GitHub Copilot app introduces bidirectional canvases for custom agent workflows

The GitHub Copilot app now supports canvases, customizable interfaces that allow users and AI agents to collaborate on shared workspaces. Unlike fixed screens, these canvases—such as kanban boards, issue triage boards, or release checklists—are shaped around the user's specific workflow.

Creating and Refining Interfaces

Users generate a canvas by entering the /create-canvas skill in an agent session and describing the desired workflow in plain English. The prompt should specify:

• The workflow the canvas supports.
• Actions the user can perform in the interface.
• Actions the agent can perform.

The agent builds the interface in the right-side panel without requiring manual coding or design. Users can refine the canvas by asking the agent to add columns, filters, or convert the layout into a checklist. Once created, the canvas is saved as an extension, allowing it to be shared with teams or kept as a personal tool.

Bidirectional Collaboration

Canvases operate on a bidirectional model, enabling real-time interaction between the user and the agent. When a user clicks a button, updates a field, or moves a card, the shared state changes immediately, and the agent sees the update without a separate sync step. Conversely, users can instruct the agent to use the canvas's native capabilities to add notes or move items, with changes appearing instantly in the interface. This replaces the traditional "command and wait" interaction with simultaneous steering of the work.