The AI agent accepts both text and images as input. To complete tasks, the CUA processes raw pixel data of the screen and uses a virtual keyboard and mouse to execute actions. OpenAI claims it can ...