GitHub's new desktop hub for AI agents

Coding alongside AI used to mean an autocomplete pop-up helping you finish a function. Then agents arrived, and workflows descended into chaos: terminal windows everywhere, forgotten background tasks, and pull requests flooding in with zero context on what was actually tested.

At Microsoft Build, GitHub unveiled the GitHub Copilot app, a standalone desktop control center designed specifically for agent-native development. It is available in technical preview for existing Copilot Pro, Pro+, Business, and Enterprise subscribers.

The launch comes alongside staggering platform numbers. Monthly commits on GitHub have nearly doubled year over year, topping 1.4 billion, while developers run over 2 billion GitHub Actions minutes each week. The agent boom is already here, and GitHub wants to be the interface that directs traffic.

An actual cockpit for multiple agents

The app revolves around a centralized My Work dashboard. Instead of alt-tabbing through separate IDE windows, you can see all running agent sessions, active issues, and pull requests in one place.

To prevent agents from colliding, every session automatically spins up its own isolated git worktree. There is no manual branch juggling or cleanup required; the agent gets a private sandbox to test and iterate.

GitHub is also introducing Canvases, which step away from endless chat logs. A canvas acts as a bidirectional, real-time board showing plans, diffs, terminals, or preview browsers. The agent updates its progress live, and the developer can jump in to edit steps, reorder tasks, or redirect the run directly on the board.

Once code is ready, Agent Merge monitors CI pipelines, fixes failing tests, pings reviewers, and handles merge conditions autonomously based on rules you set.

Heavyweight reviews and terminal perks

Because more agents inevitably mean more pull requests, Copilot code review now includes a medium tier that routes diffs to higher-reasoning models for mission-critical repositories, while leaving lightweight models on low-risk repos. Azure DevOps users also get native review integration.

For developers anchored to the command line, Copilot CLI received a redesigned TUI, an /every cron command for background tasks, and an on-device local voice mode where audio never leaves your machine.

Meanwhile, the GitHub Copilot SDK is now generally available across Node.js/TypeScript, Python, Go, .NET, Rust, and Java, letting teams build custom internal agents using the same core runtime.

Why it matters

For individual developers, software engineering is shifting from hands-on typing to project management. Instead of writing boilerplate, you are now managing a small squad of agents tackling backlog tickets and bugs in parallel. If you already pay for a qualifying Copilot tier, this preview is well worth testing just to escape the chaos of scattered terminals.

Strategically, GitHub is playing to its ultimate advantage: platform gravity. Third-party agent startups can generate clever code, but GitHub already owns your repositories, your pull requests, your CI runners, and your review history. Building the control room on top of git worktrees and GitHub Actions is a natural defensive moat that competitors will struggle to match.

The real risk here is automation fatigue. When agents write the code, agents run the tests, and agentic review models approve the PRs, human oversight can easily become rubber-stamping. If verification standards drop, shipping bugs faster is the only guarantee.