Cursor、コード生成物を視覚的に整理する Canvases を追加 Canvases

Cursor, the AI-assisted code editor, has introduced a new workspace feature called Canvases. Unlike the conventional tab and chat-history paradigm, Canvases offers a visual surface where agent outputs, code snippets, and file references can be freely arranged, marking a shift in how developers organize AI-driven work.

According to Cursor, Canvases lets users place multiple agent sessions and their generated results side by side on a single screen for direct comparison. Typical use cases appear to include lining up alternative implementations for the same task to inspect differences, or grouping related files together to get a holistic view of a change. The feature can be read as an attempt to translate the kind of parallel thinking that linear chat threads struggle to accommodate into a spatial layout.

The broader context is that AI coding is moving away from one-off question-and-answer exchanges toward long-running autonomous work carried out by multiple agents in parallel. Cursor has already shipped Background Agents and the ability to run several models concurrently, and as the volume of artifacts produced by those workflows grows, the need for a way to organize them grows with it. Canvases may serve as that organizational layer, providing a home for the outputs that would otherwise be scattered across tabs and conversations.

Similar concepts have emerged elsewhere in the industry. OpenAI's ChatGPT Canvas and Anthropic's Artifacts both detach generated content from the chat stream and place it into a separate, editable surface. Cursor's Canvases, however, appears to put more weight on editor integration and on the spatial arrangement of multiple concurrent tasks, an approach closer in spirit to a whiteboard than to a document editor. It also aligns with a broader trend of infinite-canvas interfaces — exemplified by tools such as tldraw and Excalidraw — making their way into developer tooling.

In practice, the feature looks well suited to review workflows where multiple changes need to be compared, to exploring alternative refactoring strategies, and to capturing design discussions in a form that survives beyond a single chat session. Details on sharing, persistence, and collaboration behavior are not fully clear from the announcement, and developers interested in adopting Canvases are likely to want to consult the official documentation for the specifics of how state is stored and how canvases interact with existing Cursor primitives such as workspaces and agents.

Whether Canvases becomes a central part of the Cursor experience or remains a complementary view will likely depend on how well it integrates with the editor's existing agent and indexing infrastructure. If it succeeds in making parallel agent work easier to manage, it could influence how rival AI coding environments think about presenting multi-agent output. For now, it stands as another data point in the ongoing reshaping of the IDE around AI agents rather than around files and tabs alone.