Anthropic announces Claude Duet, a new proactive AI agent

Anthropic has been building Claude as a general-purpose AI assistant, and Claude Duet is a newer product in that line aimed at taking actions rather than only responding to prompts. The October 21 release places it in a fast-moving market where AI firms are racing to add more agent-like behavior to their products. That matters because this kind of feature shifts the service from drafting and answering toward acting on behalf of a subscriber.

For people who use Anthropic’s Pro and Max plans in the US, Claude Duet adds a workflow for shopping and purchasing that is tied to approval steps rather than fully autonomous spending. The feature’s handling of card numbers and government IDs points to a design that tries to keep sensitive data out of the agent’s direct view. Builders integrating AI agents into commerce will need to think about checkout flows, permission checks, and what data an agent can actually see.

Claude Duet sits at the intersection of convenience and control, which is where agentic AI is drawing attention from both product teams and privacy-minded users. A system that can buy items for a user also raises the question of how much trust belongs in the software layer, especially when that software sits between a customer and a wallet or store. The tension here is not just about productivity; it is about how much agency people hand to an AI system when money and identity are involved.

Still open is how broadly Anthropic will roll out Claude Duet beyond the initial US Pro and Max subscriber group, and how the waitlist will be handled over time. The release notes do not settle how widely the product will be available, what exact limits apply, or how the preview will evolve after launch. Those details will determine whether Claude Duet stays a narrow test or becomes a more general shopping tool.

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.