From Coding Assistant to Desktop AI Agent Operating System
GitHub Copilot App is a desktop-native AI development environment that centralizes the control, supervision, and coordination of multiple AI agents working across software repositories at the same time. Announced at Microsoft Build, the GitHub Copilot app reframes Copilot from an inline coding helper into what GitHub calls an “agent-native” desktop experience. Instead of juggling separate Copilot chats in the browser or editor, developers get a single AI agents desktop where every task, from feature work to bug fixes, shows up as a trackable session. This shift supports a new model of multi-agent development, where AI systems can act more like semi-autonomous teammates than autocomplete tools. According to GitHub’s description, the app is positioned as a dedicated operating system for AI agents, with support for Windows 11, Windows 11 on Arm, Mac, and Linux in technical preview.
My Work: Unifying Fragmented AI Workflows Across Repositories
At the center of the GitHub Copilot app is “My Work,” a consolidated dashboard that replaces scattered chat windows with a single view of ongoing AI activity. Instead of switching between browser tabs, IDE panels, and issue trackers, developers see active agent sessions, issues, pull requests, and automations in one place. The interface shows which AI agent is building a new feature, which is handling bug fixes, and which is responding to code review feedback. This unified AI development environment is designed to solve the growing coordination problem of multi-agent development: as more agents run in parallel, it becomes harder to know who changed what, where tests ran, and how results were verified. The app turns those moving parts into a manageable queue of work items that can be supervised much like a team lead tracking human contributors across multiple repositories.
Parallel AI Agents, Isolated Worktrees, and Canvas Work Surfaces
To support multiple AI agents working on the same codebase without collisions, GitHub’s Copilot app runs each agent session inside its own isolated Git worktree. This setup lets parallel agents modify related areas of a repository while avoiding direct conflicts, similar to multiple developers working on separate branches. On top of that, the app adds “canvases” as a new kind of work surface. A canvas can display a plan, a pull request, terminal output, a browser session, or deployment status while agents update it as they progress. Developers can edit, reorder, approve, or redirect work on the same surface, turning prompts into traceable, verifiable artifacts. GitHub describes this as the beginning of “agent experience (AX)” in the Copilot app, where chat remains the place to reason about intent and canvases become the place where that intent shows up as visible work.
Sandboxes, Agent Merge, and the Path to Desktop-Native AI Development
Beyond coordination, the GitHub Copilot app introduces sandboxed execution and more autonomous pull request handling, reinforcing the idea of the desktop as an AI agents operating hub. Local sandboxes run on the developer’s machine with restricted filesystem and network access governed by central policies, while cloud sandboxes spin up ephemeral Linux environments that can be resumed from any device. These spaces let AI agents run tests, validate behavior, and experiment without touching production systems. Agent Merge extends this by carrying pull requests through CI checks, reviewer requirements, and merge conditions, with developers choosing how much autonomy the agent gets. Combined with upgraded Copilot code review capabilities and an SDK that exposes the same agent runtime to popular languages, these features mark a shift away from browser-centric tools toward a desktop-native, multi-agent development model where the operating center for AI work lives alongside the developer’s local tools.
