Computer use.

Computer use is when the model gets its own computer, or takes control of yours. As Remy puts it, the model "looks at the screen, decides what to click or type, does it, checks the result, and repeats." It works a screen the way you would.

That makes it powerful, because anything with a screen is now something your agent can operate. It's also the slow, clumsy option, so a good agent saves it for last. It checks for Connectors first and uses an app's own API when one exists. Then it tries a browser. Full screen control is the last resort, for when nothing cleaner is available.