Claude Duet, Explained: What Anthropic’s Proactive Agent Does, Who Can Get It, and How It Compares

By Ammar Ahamed, Founder, Cocoon · 22 October 2026 · Facts checked against Anthropic’s product page and help centre on 22 October 2026

What Claude Duet is

The demo Anthropic chose for the Duet page is a phone lock screen at 9:41 on a Monday with a single notification: “Morning, Maya. 5 things in progress and 2 that need you.” That one line is the product. Not the five things in progress. The two that need you.

Every AI assistant until now has waited for you to open it. Claude Duet is built to come to you. Anthropic describes it as “a proactive agent that understands what needs your attention”: it watches the apps you connect, sends a short brief each morning, drafts the next steps, then goes quiet until something genuinely needs a decision from you. When you hand it a job, it works on its own computer in the cloud, so the work carries on after you close the app, and it messages you when the job is done or when it needs your approval.

Anthropic says it built Duet for “professionals and small business owners” with more on their plates than they can track. That focus shows in the examples it gives: chasing unpaid invoices in Stripe, preparing payroll for you to review, getting contracts ready to send, and building a brief before a 2 pm meeting.

Claude Duet at a glance

How Claude Duet works, day to day

One thread, like messaging a person

There are no prompts to write and no setup screens. You type or record a voice message, and your main thread stays in one place, so Duet picks up where you left off. You can open separate channels for different topics, and they all share the same memory. Because Duet runs on your Claude account, the memory, skills and connectors you have already set up in Claude are there from your first message. It can also search your past Claude chats (outside projects) for context, so you don’t have to repeat yourself.

A morning brief, then silence

Each morning Duet looks across your email and calendar, picks out what most needs your attention, and can start the first steps, such as drafting replies for you to review. Through the day it checks your connected apps at set intervals (MacRumors reports every few hours), sends updates on the work it is doing, and follows up on threads that have gone quiet. The design goal is restraint: it should reach out when something matters and stay quiet when it can wait.

Larger jobs, returned as files

Describe an outcome and Duet works out which apps to use and how much effort the job needs. Anthropic’s examples range from scheduling meetings to building dashboards across many tools, financial and sales forecasts, prototypes and web apps. Bigger results come back as files you can review and edit: a research brief, a financial model, a slide deck, a dashboard. Finished work lands in the Make tab.

Tasks you can track

The Tasks tab shows everything Duet is working on and highlights what needs your input, which keeps the main thread quieter. Duet may suggest a task looks finished, but only you can mark it complete. That small rule matters: the agent never gets to decide its own work is done.

Plugins and a browser for everything else

Duet supports Claude plugins, such as the small business plugin, and can suggest ones that fit your work. Anthropic says it works with more than 6,000 connectors and plugins. For websites with no connector, it uses its own browser in the cloud.

Who can get Claude Duet, and how the waitlist works

During the research preview you need all three of these:

• A Claude Pro or Max plan
• A Claude account in the US
• An invitation from the waitlist

How to join the waitlist:

1. Go to the Claude Duet page and select Join the waitlist.
2. Enter your name and email. If you already use Claude, use the email on your Claude account.
3. Choose your Claude plan and phone type (iPhone or Android). There is an optional box for what you do and what you’d hand off first.
4. Submit. Invitations go out in waves, and you’ll get an email when it’s your turn.
5. Open the invitation on your phone, install the app with the Apple or Google account linked to that same email, and sign in with your Claude account. It can take a few minutes to appear after you accept.

A tip on the optional question. Anthropic asks what you do and what you would hand off first. A specific answer (“I run a six-person agency; I’d hand off invoice chasing in Stripe and pre-meeting client briefs”) is more useful to a team shaping a preview than “everything”. We can’t promise it moves you up the list. Nobody outside Anthropic can.

Claude Duet pricing and usage limits

Price: during the research preview, Duet is included with Pro and Max at no extra cost.

Usage: this is the part most people will miss. Duet uses your plan’s usage limits, including the five-hour session and weekly limits, at the same rate as chats in Claude. Unlike a chat, though, it also works when you are not talking to it. The morning brief and the check-ins count too, so Anthropic warns you may reach your limits sooner than with chat alone. Larger jobs, like a financial model or a deck, take more model turns and more background work. To stop background work eating your allowance, Duet checks your accounts at set intervals rather than constantly.

The launch bonus: when you are admitted, you get one full limit reset that clears both your five-hour and weekly limits. It is available until 30 November 2026. Save it for a heavy day, not your first ten minutes.

Pro or Max? Anthropic hasn’t published Duet-specific allowances. Our rule of thumb: if you already bump into limits on Pro with ordinary chat, an always-on agent will make that worse, and Max is the more realistic plan for heavy daily use. If you rarely hit limits today, start on Pro and watch your usage for a week.

What Claude Duet can’t do yet

Anthropic is unusually direct about the gaps. During the preview:

• You can’t choose the model. It runs on Opus 5.5.
• You can’t watch it work in its browser or take over the browser. This is a real difference from ChatGPT Dots and Meta Muse, which let you watch.
• It checks your accounts at set intervals, so something urgent can arrive between checks.
• It can’t see all your projects or the files on your computer, and it can read your work in Claude but not change it.
• It doesn’t make phone calls or send text messages to other people. With your approval, it can send email from accounts you have connected.
• Some actions are off-limits even with your approval: trading securities, moving money between accounts, or accessing someone else’s account.
• Microsoft 365 isn’t available yet, and some connectors, such as Slack, work with limits.
• It makes mistakes. Anthropic says it may misread a thread or misjudge what’s urgent, so review drafts and approval requests before you confirm them.

