On October 28, 2026, right after its Universe conference, GitHub published a changelog entry adding three capabilities to the Copilot app: multi-repo workspaces called Projects, connectors to Slack and Figma with image generation, and a public-preview computer use feature for macOS and Windows. Together they push Copilot beyond the single repository, toward cross-cutting workflows driven by agents.
Projects: a multi-repo workspace
Projects, in public preview, gives work that spans more than one repository, or isn’t code at all, a single workspace for context, agents and results. Users describe a goal, add context (repositories, design files, chat threads), then choose which models and tools get used. Copilot then runs several agents in parallel and returns results, such as pull requests, ready for review. Single-repo sessions remain available for more focused work, and project automations let teams schedule repeatable tasks.
Figma, Slack and image generation inside chat
Copilot can now draw on context from Slack and Figma to research a request, draft a spec, or work on a design alongside the code. Generating and editing images from natural-language prompts also lands directly in chat, bringing the tool closer to consumer image generators for quick illustration or mockup needs.
Computer use: a door into apps without an API
The most structural addition is computer use, in public preview on macOS and Windows. Copilot can read what’s displayed in an app, click interface elements, enter or edit text, and move across applications, including ones with no API, CLI or MCP integration at all. GitHub notes that the agent asks for approval before taking control of an app, and that organizations can disable the feature through managed settings.
It’s turned on per machine, in the app’s settings: Settings > Computer Use > Enable Computer Use. On macOS, it also requires granting Accessibility and Screen Recording permissions.
| Capability | Status | Platforms | What it changes |
|---|---|---|---|
| Projects | Public preview | Copilot app | Multi-repo workspace, parallel agents, PRs as output |
| Slack / Figma connectors + images | Available | Copilot app (chat) | Product/design context, prompt-based image generation |
| Computer use | Public preview | macOS, Windows | Drives apps without an API, CLI or MCP |
# Update the Copilot app before turning on the new capabilities
# macOS, via Homebrew
brew upgrade --cask github-copilot
# Windows, via winget
winget upgrade GitHub.CopilotApp
Copilot can now read what’s on screen in an app, click its controls, and move from one piece of software to another, including ones with no API or CLI at all.
Computer use grants an agent “user-like” access to third-party applications, sometimes without the audit trail an API call would leave. Before rolling it out across an organization, it’s worth keeping the feature off by default through managed settings and limiting it to pilot accounts on dedicated machines.
Key takeaways
With Projects, the Slack/Figma connectors and computer use, GitHub is shifting Copilot from a coding assistant toward an agent capable of operating entire applications, including ones never designed to be automated. Both public previews, Projects and computer use, are worth testing on a narrow scope before any broader rollout, drawing on the same governance principles already applied to editorial agents running in production.
Computer use is the feature in this announcement I find most interesting, and the one that worries me most at the same time. Being able to automate an application with no API and no export is genuinely useful in the field, I’m thinking of certain line-of-business tools or third-party extensions with no integration of their own. But until the ecosystem around the MCP protocol sets an equivalent standard for auditing what an agent does on a graphical interface, I’ll keep recommending clients treat this as an experimental feature, not a production automation. — Simon Janvier
Further reading: the official changelog entry at github.blog/changelog.
