Cursor agents can now control your computer to test the apps they build

Cursor's AI agents can now take control of a user's computer and drive desktop apps such as Apple's iOS Simulator to test the features they build, the company said on X on 23 October 2026 [1]. The new Computer Use tool lets agents click and type across any app, not just the browser [1].

What does Computer Use in Cursor do?

Cursor says its agents can now control a user's computer and drive desktop apps such as the iOS Simulator. The company's announcement frames this as a way for agents to test the features they build, rather than handing back untested code [1].

Cursor says agents use Computer Use on their own when a task needs it, or when a user @-mentions it in a prompt. Users get the feature by updating the Cursor app, according to the announcement [1].

What does the demo show?

In a 21-second demo video, a user in a project called "hiking app" asks an agent running on "This Mac" to add the ability to favorite trails and demo it [1]. The agent edits the app, launches it in an iPhone simulator and taps through it while a banner reads "Cursor is using Simulator" [1].

How is this different from Cursor's earlier computer use?

Before this update, Cursor documented computer use for cloud agents' virtual machines and self-hosted workers. In February the company said it gave cloud agents their own virtual machines so they could test their changes and record videos, screenshots and logs [3].

Cursor's documentation describes computer use on self-hosted macOS and Linux workers as an explicit opt-in started with a command-line flag [2]. On a Mac, a helper app called Cursor Computer Use needs the Accessibility and Screen Recording permissions before the agent can click or take screenshots [2].

That route took setup. In a September forum thread asking for iOS Simulator support, a Cursor staff member said there was no built-in Simulator panel in the Agents Window and pointed users to the self-hosted worker [4]. The new posts describe agents working on the user's own computer after an app update [1].

Our read

Opinion: Letting the agent run the app it just changed closes the most annoying gap in AI coding, the part where you find out it does not actually work. The trade-off is trust: an agent that can click anywhere on your own machine deserves clear, visible limits, and we found no published rules for local use yet.

What we could not verify

• Which operating systems and Cursor versions support Computer Use for local agents; Cursor's posts do not say, and the self-hosted documentation covers only macOS and Linux workers.
• Which plans include the feature and whether it is rolled out to all users; Cursor's announcement only says to update the app.
• How reliably agents test apps in practice; the only demonstration we saw is Cursor's own 21-second promotional video.
• What permissions and safeguards apply on a personal computer; Cursor's published permission steps describe self-hosted workers, and no changelog entry for the launch was online at the time of writing.