GitHub Copilot introduces canvases for building interactive AI experiences

Canvases create persistent workspaces for multi-turn interactions

GitHub introduced canvases as a new interaction model within Copilot, allowing developers to build persistent visual workspaces where users can explore information, visualize workflows, and take actions across complex tasks. The feature moves beyond single-turn code generation into multi-step, iterative experiences.

Developers can create custom canvases that display structured data, accept user input, and maintain state across multiple AI interactions. GitHub positions this as infrastructure for building richer developer tools and internal applications powered by language models.

Technical implementation uses React components and API extensions

The canvas system uses React components that developers can customize and extend. GitHub provides APIs for handling user interactions, managing canvas state, and coordinating between visual elements and underlying AI model calls.

Early examples demonstrate canvases for visualizing code architecture, managing task lists, and building custom debugging interfaces. The feature requires GitHub Copilot access and works within the company's existing extension ecosystem.

Strategic move toward agentic and workflow-centric AI

Canvases reflect broader industry movement toward AI interfaces that support longer, more complex interactions rather than isolated queries. Companies including Anthropic and OpenAI have introduced similar concepts, framing them as steps toward more capable AI agents.

GitHub's implementation focuses specifically on developer workflows, where multi-step tasks like refactoring, debugging, and documentation often require sustained context and iterative refinement. The feature competes with standalone AI coding tools that already offer visual workspace capabilities.