Claude Duet: What Anthropic's Proactive Agent Means for Your Business

Muhammad Shahid·AI-Powered Digital Marketing Consultant

Google Ads Certified

Meta Blueprint Certified

Google Analytics Certified

Updated October 2026

Claude Duet is a new app from Anthropic in which Claude stops waiting for you to ask. It watches your email, calendar, invoicing and other connected apps, tells you what needs your attention each morning, finishes tasks in the background, and checks with you before it acts on your behalf. Anthropic announced it on 21 October 2026 as a research preview for Claude Pro and Max subscribers in the United States, and it is aimed squarely at professionals and small business owners.

Almost everything published about it so far is a launch recap: the same feature list, re-ordered. That is useful for a day. What an owner actually needs to know is narrower. What will it do without asking? What will it cost you in usage? Which jobs is it ready for, and which ones should you keep? And if your business is not in the US, what should you do while you wait? I build agent workflows on Claude for service businesses, so that is the lens here. Every fact below is from Anthropic's own announcement, product page and support documentation unless I say otherwise.

What is Claude Duet?

Claude Duet is a proactive AI agent: software that watches, decides what matters and does work on your behalf, rather than a chatbot that answers when spoken to. Anthropic's announcement describes it as an app where Claude “comes to you with what needs your attention, completes tasks you ask of it, and checks with you before it acts on your behalf.” It is built on Claude Opus 5.5, the current top model in the Claude family.

You use it through a single ongoing message thread, typed or spoken. There are no prompts to write and no workflows to configure. You tell it what you care about, connect the apps you already use, and it reports back. The 9to5Mac write-up captured Anthropic's own framing well: Duet is something “you message the way you'd message a person.”

How it differs from the Claude app you already use

The ordinary Claude app is reactive. Nothing happens until you open a chat. Duet reverses that. It runs on your existing Claude account, so the memory, skills and connectors you have already set up are there from the first message, but it lives in its own standalone app on iOS and Android and on claude.ai. Anthropic's product page lists the practical differences:

• It starts the conversation. A morning brief, then a message only when something needs a decision.
• It keeps a task list. Work it is doing, work it needs from you. Only you can mark a task finished.
• It returns finished work. Docs, sheets, decks and dashboards, for you to review and edit.
• It keeps going when you close it. The work runs on a cloud computer, not your phone.

How does Claude Duet work day to day?

It runs a loop: read your connected apps, decide what matters, draft the next step, and ask you before anything leaves the building. The loop starts each morning with a brief and repeats every few hours through the day. Between check-ins it works through whatever you have handed it, on its own machine, and messages you when a task is finished or needs your review.

The morning brief and the check-ins

The morning brief is a short summary of the most important items across your connected apps, with drafted next steps attached. After that, Anthropic says, “Claude Duet checks your connected accounts every few hours”, and messages you only when something needs attention. You choose what you want to hear about and how often.

The “every few hours” is deliberate. Anthropic spaces the checks out to conserve your plan limits. It also means Duet is not an alarm system. Its own support article warns that because it checks at set intervals, urgent items can arrive late. For a plumber, a missed emergency call is still a phone problem, not an agent problem.

Its own computer in the cloud

Duet runs its own browser on a virtual machine, so your laptop and phone can be switched off while it works. For sites with no connector, it uses that cloud browser directly. Two practical limits follow: you cannot yet watch or take over that browser while it runs, and it cannot reach files sitting on your own computer. It can read the Claude work on your machines, but not change it.

The 6,000 connectors, translated into business jobs

Duet has access to over 6,000 connectors and plugins in the Claude Marketplace. That number means little on its own. The examples Anthropic chose are more telling, because nearly every one is back-office work a small business owner does at 9pm:

• Find missed messages in Gmail and Slack and help you reply.
• Check QuickBooks or Stripe for unpaid invoices and draft the reminders.
• Prepare and run payroll in Gusto, for your review.
• Prepare agreements for signature through Docusign.
• Draft follow-ups from HubSpot deal history.
• Turn a product launch into a social campaign in Canva.
• Set up and manage a Shopify store.

Read that list as a map of where your time leaks. If none of those apps are connected to anything today, the agent has nothing to read, and the morning brief will be thin. That is the part most launch coverage skips.

What will Claude Duet do without asking you?

