Anthropic announces Claude Duet, a new proactive AI agent

Claude Duet sits in Anthropic’s broader move from a chat assistant toward an agent that can keep working after a user steps away. The launch arrived as a research preview on October 21, with access starting for US Pro and Max subscribers. That puts the product in a familiar AI rollout pattern: limited release first, then wider access if the preview holds up in use.

For people who use Claude Duet, the practical change is that the agent can help with purchases while Anthropic sets approval steps around spending. It is designed to shop online through Stripe’s Link wallet and complete checkout with Shop Pay, while leaving each spend for the user to confirm. The credential and ID limits mean the agent is framed to act without seeing credit card numbers or government IDs.

Agentic AI is moving the line from suggestion to execution, which raises the stakes for trust, access control, and user consent. Anthropic’s approval checks and credential isolation show one way builders are trying to keep an autonomous system from becoming a free-ranging account holder. The trade-off is clear: more automation can save time, but it also increases the need to define exactly what the software may touch and when a human must still step in.

The move was larger before the first headline than after. That is an observation, not an explanation: the market also moves on things nobody publishes.