GitHub on May 14 opened a technical preview of a standalone Copilot desktop application, a GitHub-native workspace designed specifically for running coding agents end-to-end — from issue triage to pull request merge — without bouncing between a terminal, an IDE, and a browser. The release marks Microsoft and GitHub's most ambitious attempt yet to compete directly with the wave of agent-first tools that have reshaped how developers write software over the past year.
The app is available on Windows, macOS, and Linux from the start of the preview. GitHub Copilot Pro and Pro+ subscribers can sign up for early access immediately, while Business and Enterprise customers will be enrolled progressively through the week, with organization admins required to enable preview access and the Copilot CLI in policy settings before their developers can join.
A workspace built around parallel agents
The core design idea is that each agent session lives in its own git work tree, branch, and task state. That isolation, according to GitHub's changelog, lets developers run multiple agents against the same repository without them stepping on each other's changes — a recurring pain point with current agent tooling, where parallel runs frequently collide over shared files and branches.
A session can start from an issue, a pull request, a free-form prompt, or a previous session, and the app keeps issue details, repository state, review comments, and CI checks connected throughout the run. Developers can review diffs, run shell commands, and test changes inside an integrated terminal and browser before pushing the work into a pull request.
Agent Merge and follow-through
A feature the company is calling "Agent Merge" handles the post-PR loop that often eats developer time: it can address review comments left by humans, fix failing checks, and then merge the pull request once the user's stated conditions are met. The pattern is meant to give developers a way to delegate not just the initial implementation, but the back-and-forth of code review and CI fixes — an area where today's agents typically hand control back to a human.
Implications
The launch lands in a crowded market. Cursor, Cognition's Devin, Anthropic's Claude Code, and OpenAI's Codex have all expanded aggressively into agentic coding over the past twelve months, with several reaching valuations above $25 billion. By shipping a dedicated, OS-native client tied to the GitHub graph — issues, PRs, branches, checks — GitHub is betting that proximity to the source of truth for code beats the slicker UX of independent IDE startups.
For enterprise buyers, the admin-controlled rollout and policy gating are likely as important as the agent itself. Multiple large organizations have spent 2026 wrestling with how to govern parallel agent activity inside their repositories, and a vendor-supplied tool with policy hooks may prove easier to approve than third-party alternatives. The technical preview tag, however, is a reminder that GitHub is shipping fast rather than shipping finished: the first weeks of broader testing will determine whether the parallel-session model scales to the way real teams actually work.
