Cursor announced Computer Use on 23 October: its coding agents can now operate desktop applications directly, with the iOS Simulator given as the headline example. An agent that writes a feature can open the app, drive the interface and check the behaviour, rather than stopping at a successful build.
That is a meaningful change in what a coding agent can verify about its own work. Until now the feedback an agent could gather was mostly textual - compiler output, test results, logs. Clicking through a running application covers the class of failure that compiles cleanly and behaves wrongly, which is most of the interesting ones in mobile development. For iOS specifically, where the simulator is the normal test surface, it removes the step a human was previously required for.
It also lands the same week as a Shandong University study finding that harness complexity drives cost up to 3x without a consistent accuracy gain. Computer Use is not scaffolding in that sense - it adds a capability rather than rearranging context - but the cost question applies. Driving a GUI means screenshots, and screenshots mean tokens.
Sources
Cursor · Flavio Copes