Most AI apps wait for you to type. Claude Duet is Anthropic's attempt at the opposite: an agent that looks at your inbox, calendar and business tools, tells you what needs you, and does the work in the background. It's out now as a research preview, by waitlist.
What it does
Anthropic says Duet is built around Claude Opus 5.5 and aimed at professionals and small business owners. In practice:
- A morning brief. Each morning it sends a short summary of the most important items across your connected apps.
- Check-ins during the day. It checks your connected accounts every few hours, not nonstop, because constant checks would burn through your usage limits. It messages you when something needs you, and you can tell it what you want to hear about.
- Background work. Duet has a virtual machine with its own browser, so your laptop and phone don't need to be on. You describe a task in one message and get a notification when it's done or needs a review.
- One thread. You talk to it by typing or by voice, in a single ongoing conversation.
It carries over your memory, skills and connected apps from your Claude account. If you've already built agent skills, Duet can use them.
What it connects to
Anthropic says Duet can use the more than 6,000 connectors and plugins in the Claude Marketplace. Its examples include finding missed Gmail and Slack messages, checking QuickBooks or Stripe for unpaid invoices, running payroll in Gusto, sending agreements through Docusign, drafting follow-ups from HubSpot, and building a Shopify store.
For purchases, it uses Link, Stripe's digital wallet. Link asks you to approve the spend, then pays with a virtual card so your real card number isn't shared. It can also check out with Shop Pay. Signing in with logins saved in 1Password is listed as coming soon.
The guardrails
This is the part worth reading twice, because it's the template a lot of agent products will copy:
- Approval before anything irreversible. Before Duet sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it's about to do and waits.
- Some things are off-limits even with approval. Anthropic names trading securities and moving money between accounts.
- A classifier for sensitive data. A new dedicated classifier checks each message and refuses to share the strictest tier of information, like credit card numbers and government ID. Health and personal-finance details go out only when you've explicitly told Duet to send them.
- No direct access to your credentials. Credential handling follows the Claude Managed Agents design, so Duet never directly sees your passwords or sign-in tokens.
- Testing. Anthropic says red teams and an external firm pen-tested it, and it's covered by the public bug bounty.
If you build agents yourself, that list (approve irreversible steps, hard-block a few, keep secrets out of the model's view) is a good checklist. We cover the same ideas for coding agents in Run coding agents safely.
The limits
Anthropic is upfront that this is early:
- It may misread a thread or get urgency wrong.
- It's built for work, so personal errands like booking tickets are weaker.
- It can't show you its browser yet.
- It lives in its own app, separate from the main Claude app for now.
Who can get it
The preview starts with Pro and Max subscribers in the US, by waitlist. It's included in those plans and uses the same usage limits as the rest of Claude. The background checks count toward them, and big jobs like building a financial model take more. If you're outside the US (Canada included), there's no date yet.
