Cursor cloud agents can leave a canvas your team can inspect

You ask a background coding agent to investigate a project, then receive a long report that is difficult to compare. Cursor’s October 16, 2026 update lets cloud agents create interactive canvases as part of their work, giving a team a view to inspect after the run finishes. Original changelog.

A useful view still needs traceable numbers

Canvases can organize an analysis into sections, tables and charts beside the conversation. They are saved so you can reopen the result and ask for revisions. A published share captures the view for teammates with read-only access. Refreshing that share is a separate action; a shared snapshot is not proof that its underlying information has been refreshed. Canvas workflow.

For an illustrative first report, ask for unresolved launch checks from a small project: the evidence, current status and next action for each item. Verify one completed item and one unresolved item against the actual files or check output. If the chart says “ready” but its source still contains a blocker, fix the report before distributing it. We have not run this example.

Check who can open the finished view

Sharing requires an eligible paid plan, membership in a team and a privacy mode that allows data storage. Legacy Privacy Mode blocks sharing, and an administrator can disable it. The documented recipients are teammates, so do not plan an external client handoff around an anonymous link. Sharing requirements.

Current pricing lists cloud agents from Premium upward and bills third-party models separately. Cloud-agent documentation also ties usage to the selected model and notes that larger contexts can increase cost. A polished canvas is not a fixed-price report. Current plans, cloud-agent billing.

Use our Cursor guide for the wider editing workflow and AI output evaluation guide to define the report’s acceptance checks. Start with a result your team can verify and actually open.