Cursor Agent can now create dashboards: Canvas turns AI-generated text into interactive interfaces

According to Supervise Beating monitoring, Cursor 3.1 introduces the Canvas feature. After the Agent completes a task, the output is no longer just code or markdown text; it can now directly generate an interactive React interface containing charts, tables, a diff view, or custom logic. The Canvas, as a persistent artifact, resides in the Agents Window, alongside the terminal, browser, and version control.

The core change lies in information density. Taking incident response as an example: previously, the Agent would organize time-series data from Datadog, Databricks, and Sentry into a markdown table, which was hard to interpret and required additional visualization. Now, the Agent directly consolidates data from multiple sources into the same Canvas chart. Similarly, for PR reviews: the Agent can group changes by importance, generate pseudocode for complex algorithms, instead of laying out all the diffs.

The Cursor team also used Canvas in their Eval analysis. Previously, engineers had to manually look up request IDs to investigate failure patterns, at one point considering developing a standalone web app. They ultimately decided to create a Skill directly within Cursor to allow the Agent to read all rollouts, classify failure reasons, and generate an interactive analysis interface. This workflow significantly reduced the troubleshooting time for the team during the recent rollout of two new models.

Canvas supports custom extensions: the Docs Canvas Skill, available on the Cursor Marketplace, allows the Agent to generate interactive architecture diagrams for any code repository. Users can also write their own Skills to define the types of Canvas the Agent should generate.