GitHub has introduced canvas extensions for Copilot, turning AI from a conversational tool into a shared workspace where developers and agents collaborate in real time. A canvas is an interactive surface in the Copilot app that the agent can update as it works, while the user interacts through clicks, edits, and other actions. Those interactions feed back to the agent, processed locally by the canvas.
How Canvases Work
Developers create a canvas with /create-canvas in an agent session, describing what they want and what capabilities it should support. Because canvases are generated from prompts and evolve with the workflow, they can take many forms. The agent updates the canvas as it works, and the user shapes the experience by asking Copilot to iterate, adding functionality or refining features.
Practical Examples
One example is an issue triage interface where each issue appears as a card. Users swipe right to ship, left to reject, with the canvas tracking decisions and sorting items automatically. Another generates an interactive architecture diagram of a codebase, with nodes representing system components that users can hover, drag, and filter. A worktree view shows active and stale Git sessions with one-click cleanup. A prompt coach analyzes previous interactions and suggests improvements for missing context or syntax errors. A people finder searches across Slack, Teams, email, and documentation to surface who has knowledge of a specific file or topic.
Why It Matters
Canvases address a real limitation of conversational AI. Some tasks are easier when you can see and interact with information directly. Triage, visualization, and multi-source aggregation all benefit from a visual interface that makes patterns and connections visible. Instead of working through long chains of prompts and responses, users interact directly with the information in front of them.
The feature is available now in the GitHub Copilot app.