# GitHub Copilot Desktop App Targets Parallel Agentic Workflows > Source: > Published: 2026-06-17 08:00:00+00:00 GitHub has introduced the GitHub Copilot app, a desktop control centre for agent-native development that aims to keep engineers in charge while AI agents handle more coding work. [Mario Rodriguez writes on the GitHub blog](https://github.blog/news-insights/product-news/github-copilot-app-the-agent-native-desktop-experience/) that the recent wave of coding agents has brought faster delivery but also "disjointed workflows, more context switching, and too much time spent reviewing agent-generated code". The Copilot app is presented as a place where these agents can be directed, observed, and tied to existing GitHub workflows rather than running as opaque background processes. The Copilot app is a standalone desktop application in technical preview for Copilot Pro, Pro+, Business, and Enterprise users. It presents a "My Work" view that shows sessions, issues, pull requests, and background automations across connected repositories, with each agent session running in its own git worktree to keep changes isolated. [Rodriguez explains](https://github.blog/news-insights/product-news/github-copilot-app-the-agent-native-desktop-experience/) that this isolation is intended to let teams run several agents in parallel without them interfering with one another or with a developer's local branches. "The GitHub Copilot app is the latest in a line of AI tooling from GitHub that is transforming our business. Moving beyond AI assistance, the app has provided a much-needed control center for agentic development." -- David Jobling Git worktrees underpin the app's model of parallel work. Each session has its own branch, files, and task state, which means that one agent can investigate a production bug while another implements a backlog issue and a third responds to review comments. When an agent is ready to propose changes, the app leans on GitHub's pull request mechanism, including a feature called Agent Merge that follows a change through review, checks, and merge, and can be configured to drive continuous integration back to green or address review feedback before merging. GitHub is also introducing canvases as a way to make agent work visible. In the announcement, canvases are described as "bidirectional work surfaces" where plans, pull requests, browser sessions, terminals, deployments, or dashboards become artefacts that both humans and agents can update. Rodriguez describes this as the start of an "agent experience" layer, in which chat remains the place for instructions and discussion while canvases show the concrete work that results. "Chat is where you instruct, discuss, and reason through ambiguity. Canvases are where that intent becomes visible work you can inspect, steer, and verify." -- Mario Rodriguez To keep agents contained, the Copilot app offers both local and cloud sandboxes. Local sandboxes run in an isolated environment on the developer's machine with restricted access to the filesystem, network, and system capabilities, and can be governed by centrally managed policies. Cloud sandboxes run agents in ephemeral Linux environments hosted by GitHub, allowing organisations to define guardrails and to resume sessions from different devices. The aim is to allow agents to run code, inspect results, test changes, and iterate without touching production systems. Microsoft principal developer advocate Burke Holland published a [YouTube walkthrough](https://www.linkedin.com/posts/burkeholland_github-just-released-a-new-ai-development-activity-7462170276634681344-VROO/) of the technical preview showing workflows where agents pick up issues, work in isolated worktrees, and use Agent Merge to push changes through CI and code review. The Copilot app ships alongside updates to Copilot's code review and automation features. Copilot code review can now be tuned per repository, including a medium-depth review option that routes pull requests to a higher-reasoning model, and custom skills such as `/security-review` and `/rubberduck` that focus on security analysis or critical commentary on an implementation. GitHub has also brought the same code review experience to Azure DevOps, extending Copilot's reach beyond GitHub-hosted repositories. GitHub is also consolidating its agent runtimes. The GitHub Copilot SDK, now generally available in languages including Node.js, Python, Go, .NET, Rust, and Java, exposes the engine that powers the app and Copilot CLI. This lets teams embed Copilot-style planners and tool loops into their own tools and workflows rather than building agent orchestration from scratch. At the same time, the Copilot CLI has gained a redesigned terminal UI, voice input, and the ability to schedule recurring prompts or background tasks, with cloud automations that run on GitHub infrastructure but still ask permission before performing write actions. In a comparative review on Pickuma, Owen Smith argues that the Copilot desktop app marks a shift in how GitHub presents Copilot, moving beyond inline completions towards a workflow where agents own longer running tasks. [He contrasts](https://pickuma.com/for-dev/github-copilot-desktop-vs-claude-code-codex/) the GUI-based Copilot app with terminal-first agents such as Anthropic's Claude Code and OpenAI's Codex CLI, noting that all three tools can read repositories, propose multi-file edits, and execute commands but differ in "surface and focus", approval semantics, and model neutrality. Smith highlights that Copilot's pull-request-first model offers a softer blast radius than terminal agents that work directly against a checked-out tree, and he notes the ability to switch between models from Anthropic, OpenAI, and Google within a single Copilot interface. Dr Matthias Liebeck's ongoing GitHub Copilot newsflash series offers further context on the app's place within GitHub's agentic strategy. [In his coverage](https://www.linkedin.com/pulse/episode-7-first-look-new-github-copilot-app-dr-matthias-liebeck-yniue/) of AgentHQ and Copilot's agent primitives, Liebeck describes a three-layer model of ad-hoc prompting, reusable skills, and orchestrated workflows that run across Copilot CLI, VS Code, and GitHub.com. The Copilot app sits on top of this stack as a "mission control" surface where those workflows can be directed and monitored. InfoQ has previously reported on related parts of GitHub's agentic roadmap, including the technical preview of GitHub Copilot Workspace as a task-centric environment for planning and executing changes,[ the introduction of Copilot Extensions](https://www.infoq.com/news/2024/05/github-copilot-workspace-preview/) for integrating external services into Copilot chat, and the Copilot SDK that lets teams build their own agentic tools on top of the Copilot engine. [Other vendors](https://www.infoq.com/news/2025/02/github-copilot-extensions/) have taken different approaches: Anthropic's Claude Code and subagents focus on terminal-first workflows and specialist agents, while LangChain's Open SWE and Google's Jules emphasise asynchronous cloud-based coding agents. The Copilot app instead keeps the locus of control inside GitHub, making agents part of the pull request and issue lifecycle. Rodriguez closes the announcement by arguing that professional software still "demands judgment, verification, and accountability" and by positioning the Copilot app, sandboxes, code review, and partner ecosystem as parts of a single system where agents "do more of the work" while developers retain control over quality and delivery. For teams already standardising on GitHub for source control and review, the Copilot app offers a way to try parallel coding agents without leaving their existing workflows, while competitors continue to explore terminal, IDE-first, or self-hosted routes to agentic development.