Interactive Canvases in Agents Window
TL;DR
Cursor now supports interactive canvases that allow agents to create dashboards, diagrams, and custom interfaces using first-party components like charts and tables. These durable artifacts live in the side panel for easy reference during development.
What changed
Cursor's Agents window now renders interactive canvases directly inside the chat -- charts, diagrams, and visual scratchpads update as the model reasons.
In practice that means you can ask the agent to build a mental model, and the canvas shows the structure evolving turn by turn rather than dumping a final answer. This is Cursor's bet that developers want to see the reasoning, not just the output.
Paired with the new tiled-layout release from April 13, the Agents window feels closer to a working IDE surface than a pop-out chat.
Who this matters for
- **Basic User:** You can now ask Cursor to plan a project and literally watch the diagram form -- lower cognitive load than reading a wall of text.
- **Vibe Builder:** Interactive canvases let you iterate on architecture visually before committing to code -- faster loop from idea to scaffold.
- **Developer:** Canvas state is exposed to tools, so custom MCP integrations can render side-panels without hacking the chat UI.
What to watch next
This is the first real UX win for agent-style coding in 2026 -- the canvas makes the model's reasoning auditable. Watch whether Codex and Claude Code's IDE extensions match within a month; if they don't, Cursor's pulled ahead of the Copilot-style baseline.