Claude Duet: the assistant that comes to you

What Claude Duet is

Claude Duet is a new app from Anthropic in which Claude works as a proactive agent: it comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.^[1]

Anthropic announced it on 21 October 2026 as a research preview, available by waitlist, starting with Pro and Max subscribers in the US.^[1] The company’s account on X posted the same description with a 39-second video, and followed with posts on the model behind it, safety and availability.^[2]

Anthropic says it built Duet “for professionals and small business owners with more on their plates than they can track.”^[1] It is available as a standalone app on iOS and Android, and on claude.ai.^[1]

The difference from the Claude app, in Anthropic’s words, is who goes first. The Claude app waits for you to open a chat. Duet “comes to you”: it tells you each morning what needs attention, takes tasks by message, and reaches out when one is done or needs approval.^[3]

A day with it

Each morning Duet sends a short brief on the most important items across your connected apps. Through the day it keeps tabs on those apps and messages you only when something needs your attention. You can also hand it tasks of all shapes, from scheduling meetings to building dashboards with dozens of tools, creating financial or sales forecasts, prototyping and building web apps.^[1]

Anthropic’s product page runs a sample conversation in a phone mock-up. The sample shows Duet reporting on a Chicago location opening and listing items that need the user’s attention. It is Anthropic’s demonstration, not a real user’s account.^[3]

Anthropic also published short stories from testers. A broker named Jess Provencher said Duet flagged, unprompted, a showing on her calendar on the wrong day, calling it “exactly the kind of thing I want caught without having to ask.” Amin Ebrahimi, owner of Shahs of Kabob, said of using it for accounting, customer support and marketing, “I don’t need to open up 100 tabs.”^[1]

How it works

Your context from Claude is the foundation. Duet carries over memory, skills and connected apps from your Claude account, so there is little to set up, and you message it by typing or voice in one ongoing thread.^[1] It runs on a virtual machine with its own browser, which is why it does not need your laptop or phone to be on.^[1] Anthropic’s X thread adds that it was designed around Claude Opus 5.5, which gives it the judgment to decide what needs you, what can wait, and when to check with you first.^[2]

One detail shapes how you will use it. Duet “checks your connected accounts every few hours”, Anthropic says, because checking nonstop would use up your plan’s limits much faster.^[1] The product page says plainly that it can miss something that arrives between check-ins.^[3] It is proactive on a schedule, not a live alarm.

What it asks you first

Duet reads, drafts and prepares work on its own. Before it sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it is about to do and waits for your approval.^[1] Some things are off-limits even with your approval, such as trading securities or moving money between accounts.^[1] You can also disconnect any app it has access to.^[1]

How it is kept safe

An agent that reads your email and acts on it is a target. Anthropic names five layers for Duet: its framework for trustworthy agents, a credential design that means Duet never directly sees your passwords or sign-in tokens, an auto mode classifier that approves actions, defences against hidden instructions in web pages and messages, and adversarial testing by red teams and an external firm, with Duet covered by its public bug bounty program.^[1]

The framework behind the first layer has five principles: keeping humans in control, aligning with human values, securing agents’ interactions, maintaining transparency, and protecting privacy.^[4] Its authors say that as agents become more capable, the risks of misreading intent and of prompt injection both intensify.^[4]

On credentials, Anthropic’s Managed Agents post explains the design Duet says it follows. In an older, coupled design, code Claude generated ran in the same container as credentials, so a prompt injection only had to convince Claude to read its own environment. The fix was to keep tokens out of the sandbox: OAuth tokens sit in a vault, and Claude calls tools through a proxy that fetches them, so the harness is never made aware of any credentials.^[5]

On approvals, Anthropic’s auto mode post describes a classifier that decides whether an action matches what the user asked for. It was built for Claude Code, where it says users accept 93% of manual permission prompts anyway, which leads to approval fatigue.^[6] Our reading, not Anthropic’s, is that a proactive agent makes this worse, because many small decisions arrive while you are not watching, so automated checks matter more than a prompt for every step.

