Anthropic, a startup backed by tech giants Alphabet and Amazon.com, has introduced updated artificial intelligence models with a new feature that automates computer tasks. This “computer use” capability allows the AI to understand where to move the mouse, where to click, and what to type to accomplish complex tasks, according to Jared Kaplan, Anthropic’s Chief Science Officer.
Designed specifically for software developers, this feature represents a significant step toward AI agents. These agents are programs that can perform multi-step actions with minimal human intervention. Researchers consider AI agents the next frontier in AI development, moving beyond chatbots that can generate text or code but cannot execute actions.
Anthropic showcased how this feature can be used to code a basic website and plan a sunrise outing using tools like Google Search and Apple Maps.
The company offers three versions of its AI model family, Claude, catering to different performance needs and price points. The latest updates apply to Sonnet, the mid-tier model, and Haiku, the most affordable option. The new 3.5 version of Haiku can generate computer code almost as well as the version of Sonnet released in June, Kaplan noted. CEO Dario Amodei previously mentioned plans to update Opus, the most advanced model, by the end of the year.
Currently, the “computer use” feature is limited to the new version of Claude 3.5 Sonnet and includes safeguards to prevent misuse such as spam, fraud, and election interference. Kaplan acknowledged that the AI still makes mistakes.
Mike Krieger, co-founder of Instagram and now Anthropic’s Chief Product Officer, said the company is seeking feedback from business customers to determine where to focus development efforts. An internal labs team is also exploring how to make this capability available to consumers, a feature Krieger personally wants.
“I was booking flights,” Krieger shared. “I really just want this to be completely automated.”
