GitHub Copilot canvases let developers visualize and interact with AI work

GitHub has rolled out canvases, a new feature within the Copilot app that transforms how developers interact with AI-assisted work. Canvases are shared, interactive surfaces—called canvas extensions—where developers and AI agents can collaborate in real time. Rather than cycling through a long chain of prompts and responses in a conversation, developers can work with visual interfaces that agents update as they operate, and developers can shape the experience through direct interaction: clicks, edits, and other actions that either feed back to the agent or are processed locally within the canvas.

To create a canvas, a developer types /create-canvas in an agent session and describes what interface they want and what capabilities it should have. Because canvases are generated from a prompt and evolve with the workflow, they can take many forms. The article walks through five real-world examples. An issue triage helper uses a card-based interface where each GitHub Issue appears one at a time and the developer can swipe right to ship or left to reject, with the canvas updating in real time to track decisions. An interactive codebase diagram renders a dynamic visualization where each node represents a part of the system and developers can hover, drag, and filter to explore layers and relationships. A sessions worktree view shows all active GitHub Copilot app sessions and their git worktrees, clearly marking which are active or stale, with one-click cleanup. An agent prompt coach reviews past interactions and suggests improvements—flagging missing context, spelling errors, and syntax issues. A knowledge finder searches across Slack, Teams, email, and documentation to surface people most connected to a given file or subject, showing where the connection was found so developers know who to reach out to.

The key insight GitHub emphasizes is that canvases turn AI from a conversational tool into a place where developers can visualize information, explore workflows, and interact directly with the work. Rather than reading through agent suggestions, developers see patterns, spot connections, and take action immediately. GitHub notes that canvases are available now in the GitHub Copilot app, and developers can read the documentation to learn more about how to build and use canvas extensions.

GitHub's introduction of canvases reflects a shift in how developers work with AI—moving beyond pure conversation to visual, interactive collaboration. The article frames this as a solution to a specific problem: some tasks are fundamentally harder to handle through a linear chain of prompts and responses. When a developer needs to triage a backlog, visualize how components of a system interconnect, or find information scattered across multiple tools, a visual interface enables faster pattern recognition and more direct action.

Canvases are positioned as extensions of the Copilot app that preserve the agent's role while giving developers a workspace where they can see changes in real time and steer the work through direct interaction—clicking, dragging, editing, and swiping. The iterative nature is key: a canvas is not a static output but something that evolves alongside the developer's workflow. The five concrete examples (issue triage, codebase diagrams, worktree management, prompt coaching, and cross-tool knowledge search) suggest that GitHub is targeting high-friction workflows that benefit most from visualization and hands-on interaction rather than conversation.