Claude Duet: Anthropic's Proactive AI Agent for Work

Who Can Try Claude Duet and Where It Works

The Verge and VentureBeat report that Claude Duet is available in the United States to Claude Pro and Max subscribers through a waitlist. They describe access through Claude on the web and standalone apps for iOS and Android. Anthropic has not been reported as making the preview broadly available to all users.

The publications also identify Claude Opus 5.5 as the model behind Duet. Because access is limited, the research preview offers an early look at the agent rather than a general release. Its availability and features may change as Anthropic tests the product.

How Duet Handles Connected Apps and Background Tasks

Duet is built to use connected services and a remote computing environment with a browser to complete multi-step work. VentureBeat reports that examples include preparing morning briefings, checking invoices and creating reminders, following up with customers, and building financial forecasts or web applications. The report names QuickBooks, Stripe, HubSpot and Docusign among the services involved in example workflows.

The agent can continue working in its remote environment after a user closes a laptop or turns off a phone, according to VentureBeat. That lets a task run without the device remaining open. Duet is also described as having memory and reusable skills, which could help it handle recurring work with less repeated setup.

These capabilities move the product beyond a conventional chat exchange: a user can delegate a task, then return to a result or an update. But the examples should be understood as reported preview capabilities, not a guarantee that every connected app or workflow is available to every user.

What Users Should Know About Monitoring and Approvals

“Proactive” does not mean constant surveillance. VentureBeat reports that Duet checks connected applications periodically, rather than monitoring them continuously in real time, and can surface developments it finds. Users should therefore not treat it as an immediate alert system for every change.

There are also approval safeguards. The Verge reports that Duet asks users to approve significant actions, including sending a payment or deleting something. That distinction matters: an agent may prepare or organize work, but consequential actions can still require a person’s confirmation.

Claude Duet places Anthropic in a growing field of agents intended to take actions, not just generate responses. Its practical test will be whether users find its background work useful, its app connections reliable and its permission controls clear. For now, it remains a limited research preview, with rollout, supported apps and monitoring behavior worth watching as testing progresses.

Follow our coverage for updates on Claude Duet’s rollout, supported apps and how it compares with other AI agents.

Frequently Asked Questions