text, and navigate across apps, including ones without an API, command-line interface, or MCP integration. The company’s example is an expense-report workflow. Before controlling an app, Copilot asks for approval, and organizations can disable the feature through managed settings. To enable it, users open Settings > Computer Use and turn on Enable Computer Use. On macOS, the feature requires Accessibility and Screen Recording permissions. The post gives no pricing details or timeline for general availability.
What the archive adds
This is a new fact on a known GitHub Copilot story, not an entirely new thread. Our earlier Field Note on GitHub’s Copilot Universe releases covered the broader Universe event, including Copilot app, projects, session continuity, automations, and code review. The October 28 changelog supplies concrete details on how Projects gather context and agents, and how computer use is gated and managed.
Trend signal
This supports agent-ready software exposes capabilities: Copilot’s desktop control is an explicit capability that can cross app boundaries, but it is bounded by approval and organizational settings. It also illustrates the direction in AI-native organizations move toward review and orchestration: Copilot coordinates agents and returns artifacts such as pull requests for human review, rather than treating output generation as the whole workflow.
Why now
GitHub published this changelog on October 28, 2026, during its Universe releases. The source presents the features as a coordinated expansion of the Copilot app, but does not explain a separate market trigger or announce a general-availability date.
Why it matters
For teams, the useful shift is not simply that an agent can click a screen. Projects put multi-repository work and non-code context in one place, while computer use offers a route into desktop workflows that lack an API, CLI, or MCP integration. That may reduce the number of manual handoffs needed for tasks that mix code, design, and business applications.
The controls are equally relevant to deployment. Approval before app control and an organization-level disable setting provide visible points where teams can govern access. They do not, by themselves, establish that every workflow is safe or that the agent can reliably complete it. Teams evaluating the preview should identify which apps and tasks are appropriate, keep a person responsible for reviewing results, and check what permissions the setup requires.
Limits and uncertainty
The changelog does not specify which Copilot plans include Projects, automations, integrations, or computer use, nor does it give usage quotas, pricing, or a general-availability schedule. It describes capabilities and examples, not measured reliability, task-completion rates, or security results. The approval prompt and managed setting are stated controls, but the source does not explain the full permission model, audit trail, or how actions are logged. It also does not enumerate supported connected tools beyond Slack and Figma, or define the full range of apps that computer use can operate.
Skill to build now
Choose one low-risk workflow that crosses tools—for example, turning a design discussion into a draft specification—and write down its required inputs, expected output, and human review step. If testing computer use, start with a non-sensitive desktop task, confirm the requested operating-system permissions, and agree on who may approve actions. That gives a team a concrete way to assess whether Projects or computer use reduces handoffs without assuming the preview is production-ready.
Reader takeaway
GitHub is combining agent coordination, reusable project workflows, connected context, and desktop interaction inside Copilot app. Projects and computer use are public previews, and computer control has an approval step plus an organization-level disable setting. The announcement points to a broader work surface, but leaves plan eligibility, limits, reliability, and general availability unanswered.