It will read, summarise, draft and build on its own. It will not send a message, make a payment, sign a document or delete anything until you approve. Anthropic's wording is precise: before any of those actions, “it shows you exactly what it's about to do and waits for your approval.” A short list of actions is blocked even with your approval: trading securities and moving money between accounts. The support article adds accessing someone else's account.

This design matters more than any feature on the list. An agent that drafts the invoice reminder and waits is a time saver. An agent that sends forty reminders to the wrong customers at 3am is a reputation problem. The line Anthropic drew, read freely and act only with consent, is the same line I set on every agent I build for a client, and it is the right default for any business that has customers it cannot afford to annoy.

Purchases go through a one-time card

When Duet buys something, it uses Link, Stripe's digital wallet, or Shop Pay from Shopify. It confirms the item first, Link asks you to approve the spend, and the payment goes through a virtual card that works for that one purchase only, so your real card number is never shared. Order and refund problems still go to the store, not to Anthropic.

Passwords, sensitive data and prompt injection

Duet never sees your passwords or sign-in tokens directly. It follows the same design as Claude Managed Agents, and when you sign in to a site, the password goes into a separate form on your device. A dedicated classifier checks each message for sensitive information: requests to share credit card numbers or government ID are refused outright, and health or personal finance details are shared only when you explicitly instruct it. Anthropic says the product was tested by internal red teams and an external firm's penetration test, and it sits inside the company's public bug bounty program.

The honest risk is prompt injection: an email or web page written to trick the agent into doing something you did not ask for. Duet is built to treat content from emails and pages as information, not instructions, but the support article says plainly that no defence is perfect. The approval step is your real control. A booby-trapped email can make the agent propose something odd. It cannot make you approve it without reading.

Who can get Claude Duet, and what does it cost?

Claude Pro and Max subscribers in the United States, through a waitlist, at no extra charge during the preview. Claude Pro is 20 US dollars a month (17 on annual billing) and Max starts at 100, according to Anthropic's pricing page. Team and Enterprise plans are not included yet. Joining the waitlist takes a few minutes:

1. Check eligibility. A US Claude account on Pro or Max.
2. Join the waitlist on the Claude Duet product page with your name, email, plan and phone type.
3. Match your emails. Access is tied to the waitlist email, which has to match your Claude account and the Apple or Google account on your phone.
4. Wait for your wave. Invitations go out in batches, with no guaranteed date.
5. Install and sign in with your Claude account when the invite arrives.
6. Use the reset wisely. New users get one full reset of their five-hour and weekly limits, available until 30 November 2026.

The real cost is your usage limit

“Claude Duet is included in Pro and Max plans and uses the same usage limits as the rest of Claude,” says Anthropic. The part that matters for budgeting comes next: the background checks count, and so does the morning brief. Larger jobs like a financial model or a deck take more model turns. You also cannot pick a cheaper model. Duet always runs on Opus 5.5 and decides for itself how much effort a job needs.

So on the 20 dollar Pro plan, Duet and your normal Claude chats draw from the same pool. If you already hit Pro limits by Thursday, adding an agent that works in the background will get you there sooner. That is the most likely reason a Pro owner ends up on Max, and it is worth knowing before the invite lands.

For comparison, OpenAI's equivalent, dots, is limited to its Pro tiers, Business Premium and Enterprise. The entry Pro tier is the 100 dollar plan OpenAI introduced in April, reported by TechCrunch, and Bloomberg covered the new 500 dollar tier that launched alongside dots. On sticker price, Duet is the cheaper way in by a factor of five.

How does Claude Duet compare with OpenAI dots and Meta Muse?

All three are always-on agents that ask before acting, and all three business versions launched within 22 days of each other. The differences are price, where you can talk to them, and which business tools they reach. OpenAI announced dots on 29 September 2026. Meta launched Muse for Small Business the same day, built on the personal Muse agent it released earlier in September.

OpenAI's dots documentation confirms one real advantage: “Conversations with your dot don't count toward your ChatGPT usage limits.” Talking to Duet does. Muse has the opposite strength. It connects directly to your Instagram analytics, Facebook Pages and Meta ad accounts, which neither of the other two do natively.

The one independent hands-on review so far is from Dan Shipper at Every, who had been running his workday through a dot. His verdict: “Claude Duet is the first personal agent I’d call a pro tool.” He also found Duet's email drafts weaker than his dot's, and its scheduled runs could not reach his custom connectors. One reviewer is not a verdict, but it matches my read: Duet is the stronger back-office operator, and the writing still needs your hand on it.

