Intel / blog

GitHub Copilot app for Beginners: How to build custom workflows with canvases

Source: github.blog
Author: Kayla Cinnamon
Date: Published Sep 25, 2026

Context

A canvas in the GitHub Copilot app gives the user and a shared interface for a workflow. In the release-notes example, the user describes the workflow, what they should be able to do in the interface, and what the agent should be able to do. The agent builds the interface, which can then be refined and saved as an extension for reuse.

The useful difference is shared state: a person can edit a field or move a card and the agent sees the change without a separate send or sync step, and the agent can use the same actions to update the interface. GitHub's earlier canvas overview describes the same feature as a durable shared surface, separate from chat, that keeps workflow state explicit so people can inspect and steer the work.

Key takeaways · AI-distilled

Canvas state is shared: when you click a button or move a card, the agent sees the change immediately with no separate send or sync step, and it can use the same actions the canvas offers you.
Canvases are built with the /create-canvas and a plain-English prompt; GitHub advises covering the workflow, what you should be able to do in the interface, and what the agent should be able to do.
A finished canvas is saved as an extension that can live with the project for a team or stay personal, and ready-made ones such as kanban boards and issue triage boards are shared through Awesome Copilot.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.

Read the source github.blog