Cursor Cloud Agent Canvases: Deciding Project Direction With Clients

I use Cursor Cloud canvases in ongoing projects, in the areas where I need to make decisions together with my clients. A canvas lets me evaluate the direction of the project interactively.

That is my impression from working use rather than a measured test, and I did not record the dates. The rest of this note describes what Cursor documents about canvases and sharing.

What changed

Cursor's changelog entry on cloud agent canvases says cloud agents can now build interactive canvases as part of their work, and that a finished canvas can be shared with a team by sending its link. The same entry introduces an Environments API in beta and says cloud agent environments now support up to 300 repositories, with large environments loading five times faster.

Cursor's canvas documentation describes a canvas as an interactive artifact that renders next to the chat, laid out with sections, stats and tables. An agent can open its result in a canvas when that fits better than markdown, or you can ask for one directly. Canvases are saved, so they can be reopened, edited and rerun with fresh data.

Why it changes the handoff

The earlier Field Note on Cursor Cloud Agents is about a handoff you can inspect: a branch, a pull request and a record of what the agent did. A canvas adds a different kind of output. For a dashboard, an audit or a report, the deliverable is not code, and a link to a live snapshot is easier to pass to a teammate than a chat transcript.

According to the documentation, publishing happens from the canvas toolbar, and anyone on the team can open the link in a browser with read-only access. Common canvas layouts can also be packaged as skills, so a short prompt regenerates the same structure with new data.

My own use is about deciding things with clients. Because a shared link opens only for members of the publisher's team, this note does not test how a canvas reaches someone outside the team.

What the documentation limits

Sharing has conditions. The documentation lists shared canvases on paid plans, including Premium and higher, Teams, Enterprise and legacy Pro, Pro+ and Ultra, and says free accounts cannot create shares. Because only team members can open a share, the publisher needs to be on a team, and sharing requires a privacy mode that allows data storage. Team admins can turn shared canvases off for the organization.

The Shared Canvases page lists only the canvases you published yourself, so opening a teammate's canvas requires its link. The documentation also suggests reverting and prompting again for large reworks instead of nudging a layout through many small follow-ups.

Where I would draw the line

For me, canvases are most useful at the points in an ongoing project where I need to make a decision together with a client and want to review its direction interactively. I have not compared this with working from the raw data or a pull request, so I would not treat a canvas as a replacement for either.

A useful next test would record the plan, the task, how many follow-up prompts it took to get a correct canvas, and whether the numbers in the canvas matched the underlying query.