Two customer emails waiting since FridayReplies drafted for you to reviewDrafted
Three invoices past their due dateReminder emails ready to sendNeeds you
Tomorrow’s supplier call overlaps the site visitMove the call to the afternoon?Question
Through the day
Tasks
On its own computer in the cloud, with your laptop closed
Read the three unpaid invoices in Stripe
Drafted three reminders in Gmail
Waiting for you before sending
Checks your apps again every few hours and messages you only when something needs you
When it needs you
Waiting for your approval
Send three payment reminders
Action
Send an email
From
Your Gmail account
To
accounts@ first customeraccounts@ second customeraccounts@ third customer
Content
“A reminder that the invoice below is now past its due date…”
Checked separately: matches what you asked
ApproveEdit
Illustration with a fictional business, built from Anthropic’s announcement of 21 October 2026 and its safety post of 23 October. It is not Anthropic’s interface.
On 21 October 2026 Anthropic opened a research preview of Claude Duet, a separate app in which Claude works as an agent that brings you what needs your attention instead of waiting for a question. Each morning it sends a short brief on what matters across the apps you have connected, it checks them again every few hours, it carries out tasks in the background on its own computer in the cloud, and it shows you what it is about to do before it sends a message, makes a payment, signs a document or deletes anything. The preview is open by waitlist to Pro and Max subscribers in the United States. This page sets out what Duet does, what it refuses to do, how it handles what it reads and who can use it, from Anthropic’s announcement, its safety post and its help page.
By The AI Adoption · Published on 25 October 2026 · Sources: Anthropic, 21 and 23 October 2026
What Anthropic announced
Claude Duet is a new app, on iOS, Android and claude.ai, where Claude works as a proactive agent. Anthropic says it built Duet for professionals and business owners with more on their plates than they can track, and designed it around Claude Opus 5.5. Each morning Duet sends a short brief on the most important items across your connected apps; through the day it keeps track of those apps and messages you only when something needs you; and you can hand it tasks of every size, from scheduling meetings to building a dashboard, a financial or sales forecast, or a prototype.
Duet starts from what your Claude account already holds: your memory, your skills and the apps you have connected. You message it by typing or by voice in one ongoing thread, and it works on a virtual machine with its own browser, so the work carries on after you close the app or switch off your laptop. Larger results, such as a research brief, a financial model, a slide deck or a dashboard, come back as files you can review and edit, and a Tasks tab shows what Duet is working on and what needs your input. When it gives you a number or a summary, Anthropic’s help page says it tells you which email, file or spreadsheet it came from.
Duet uses the connectors and plugins of the Claude Marketplace, which Anthropic puts at more than 6,000. The examples in the announcement are business tasks: finding the Gmail and Slack messages you missed and helping you reply, checking QuickBooks or Stripe for unpaid invoices and drafting reminders, preparing agreements and sending them for signature through Docusign, drafting follow-ups from the contact and deal history in HubSpot, and setting up and running a Shopify store.
What it does alone, and what waits for your approval
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 the action, where it is going and the content, including every recipient, and waits for your approval. If you cannot approve straight away, the help page says it finishes the rest of the task and brings the approval back to you later.
A second check runs alongside each action, separate from the model doing the work. Anthropic built it from the auto mode classifier of Claude Code and extended it for Duet. It assesses whether the action matches what you asked, and because it runs outside that model, instructions hidden in a web page or an email cannot change its answer.
What Duet does before an action
From Anthropic’s safety post of 23 October 2026
Goes ahead on its own
Reads your inbox and apps
Drafts replies
Prepares a model or a deck
Checks your accounts every few hours
Shows you, then waits
Sends a message
Makes a payment
Signs a document
Deletes anything
Asks every time, no “don’t ask me again”
Paying or sending money
Deleting an order
Creating an account
However often you have approved them before
Refuses, even with your approval
Trading securities
Moving money between accounts
Accessing someone else’s account
Before you approve, you see
The action
Where it is going
The content
Every recipient
Our drawing of the rules in Anthropic’s safety post of 23 October 2026 and its announcement of 21 October.
Some actions stay off-limits even with your approval: trading securities, banking actions such as moving money between accounts, and accessing someone else’s account. Others can never be covered by a “don’t ask me again” rule, however often you have approved them: paying or sending money, deleting an order and creating an account. The routines Duet runs while you are away follow the same rules.
For purchases, Duet pays through Link, Stripe’s wallet, or Shop Pay from Shopify, with a one-time card limited to that merchant and that amount. You see the item and the total, you approve the spend in the wallet’s own app, and your real card number is shared neither with Duet nor with the merchant.
How it handles what it reads
An agent that reads your email and sends messages for you keeps deciding where information should go. Anthropic’s example is a weekly recap for your team, which should contain only what everyone on the team can already see and never something like a performance review conversation. A separate classifier checks each message Duet is about to send, and sorts what it finds into two tiers.
