Anthropic 推出主动式 AI 智能体 Claude Duet

Anthropic PBC Wednesday introduced Claude Duet, a proactive artificial intelligence agent for professionals and small business owners in research preview, which completes tasks when asked, keeps working behind the scenes and checks in.

The company said the new assistant will send a short brief each morning of the most important items across connected apps and keep tabs on messages only when something needs attention. It can also handle numerous different administrative tasks, like a digital secretary. These include scheduling meetings and more complex tasks, such as building dashboards with interactive tools, creating financial or sales forecasts and prototyping and building web apps.

Users will message Duet like another teammate, by text or voice, with no prompts or setup. The idea is to use it like a secretary who can access apps, calendar, email and other connected information. It’s part of Claude, so it also has a running history in one thread. It remembers what a user said before, and returns finished work like a doc, sheet or deck for review.

Duet is part of the Claude that users already participate with. It can access memory, skills and apps from Claude – so there is no additional setup.

Anthropic stressed that Duet is built on the company’s framework for trustworthy agents, which also guides the company’s existing agentic systems such as Claude Code and Claude in Chrome. An automatic classifier approves any actions the agent takes by checking messages for sensitive information leaks and double-checking with the user before taking important actions.

The company added that people at Anthropic are using Duet themselves. Claude Code’s own Boris Cherny uses it on walks to talk through to-dos and save time at his desk, and the team behind Duet uses it to build the product. Karan Ahuja, who leads the team, works with it to schedule user interviews and even ship code.

The research preview will help Anthropic better understand Duet’s use cases and failure modes. Right now, it is an early, experimental version. It still might misread a thread or miscalibrate what is urgent: The agent is designed around work, so personal errands, such as booking tickets, won’t be treated as urgently and it can’t show its browser. The company will use what it learns from early users to fine-tune how it works.

Image: Anthropic