GitHub's October 28, 2026 Universe releases expand Copilot from individual coding sessions toward ongoing project work and desktop application tasks. The app announcement places Projects and computer use in public preview. A separate release makes Copilot Fusion generally available. Those availability labels belong to different capabilities and should stay separate in rollout plans.
Projects provide an ongoing place for work
The Projects documentation describes goals spanning repositories, sessions, recurring tasks and generated artifacts. A project can also operate without a repository. Its overview groups work by status, including items needing input, open pull requests, active sessions and completed work.
For a team preparing a product change across two repositories, that offers a shared context for implementation and supporting material. The useful discipline is to state which deliverables belong together: code, release notes and a decision log might share context while still having different reviewers. A coordinated goal does not eliminate those ownership distinctions.
Computer use adds a different permission surface
The announcement introduces desktop computer use on macOS and Windows. It can inspect application content and interact with controls, including in applications without a conventional integration. That extends the reachable workflow beyond repository files.
The operational documentation adds an important qualification to the announcement's approval language: approval depends on Tool Permissions and saved app approvals. macOS requires Accessibility and Screen Recording permissions. Removing a saved approval does not revoke an already running session's access; operators must stop the operation or end the session as appropriate.
A sensible initial task would compare information across two applications and produce a draft for review. Specify the applications, records and expected artifact. Add submission or account-changing steps only after the team has established what the permitted interaction actually covers and how a person can interrupt it.
Fusion changes model selection
Copilot Fusion selects and orchestrates models rather than identifying one fixed model. GitHub offers Standard, High and Max effort levels, with single-model, cascade and critique execution patterns. Enterprise and organization model policies constrain the models it can use.
GitHub says usage is billed for each model involved at that model's standard rate, with a temporary paid-plan discount announced at launch. Consequently, the effort setting is an operational choice as well as a convenience. Model orchestration should be evaluated on total accepted-task cost and elapsed time, including revisions and reviewer work.
Test a complete, reviewable result
These releases can be evaluated together without treating them as one undifferentiated feature. Select a bounded project, record its allowed tools, choose an effort level and define acceptance evidence. Compare the result with a familiar workflow using the same input material.
For a cross-repository change, acceptance might require passing checks in each repository and a human review of the combined behavior. For a desktop task, it might require a saved draft and a comparison against the original records. Record who owns each approval, which evidence closes the task and what should happen if a session stops before producing its intended artifact. Keep that record with the project. Measuring those outcomes exposes whether broader coordination reduces work for the team or merely moves it into a different review queue.