The AI developer Anthropic, backed by Amazon and Google, has endowed its artificial intelligence Claude with the ability to perform actions on a computer—including internet searches, clicking buttons, and entering text. With this expansion, the company aims to create virtual agents that can automate repetitive tasks.
The new feature called "Computer Use" was introduced to developers on Tuesday. It enables Claude to act like a "human assistant" with users' consent: browsing websites, scheduling calendar entries, or filling out forms.
This kind of repetitive tasks that people absolutely hate, I call the automation of life's drudgery," said Mike Krieger, Chief Product Officer at Anthropic, to the Financial Times. "Currently, Claude can accomplish things that would normally take an hour in two minutes. We aim to enable people to focus on the creative part of tasks that is fun and leave the repetitive and less exciting things to Claude.
Anthropic's step is part of a growing trend in the industry to develop AI models that can function as virtual agents. Competitors like OpenAI, Microsoft, and Meta are working on similar technologies. Just on Monday, Microsoft announced the ability to create autonomous agents in its 365 product suite.
The goal is to create AI agents that handle everyday tasks like expense reports or travel bookings—a potentially lucrative market for AI companies to generate revenue from their powerful but expensive models.
A demonstration from Anthropic shows how a user asks Claude: "My friend is coming to San Francisco, and I want to see the sunrise with him tomorrow morning. I am coming from Pacific Heights. Could you find us a great viewpoint, check the travel time and sunrise time, and then create a calendar entry that gives us enough time to get there?
Claude researches suitable locations and sunrise times online, calculates the route and travel time, and creates a calendar entry with all the details.
The software uses computer screenshots to interpret content and can click buttons or enter text—regardless of whether it's a Mac or PC. It has real-time internet access and is available to developers for creating specific applications.
Anthropic emphasizes that the technology is still in an early experimental phase. The company acknowledges that the model may use unreliable information from the internet and is currently examining how these features can be transferred to other devices such as mobile phones.
The company is working on a consumer-oriented product based on this technology and aims to strengthen users' trust in its capabilities. Krieger compares the technology to self-driving cars, as people do not yet fully trust them with complete control. "I would say that this is more of a metaphor at the moment, rather than entrusting it with absolute full autonomy over longer periods," he added.
I see part of our mission as being a helpful guide for the future of AI in a human-centered way," said Krieger. "That's why we want to implement this in our products with computer control.
