Anthropic launches Claude Duet proactive agent for Pro and Max

Anthropic is positioning Claude Duet less as a chatbot and more as a digital secretary that already lives inside the Claude account users have. Because Duet draws on the same memory, skills and apps as Claude, there is no extra setup, and the running history stays in one thread. That continuity may make the agent feel like a teammate rather than a separate tool, which fits Anthropic's stated goal of letting users message it by text or voice with no prompts.

The launch also leans on Anthropic's existing trust framework, the same one behind Claude Code and Claude in Chrome. An automatic classifier screens the agent's actions for sensitive information leaks and checks with the user before important steps. Anthropic appears to be using its own staff as early testers: Claude Code's Boris Cherny uses Duet on walks to talk through to-dos, and team lead Karan Ahuja uses it to schedule user interviews and ship code.

Still, the company frames Duet as an early, experimental version. It may misread a thread or miscalibrate what is urgent, and because it is designed around work, personal errands such as booking tickets are not treated as urgent. Anthropic says the research preview will help it understand use cases and failure modes, and early-user feedback will shape how the agent works.