All news
AI agentsClaude by Anthropic · · 3 min read

Anthropic introduces Claude Duet, a proactive agent

Duet works in the background from its own cloud computer and asks before doing anything irreversible.

Anthropic introduces Claude Duet, a proactive agent
Image: Claude by Anthropic

What Duet is

Claude Duet is a new app, in research preview, in which Claude works as a proactive agent. It points out what needs your attention, does the tasks you hand it, keeps working while you are away and checks with you before acting on your behalf. Anthropic is aiming it at professionals and small business owners with more to track than they have time for.

Tasks range from simple to substantial: scheduling meetings, building dashboards, financial or sales forecasts, prototypes and web apps. You describe the job in a short message and Duet works out which apps to use and what steps to take.

A morning brief, then quiet

Each morning Duet sends a short summary of the most important things across your connected apps. During the day it checks those apps every few hours, rather than constantly, so it doesn't use up plan limits, and messages you only when something needs attention. You can tell it what kinds of updates you want and how often. Anthropic says it is still tuning when Duet decides to interrupt.

How it works

Duet runs on its own cloud computer with its own browser, so your laptop or phone can be off while it works. It picks up the memory, skills and connected apps from your Claude account, so there is little to set up. Everything happens in one ongoing conversation, by text or voice, whether the request takes a minute or a week, and you are notified when a task is finished or needs review.

It can use more than 6,000 connectors from the Claude Marketplace. Anthropic's examples include finding missed Gmail and Slack messages, chasing unpaid invoices in QuickBooks or Stripe, preparing payroll in Gusto, sending agreements through DocuSign, drafting follow-ups from HubSpot history and running a Shopify store.

What needs your approval

Duet can read, draft and prepare work on its own. Before it sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it plans to do and waits for a yes. Some actions are blocked even with approval, including trading securities and moving money between accounts. You can disconnect any app at any time.

When it buys something, it uses Link, Stripe's wallet, which requires your approval and pays with a virtual card so your real card number is never shared.

Security and privacy

  • Duet never sees your passwords or sign-in tokens directly.
  • A classifier checks each action to stop steps you didn't ask for.
  • Defences against prompt injection come from Anthropic's browser-agent research.
  • Highly sensitive data, such as card and ID numbers, is never sent; health or financial details are sent only on your explicit instruction.
  • It was penetration-tested by internal red teams and an outside firm, and is covered by Anthropic's bug bounty.

Limits and access

Anthropic is upfront that the early version can misread a thread or misjudge how urgent something is, is weaker at personal errands than at work, and can't yet show you its browser. Background checks and larger tasks count towards your plan's usage limits.

Duet is available by waitlist to US users on the Pro, Max and Ultra plans, on iOS, Android and claude.ai, and runs on Claude Opus 5.5.

Why it matters for your business

Asking a person before anything irreversible is the right default, and it is notable that one of the largest AI companies has made it the core of a product. Duet draws the line where we do: reading and drafting are free, while sending, paying, signing and deleting need a yes.

Its limits are worth noting too. Some actions are blocked even with approval, and it checks in every few hours rather than constantly to save usage. Those are exactly the choices every business has to make when it lets an agent act for it.

Duet is built for one person's inbox and to-do list. When the work spans a team, or has to run inside your own systems and rules, the same approach applies: approval steps, credentials the agent never sees, and a full log.

What to do next

  1. Decide which actions an agent may never take in your business, even with approval
  2. Keep sign-ins in a password manager or vault, not in agent instructions
  3. Pilot proactive agents on reading and drafting before sending or paying
Talk to us about it
Free · 30 minutes · No obligation

Want this working in your business?

We map your most repetitive processes, show what can be automated with AI and n8n, and estimate the hours you get back. You leave with a plan, whether you hire us or not.