Cursor extends its agents from code and browsers into desktop application testing

The company demonstrates Computer Use with the iOS Simulator and allows an explicit request through the agent prompt.

Cursor announced Computer Use on October 23, extending its coding agents' ability to interact with applications beyond the browser. The company's official demonstration shows an agent using the iOS Simulator to test a feature it has built.

In its announcement thread, Cursor says agents can click and type across applications on the computer. The tool can be chosen when a task requires it, or requested explicitly through a Computer Use mention in an agent prompt. The company directs users to update Cursor to try the capability.

The workflow connects two kinds of evidence. Source code and a successful build establish properties of the implementation, while interaction with a running application can expose a control that is miswired or a sequence that does not produce the intended state. Using both can make a bug report more concrete.

A saved note provides a simple example. An agent can enter text, press Save and reopen the application to see whether the note persists. That sequence checks something a screenshot immediately after pressing the button cannot establish. The expected outcome still needs to be defined by the application's actual design.

The announcement establishes a new interaction capability, rather than proof that agents can independently certify every desktop application. Its immediate value is in bringing repeatable user actions into the same working context as coding and debugging. A bounded flow, a named build and the result observed on screen make that connection useful to the developer reviewing the work.

Sources

- [Cursor Computer Use announcement](https://x.com/cursor_ai/status/2113661506634305834) — Cursor official account