Anthropic rolls out “Claude Duet” phone agent that proactively handles email and shopping, upping pressure on rivals like Meta and OpenAI

Anthropic has launched Claude Duet, a consumer-focused artificial intelligence agent designed to take actions on a user’s behalf rather than waiting for a direct prompt. Coverage of the launch characterizes the product as capable of reading a person’s email and handling shopping tasks before the user explicitly asks, placing it closer to the “do it for me” model of software than the traditional chat interface.

The announcement indicates that the biggest AI-company race is shifting from model performance to product workflows that fit daily routines. Instead of responding to questions in a single turn, an agent is expected to monitor context, propose steps, and then execute actions tied to personal information. In practice, that means consumer trust, permissions, and safety become as important as language ability.

For Meta, the launch is notable because the company has already staked out similar territory in consumer AI assistance. The report notes that Meta has claimed a user base of 10 million for its own consumer assistant push, underscoring how quickly this competitive front has expanded from enterprise tools to mass-market features.

The consumer agent concept also raises a different set of competitive requirements for large platforms. To be useful, an assistant has to integrate smoothly with everyday apps like email and commerce, or at least coordinate with them. It also has to keep users in control, including what the system can access and what it is allowed to purchase or schedule without additional confirmation.

Meta’s public communications on its newsroom emphasize ongoing work around AI products and experiences, but the provided material does not include specific details about what Meta will do next in response to Anthropic’s Duet release. Still, the market context is clear: a more proactive agent makes consumer expectations harder to contain, because users can begin to think of the assistant as a continuous background worker rather than a tool that only answers when called.

It is also not yet clear, based on the information provided here, what constraints Anthropic’s agent uses in different scenarios, such as how it handles ambiguous requests, limits spending, or escalates decisions for approval. The launch coverage points to broad capabilities like email reading and shopping execution, but does not outline the full set of safeguards, approval flows, or limits in the material available for this story.

Looking ahead, the next developments likely center on adoption and product breadth. As agent assistants move into tasks that involve personal data and spending, companies will be judged on reliability, user-perceived usefulness, and the clarity of permission and consent mechanisms. Watch for additional details on Claude Duet availability, supported actions, and how users can manage or revoke access when the assistant is operating proactively.