A new study by Singapore’s National University Show Lab evaluates Anthropic’s Computer Use mode, which enables the AI model Claude to interact with computers through standard user interfaces. According to VentureBeat’s Ben Dickson, the system shows promise in handling complex tasks involving web searches, workflow automation, and office productivity but frequently makes basic mistakes humans would avoid. The study assessed Claude’s abilities across planning, action, and self-evaluation dimensions. While the AI successfully coordinates between applications and follows multi-step processes, it struggles with simple operations like scrolling webpages or formatting text. The researchers conclude that current GUI agents cannot fully replicate human computer interaction patterns and require significant improvements before enterprise deployment.