GitHub Copilot: canvases bring shared interactive workspaces for live collaboration with AI agents

GitHub has introduced canvases into the Copilot app — a new way of working with AI agents that replaces purely text-based message exchange with a visual, shared workspace.

What is a canvas and how does it differ from chat?

GitHub Copilot is GitHub’s AI programming assistant, built into editors and the Copilot app, which helps users write code and manage repositories. A canvas is a shared interactive workspace where the user and an agent — an autonomous AI system that executes multi-step tasks — jointly edit content live, instead of the user waiting for a reply in a chat window and then manually carrying it forward. The difference from a classic prompt is direct: chat returns a text answer to a query, while a canvas stays open, updates in real time, and lets both sides directly change the same view.

Five practical uses for canvases

GitHub describes 5 concrete use scenarios: an issue triage helper for quickly sorting reported problems, interactive codebase diagrams that visualize a repository’s structure, worktree management — handling parallel working copies of a repository, prompt coaching that helps users shape better queries, and knowledge discovery, searching a team’s knowledge across communication platforms. Each of these uses gets a customized interface within the same canvas framework.

How is a canvas launched?

A canvas is created with the /create-canvas command inside the Copilot app. Once launched, the interface automatically adapts to the user’s specific workflow — for example, a canvas for codebase diagrams shows a graphical view of dependencies, while a canvas for issue triage shows a list and filters of reports. GitHub is thereby moving agentic collaboration from a one-way query toward a shared, live-updated workspace that stays open and available for further collaboration until the task is finished, instead of disappearing after a single exchanged answer.