What can't Claude Duet do yet?

Quite a lot, and Anthropic is unusually upfront about it. This is a research preview, and the gaps are the kind that matter to a business:

• It can misread a thread or misjudge what is urgent. Anthropic's words, not mine.
• It can miss something that arrives between check-ins.
• No phone calls and no text messages. With approval it can send email from connected accounts, and that is it.
• You cannot watch or take over its browser while it works.
• Personal errands are weak. Booking tickets and trip planning are its worst area.
• Connector gaps. Slack has limits, and the support article says Microsoft 365 is not available yet, even though it appears on the product page.
• It is a separate app from the main Claude app, so you have two places to check.

What should a small business hand to Claude Duet first?

Start with the work where a wrong draft costs you nothing and a missed item costs you money: unpaid invoices, unanswered enquiries and calendar clashes. Hand those over in the first week, review every draft, and only widen the scope once you trust what it flags. That is how I phase every agent rollout, and it applies here unchanged.

Week one: read-only jobs

Let it read and report, nothing more. The testers Anthropic quotes are telling here. Jess Provencher, a real estate broker, had Duet flag a property showing booked on the wrong day, without being asked: “exactly the kind of thing I want caught without having to ask.” Amin Ebrahimi, who runs Shahs of Kabob with about 110 staff in South Florida, uses it to work through invoices store by store and summed it up as “I don't need to open up 100 tabs.”

• The morning brief across email, calendar and invoicing.
• A weekly list of unpaid invoices older than 14 days.
• Enquiries that have gone more than a day without a reply.
• Calendar clashes and travel conflicts.

Week two onward: draft-and-approve jobs

Once the flags are reliable, let it draft: invoice reminders, follow-ups from the CRM, replies to routine enquiries, a sales forecast from the last quarter. You approve each one. The point of the second week is calibration. Every draft you edit tells you how much of its judgement you can trust on that job.

What I would keep away from it for now: anything customer-facing that is emotional (a complaint, a refund dispute), anything regulated, and anything where a late check-in costs a job. Your reviews, your Google Business Profile replies and your emergency phone line stay with a human. An agent that is wrong one time in twenty is a great assistant and a poor receptionist.

What if your business is outside the US?

You cannot get Duet during the preview, and there is no published date for other countries. Do not spend an afternoon looking for a workaround. Spend it on the part that decides whether any agent works for you when it does arrive: clean, connected data.

Duet's value is entirely in what it can read. A business with invoices in QuickBooks, enquiries in one inbox, appointments in one calendar and deals in a CRM gets a useful morning brief on day one. A business with invoices in a notebook, enquiries split across WhatsApp and three email addresses, and no CRM gets a brief that says very little. The agent is the same in both cases. The system under it is not.

The good news is that the groundwork carries across. Duet inherits the memory, skills and connectors from your regular Claude account, so everything you set up in Claude now is waiting for it. And if dots or Muse reach your market first, the same clean inbox, calendar and books make those agents useful too. This is the same principle I apply to AI visibility work: the machine can only work with what you have made legible to it.

Is Claude Duet worth joining the waitlist for?

If you are a US business owner already on Claude Pro or Max, yes. Joining costs nothing, there is no extra fee during the preview, and the approval design makes it safe to try on real work. Watch your usage for the first two weeks before deciding whether Pro is enough.

If you are not on Claude yet, do not sign up just for the waitlist. Pick the agent that matches where your work already lives. Back office in QuickBooks, Stripe and Docusign points to Duet. A team that runs in Slack or Teams points to dots. Sales through Instagram and Facebook points to Muse. And if you are still deciding which Claude plan or model your business needs at all, my guides to Claude Opus 5 for business and ChatGPT Work vs Claude Cowork cover that decision. For the wider picture of what agents can and cannot do in marketing, start with what AI agents are.

The bigger shift is not this one app. In a little over three weeks, Anthropic, OpenAI and Meta all shipped agents that work without being asked and stop before acting. The tools will keep changing. The businesses that benefit will be the ones whose systems an agent can actually read. If you want to know how ready yours is, a free audit is a good place to start.

Frequently asked questions about Claude Duet

The questions owners are asking most about Anthropic's new agent, answered from Anthropic's own documentation.