In the first tier, which covers sign-in credentials, full card and bank account numbers and government ID numbers, Duet is instructed never to share the information with anyone, even if you ask. In the second, which covers private information such as health or finances, it shares only with your explicit approval, and your own request counts as that approval. When a message is blocked, Duet can rewrite it without that information and bring it back to you. Anthropic says the classifier will sometimes miss something or flag a harmless message, which is why the approvals stay in place.
Duet never sees your passwords, your browser cookies or the sign-in tokens of your connected apps. When a site asks for a password, you type it into a form inside the Claude app, which passes it encrypted to the browser machine, so the model neither sees it nor can save it. Anthropic adds a warning worth passing on to a team: the Duet sign-in form only appears inside the app, during a sign-in you asked for, and never arrives as a link, a text message or an email, so anything else that asks for your password in Duet’s name is not legitimate.
Switching an app off in Claude switches it off in Duet as well. Disconnecting an app stops new access from that moment, but what Duet has already learned from it stays in its memory until you reset that memory, which you can do at any time from the app’s settings.
Who can use it, and from where
Anthropic’s help page sets three conditions: a Pro or Max plan, a Claude account in the United States, and an invitation from the waitlist. Invitations go out in waves, without a set date. Duet is not available on Team or Enterprise plans yet, and Anthropic says it plans to bring it to those plans later. For a business in Australia, New Zealand or the United Kingdom, the preview is closed for now, whatever its plan.
Who can join the preview today
From Anthropic’s help page, updated on 24 October 2026
Plan
ProBy waitlist
MaxBy waitlist
TeamPlanned later
EnterprisePlanned later
Your account
United StatesIn the preview
AustraliaNot yet
New ZealandNot yet
United KingdomNot yet
Apps it names
Gmail
Google Calendar
Google Drive
Slack with limits
QuickBooks
HubSpot
Docusign
Calendly
Not available yet: Microsoft 365
Our drawing of the access rules in Anthropic’s help page, “Get started with Claude Duet”.
Duet works with the connectors you have connected in Claude, and the help page names Gmail, Google Calendar, Google Drive, Slack, QuickBooks, HubSpot, Docusign and Calendly. Some, such as Slack, work with limits, and Microsoft 365 is not available yet, so a business that runs its email and files on Outlook, Teams and SharePoint could not put Duet to work on them today, even from a US account. We compare what Claude and Microsoft Copilot can reach in Microsoft 365 in today’s other page, on Microsoft Copilot or Claude for a business on Microsoft 365.
During the preview Duet is included in the Pro and Max plans at no extra cost and draws on the same usage limits as Claude, including the five-hour session limit and the weekly limit. Its background work, the morning brief and the check-ins, counts as well, so a plan reaches its limits sooner than with chat alone. When you are admitted, you receive one full limit reset, which you can use until 30 November 2026.
What it cannot do yet
Anthropic lists the limits of the preview. You cannot choose the model: Duet runs on Claude Opus 5.5 and decides how much effort a job needs. You cannot watch it work in its browser or take that browser over. It checks your accounts at set intervals, so something urgent can arrive between two checks. It cannot see all your projects or the files on your computer, and it can read your work in Claude without changing it. It does not make phone calls or send text messages to other people, and with your approval it can send emails from the accounts you have connected.
Anthropic also writes that Duet may misread a thread or misjudge what is urgent, that personal errands such as booking tickets are weaker than work tasks, and that for now Duet lives in its own app, separate from the main Claude app.
What the announcement does not cover
Anthropic gives no date for other countries, or for the Team and Enterprise plans. Because the preview runs on individual Pro and Max accounts, the pages we read describe the controls each person has, such as approvals, connected apps and the memory reset, and no control for a company administrator, such as limiting which apps the agent may use across a team. Anthropic also says that how often Duet asks for approval, how you sign in and how well its information checks tell private from harmless are still changing.
How we start
We work in Claude and set it up for the businesses we work with: the knowledge the AI needs, the connections each person uses, access that reflects each role and the points where a person approves. Duet runs on the memory, skills and connected apps of the Claude account it starts from, so what it can reach and what it knows about your business depend on that setup, as they do for the agents we compare in which AI agent a business should use. We explain how we decide what AI may read and do in adopting AI securely and in handing a repetitive task to AI with a person approving. If you want to talk through what an agent like Duet would mean for your business, the first step with us is a free AI pre-audit: 45 minutes on a call about how you and your team use AI today and which work to hand over first. You leave with a clear next step, with or without us.
Every fact on this page comes from Anthropic’s announcement, safety post and help pages, read on the day of publication. Claude Duet is a research preview and details may change after publication.
The AI AdoptionFree AI pre-audit 
Google Drive
Slack with limits
QuickBooks
HubSpot
Docusign
Calendly
