A JOURNAL OF ARTIFICIAL INTELLIGENCE
my ai daily

Anthropic launches Claude Duet, a personal agent that works while you are away

The research preview brings morning briefs and background tasks to US Pro and Max subscribers through a waitlist. Official documentation spells out the approval rules, shared usage limits and features still missing.

Anthropic’s official Claude Duet launch artwork, showing the product name and orange app icon on a cream background.
Image: Anthropic ↗Official source image.

THE STORY IN THREE POINTS

  • Claude Duet brings proactive briefs and background work to a separate app and the web.
  • Access starts with invited US Pro and Max subscribers; background activity counts against their existing limits.
  • Messages, payments, signatures and deletion require approval, with important restrictions during the research preview.

A personal agent moves beyond the open chat

Anthropic announced Claude Duet on October 21 as a research preview of a proactive personal agent. Its release notes confirm the launch, while the product page lists standalone iOS and Android apps and access through claude.ai. This report was published on October 22; that is distinct from the date of the original announcement.

The service runs on Claude Opus 5.5. Anthropic positions it for professionals and small business owners who want help keeping track of work across connected applications. The capabilities described here come from official material, rather than our own hands-on evaluation.

Morning briefs, then work in the background

Duet prepares a morning brief and checks connected accounts every few hours, according to the announcement. Users can hand it tasks through text or voice, from scheduling a meeting to preparing a dashboard. Its own cloud computer and browser let work continue when the user closes their laptop or phone.

Existing Claude memory, skills and connected apps carry over. The getting-started guide describes separate Tasks and Make tabs: one tracks work needing attention, while the other holds outputs such as reports and presentations. These are organized deliverables the user can review.

Who can use it, and what it costs

The product FAQ limits the initial preview to Pro and Max subscribers in the United States who receive a waitlist invitation. Team and Enterprise plans are not included yet. Joining the list does not guarantee an access date; the setup guide tells users to use the email associated with their Claude account.

Duet is included at no extra charge during the preview, but it shares the subscription’s usage allowance. Morning briefs, periodic checks and larger jobs all consume that allowance. Having a paid subscription therefore does not mean unlimited background execution.

Approval comes before consequential actions

Anthropic says Duet shows the proposed action and waits before sending a message, making a payment, signing a document or deleting something. Some actions, including securities trading and transfers between accounts, remain prohibited even with approval.

The separate Link by Stripe guide makes the payment boundary more concrete. Each purchase requires approval of an amount in Link and uses a one-time card. Link is distinct from the Stripe business connector, requires a US payment method and does not support recurring payments. A merchant can still block checkout after approval.

The help documentation sets narrower expectations

The getting-started guide says users cannot choose another model, watch Duet’s browser live or take over that browser. It cannot access all projects or local computer files, and it can read work in Claude without editing it. It does not place phone calls or send SMS messages to other people.

Connector availability also needs checking. Although the product page displays a Microsoft / Office 365 logo, the help guide explicitly says Microsoft 365 is not available yet and that Slack has limitations. That operational guidance is the basis for this report’s availability description.

The product FAQ warns that Duet can misunderstand a thread, misjudge urgency or miss an item between checks. Its proactive workflow is therefore not a guarantee of immediate monitoring or correct prioritization.

Editorial context: judge the handoff, not just the brief

For a personal agent, a useful evaluation would follow a task all the way through: what information it found, what it prepared, which decision it brought back and whether the final result matched the request. A polished morning summary establishes little about those later steps.

An early trial could start with a bounded job, such as preparing a meeting brief from selected material. Reviewing its sources, omissions and use of the subscription allowance would reveal whether the background work saves time. More consequential tasks would need equally clear evidence that the proposed action is correct.

Sources: Anthropic: Introducing Claude Duet — official launch announcement ↗ Claude Duet: product FAQ and waitlist ↗ Claude Help Center: Get started with Claude Duet — access, usage and limitations ↗ Claude Help Center: How to use Link by Stripe with Claude Duet ↗ Claude release notes: October 21 research preview ↗

02The next story

All stories

Follow your curiosity.

97 STORIES TO EXPLORESOURCES & CONTEXT INCLUDED