The Canvases feature in the GitHub Copilot app allows users to create a custom work interface from a description written in English, rather than adapting to fixed screens imposed by traditional tools. GitHub explains that these interfaces, also called Canvas extensions, can take the form of a Kanban board, an issue-triage board, a release checklist, a form, or a spreadsheet.
Creating a Canvas from a Text Description
Creating the interface does not require writing code or manually designing files. The user starts a session with the GitHub Copilot agent, then enters the command /create-canvas followed by a description of the desired interface. To obtain a useful result, GitHub suggests specifying three elements: the workflow the interface will support, the actions the user should be able to perform, and the tasks the agent can perform.
One example provided by the company is a request to create a Canvas for tracking release notes for features completed through GitHub Copilot app sessions, while providing tools for reviewing and organizing entries and allowing the agent to add and update them. The agent then builds the interface and opens it in the side panel, without requiring the user to create files or configure the layout manually.
The Interface Is an Editable Starting Point
GitHub does not treat the initial output as a final design. The user can request the addition of columns or filters, retrieve open pull requests, or turn the interface into a daily task list. The agent reshapes the Canvas based on these instructions, and it can later be saved as a reusable extension.
The extension can be kept inside the project to share with the team, or saved as a personal extension. GitHub also points to ready-made extensions in the Awesome Copilot community, including tools for release notes, Kanban boards, and issue-triage workflows, which can be installed and then customized.
Direct Collaboration Between the User and the Agent
The main practical difference is that Canvas allows both parties to modify the shared state directly. When a button is pressed, a field is updated, or a card is moved, the change appears in the interface and becomes visible to the agent without a separate synchronization step. Likewise, the agent can be asked to use the same functions available to the user to add a release note or move a card, with the result appearing in the interface.
What Changes in Practice?
Canvases offer a model that goes beyond interaction based on sending a command and waiting for a response, turning the interface into a shared workspace that the user and the agent can modify. This makes the feature suitable for workflows that require visual elements and an updateable state, rather than text answers alone. However, the material does not specify availability details, limits on the types of supported data and operations, or requirements for accessing this functionality; therefore, its suitability for teams and production environments still requires practical verification.
GitHub suggests starting with a simple board or checklist, then improving it gradually by answering three questions: What information should be displayed? What does the user want to change directly? And what should the agent be able to update or execute?