Duet adds one new safeguard. Anthropic says it is releasing it with a dedicated classifier and a safety system prompt that check each message and guard against leaks of sensitive information. They are designed to reject requests to share the strictest tier, including credit card numbers and government identification. Other sensitive information, such as health or personal finances, goes through only when you have explicitly told Duet to send it.^[1]

The X thread says the same, that the important decisions stay with you, and promises more on safety soon.^[2] Until that write-up appears, treat the design as stated intent. Anthropic has not published error rates for any of these checks in the pages we read.

What it can connect to

Duet can use the more than 6,000 connectors and plugins in the Claude Marketplace.^[1] Anthropic’s examples run from finding missed Gmail and Slack messages, to checking QuickBooks or Stripe for unpaid invoices, to preparing payroll in Gusto, sending agreements through Docusign, and managing a Shopify store.^[1]

It can also buy things for your business. Duet confirms the item with you, and Link, Stripe’s digital wallet, asks you to approve the spend, then pays with a virtual card so your real card number is not shared. It can also use Shop Pay from Shopify at online stores. Signing in to sites with logins saved in 1Password is listed as coming soon.^[1]

Who can use it, and the cost

For the research preview, Duet is included in Pro and Max plans and uses the same usage limits as the rest of Claude.^[1] The product page says Team and Enterprise plans are not included yet, and that on a computer you open it on the web at claude.ai.^[3] The X thread repeats that it is a waitlist for Pro and Max in the US.^[2]

Usage deserves a plain warning. Duet’s background checks count toward your limits, and larger tasks like a financial model or deck need more model turns and background work.^[1]

How to get ready and use it

You cannot use Duet yet unless Anthropic has let you off the waitlist, so this is a plan built from what Anthropic documents. We have not tried it. The waitlist form on the product page asks for your name, email, company, phone type and what you would hand off first, and Anthropic says it will email when Duet is ready for you.^[3]

Anthropic says you can tell Duet what you want to hear about and how often, and that you describe a task in a short message and it works out the apps and steps.^[1] So a good first message is short and concrete. Three examples of the sort of request Anthropic’s own examples suggest, written by us and not tested:

• “Each morning, tell me only about client emails and anything on my calendar that clashes.”
• “Check Stripe for unpaid invoices over 30 days and draft reminders for me to approve.”
• “Prepare a one-page brief before each external call this week.”

Limits and unknowns

Anthropic is candid that this is early. Duet may misread a thread or miscalibrate what is urgent, personal errands such as booking tickets are weaker than work tasks, and it cannot yet show you its browser. For now it lives in its own app, separate from the main Claude app.^[1] The X thread repeats that it can still misjudge what is urgent.^[2]

Beyond those, there is much we cannot say. No independent test, error rate or measured usage figure appears in the sources we read, and the testimonials are chosen by Anthropic.

The message

Claude Duet is a bet that the next step after a chat assistant is one that comes to you: a morning brief, background work on its own cloud computer, and a firm habit of asking before it sends, pays, signs or deletes. Anthropic’s safety claims are detailed and consistent with its earlier agent work. What nobody outside Anthropic can yet tell you is how often it gets the judgment right.

Join the waitlist if you are on Pro or Max in the US, start with low-risk drafts and briefs, and read every approval screen.

Quick quiz

Tap a question to check your answer.

Claude and Claude Duet are made by Anthropic; X, Stripe, Shopify, Gmail and the other apps named are made by their own companies. MSRX is independent and is not affiliated with or endorsed by any of them. Anthropic’s pages and X thread were opened in a browser on 22 October 2026; nothing was submitted or entered, and the waitlist was not joined. Quotes are word for word from the linked sources. Facts were checked on 22 October 2026 and may change, because this is a research preview. Anthropic’s images are used to illustrate the product it announces; the app’s icon and screenshots belong to Anthropic.