· By the ToolNav Team · 5 min read GitHub Copilot AI Coding AI Agents Developer Tools Build 2026

GitHub launches the Copilot App — an agent-native desktop where every session runs in its own worktree

TL;DR

At Microsoft Build 2026 on June 2, GitHub introduced the GitHub Copilot App — a standalone, agent-native desktop experience for running and steering coding agents. It centres on a 'My Work' view across connected repositories, runs each agent session in its own isolated Git worktree so parallel agents don't conflict, and ships alongside the now-generally-available GitHub Copilot SDK. The app is in technical preview for paid Copilot users on Windows, Mac, and Linux.

Technical preview

GitHub Copilot App status — available to existing Copilot Pro, Pro+, Business, and Enterprise users

Worktree per session

Each agent session runs in its own isolated Git worktree — parallel agents on one codebase without conflicts

6 languages

GitHub Copilot SDK now GA in Node.js/TypeScript, Python, Go, .NET, Rust, and Java

Win / Mac / Linux

Copilot App runs on Windows 11 (incl. Arm), macOS, and Linux

At Microsoft Build 2026 on June 2, 2026, GitHub introduced the GitHub Copilot App — a dedicated desktop application built around running and supervising AI coding agents, rather than a panel bolted into an editor. It is GitHub's clearest statement yet that agentic coding has outgrown the inline-suggestion model: the app is designed for developers who are increasingly orchestrating multiple agents at once rather than accepting one completion at a time.

What the Copilot App is. The app centres on a "My Work" view that shows work in motion across your connected repositories — active agent sessions, issues, pull requests, and background automations — in one place. The defining technical decision: every agent session runs in its own isolated Git worktree, so multiple agents can operate on the same codebase in parallel without stepping on each other. GitHub also introduced "Canvases" — bidirectional work surfaces where a plan, a pull request, a browser session, a terminal, a deployment, or a workflow state becomes something you can inspect, steer, and verify alongside the agent.

Availability. The Copilot App is in technical preview as of Build, available to existing Copilot Pro, Pro+, Business, and Enterprise subscribers. It runs on Windows 11 (including Windows 11 on Arm), macOS, and Linux. There is no separate price — it is part of the existing Copilot subscription tiers.

The Copilot SDK is now generally available. Alongside the app, GitHub announced general availability of the GitHub Copilot SDK across Node.js/TypeScript, Python, Go, .NET, Rust, and Java. The SDK exposes the same agentic runtime that powers the Copilot App, meaning developers can build their own tools and integrations on the runtime GitHub uses itself. For teams building internal developer tooling or commercial products on top of Copilot, this is the more durable announcement: a stable, multi-language interface to the agent runtime.

On the model side. The confirmed model-related change at Build was narrow: GPT-4.1 was deprecated as a selectable model on June 2. Copilot's model selector continues to expose multiple model families on paid plans, and nothing about that selection or about Copilot's pricing changed with the Copilot App launch.

For Copilot users today. If you are on a paid Copilot plan, the Copilot App is available to try now in technical preview — the isolated-worktree-per-session design is the feature most worth testing if you've run into agents colliding on a shared branch. Nothing about Copilot's pricing or the AI Credits billing system that went live June 1 changes with this announcement. See the best AI coding tools roundup for how it sits against Cursor and Claude Code.

Why It Matters

The unit of AI coding work is shifting from the completion to the session. For two years, AI coding meant inline suggestions inside an editor — accept, reject, repeat. The GitHub Copilot App is built around a different assumption: that developers are now running multiple autonomous agents in parallel and need a place to launch, watch, and steer them. The isolated-worktree-per-session design is the tell — it only matters if you're running several agents on the same codebase at once. For the broader market, this puts GitHub in more direct competition with the agent-orchestration features in Cursor (Background Agents) and the terminal-native approach of Claude Code. The Copilot SDK going GA is the quieter but more strategic move: it lets the whole ecosystem build on GitHub's agent runtime, which is how platforms entrench. For operators, the practical signal is that 'managing a fleet of coding agents' is becoming a first-class workflow with dedicated tooling — worth understanding even if you're not ready to adopt it.

Who's Affected

  • Paid GitHub Copilot users (Pro, Pro+, Business, Enterprise) — The Copilot App is available in technical preview now, at no extra cost. Worth trying if you run multiple agents or have hit conflicts from agents sharing a branch. Nothing changes with billing or your existing Copilot setup.
  • Teams building internal developer tooling — The Copilot SDK GA (Node.js/TypeScript, Python, Go, .NET, Rust, Java) is the more durable announcement: a stable, multi-language interface to GitHub's agent runtime. If you've been waiting for a supported way to build on Copilot, this is it.
  • Developers evaluating Copilot vs. Cursor or Claude Code — The Copilot App moves GitHub toward agent-orchestration parity with Cursor's Background Agents. The comparison is now about agent-fleet management UX, not just inline completion quality.

What To Do Now

  1. 1. Try the Copilot App if you already pay for Copilot and run more than one agent at a time. The isolated-worktree-per-session design directly solves the 'two agents fighting over the same branch' problem. It's a technical preview — evaluate on a non-critical repo first.
  2. 2. The Copilot SDK GA is the announcement to act on if you build tools. A supported, multi-language interface to the agent runtime is more durable than any single app feature. If you've been prototyping against unofficial Copilot interfaces, migrate to the SDK.
  3. 3. Re-check your model selection after the GPT-4.1 deprecation. GPT-4.1 was deprecated as a selectable model on June 2. If any of your workflows or saved configs defaulted to it, confirm they've moved to a current model family.

More on this topic — Best AI Coding Tools

Independent Review

GitHub Copilot

Pricing, pros and cons, real-world verdict — no affiliate spin.

Read the GitHub Copilot review

The AI Hustle Playbook Newsletter

Get one practical AI playbook each week.

Tools, workflows, and side-income ideas — curated for people who want to build, not browse forever.

No spam. Unsubscribe anytime. We respect your privacy.