The Microsoft 365 gap deserves a flag for corporate readers. A large share of companies in Sri Lanka, Singapore and the Gulf run on Outlook and Teams. Until that connector arrives, Duet is a much better fit for people whose working life is in Google Workspace.

Is Claude Duet safe? Approvals, payments and privacy

An agent that reads your inbox and can spend money is only as good as its brakes. Duet’s are worth understanding:

• Approval before consequential actions. Duet shows you what it is about to do and waits before sending a message, making a payment, signing a document or deleting anything. If you can’t approve straight away, it finishes the rest of the task and brings the approval back later, so one pending decision doesn’t stall the whole job.
• Single-use cards. Purchases go through Link (Stripe’s wallet) or Shop Pay (Shopify’s). Duet asks before each one, and each uses a one-time card that only works for that purchase.
• Passwords stay out of the chat. When a site asks for a password, you type it into a separate form on your device that sends it straight to Duet’s browser. MacRumors reports Duet doesn’t see your passwords or logins, and that a dedicated classifier and safety system prompt check each message to guard against leaks of sensitive information.
• Receipts for every number. When Duet gives you a figure or a summary, it tells you which email, file or spreadsheet it came from.
• Honesty about prompt injection. Duet reads things other people wrote, emails and web pages, and those can contain instructions meant to mislead it. Anthropic says Duet is built to treat what it reads as information, but that “no defense is perfect”, which is one reason consequential actions need your approval.

Two privacy details to know before you connect anything. First, disconnecting an app stops Duet from accessing it from then on, but memory of information from earlier conversations is kept. Second, Duet’s chat search uses the same Search and reference chats setting as Claude (Settings > Memory); turn it off if you don’t want Duet drawing on past chats, and deleting a chat removes it from search.

If your company doesn’t yet have rules for what staff may connect to an AI agent, write them before anyone gets an invite. Our free AI Policy builder gives you a starting draft in minutes.

Claude Duet vs Claude vs Claude Cowork

Anthropic now has three ways to get work done with Claude, and they are easy to confuse.

The simplest way to hold it: Claude waits for you. Cowork does what you ask. Duet notices what you haven’t asked yet. Anthropic puts it like this in Duet’s FAQ: the Claude app waits for you to open a chat, while Duet comes to you. All three share the same account, so memory and connectors carry across.

One more thing worth knowing: Anthropic is merging chat and Cowork into a single Claude, where you ask for what you need and Claude decides whether that is a quick answer or a task. That is rolling out to Pro and Max. Duet stays separate, as its own app.

Claude Duet vs ChatGPT Dots, Meta Muse, Grok Bot and Manus Cue

Duet is the last of the big labs’ always-on agents to arrive. xAI’s Grok Bot went into beta on 11 August, Meta’s Muse launched on 8 September, Manus Cue and xAI’s Team Bots on 28 September, and OpenAI’s dots on 29 September. (We cover all of them in our ultimate guide to AI agents and assistants.) Arriving last gave Anthropic the chance to define itself against the rest, and it chose judgement: knowing when to interrupt and when to wait.

Dots, Muse, Grok Bot and Cue details as reported in our agents roundup; Duet details from Anthropic, 22 October 2026.

Choose Duet if you live in Gmail and Google Calendar, already pay for Claude, and want something that triages your day and runs small-business admin (invoices, contracts, payroll prep) without you watching over it. Choose ChatGPT Dots if your team works in Slack or Microsoft Teams, or you are outside the US. Choose Meta Muse for personal life admin in the US. Choose Grok Bot if you want several specialised agents collaborating, and Cue if the job needs an agent with its own phone number or wallet.

Five things Claude Duet tells us about where AI is going

1. The scarce resource is no longer output. It is attention.

Duet’s tagline is “Get back to what needs you”, and its home screen counts the things that need you, not the things it did. That is a bet that the hard problem in 2026 is not generating a reply, a deck or a forecast. It is deciding which of the hundred things in your day deserve you at all. Notice the failure mode Anthropic warns about: not a wrong answer, but misjudging what is urgent. Proactive agents fail by interrupting badly. That is a new kind of error, and teams will need a new kind of review to catch it.

2. Your history with Claude is the onboarding

Duet has almost no setup because it inherits everything: your memory, your skills, your connectors and your past chats. That makes it better on day one for people who have used Claude seriously, and it makes switching harder later. The practical lesson, wherever you are: the work you put into teaching an assistant your preferences, tools and context compounds. Agents will reward people who have been managing their AI well, and penalise people who have been using it as a search box.

3. Approval is becoming the interface

The most interesting design decisions in Duet are about brakes, not engines: approval before send, pay, sign or delete; the rest of the task carrying on while one approval waits; single-use cards; hard limits that hold even if you say yes; and only you can mark a task done. These are good rules for any agent, including the ones your own team builds. If you are writing an AI policy this quarter, borrow them.

4. Always-on means always spending

A chatbot costs nothing when you are not using it. A proactive agent spends from your allowance every time it checks your inbox. Anthropic is open about this, which is to its credit, and OpenAI has taken the opposite line by not counting chats with your first dot toward limits. Expect pricing for agents to become the next competitive battleground, and expect the cost of “watching” to be priced separately from the cost of “doing”.

5. The skill shifts from prompting to managing

Duet is explicitly a product with no prompts. You brief it like a colleague, check its work, approve what matters and correct its sense of priority. Those are management skills: delegation, clear outcomes, review, and knowing what never gets delegated. That is the core of what we teach at Cocoon, and it is why we think being AI native is about habits, not tools.

15 jobs to hand Claude Duet first

Every request below uses a capability or connector Anthropic documents for Duet. The first three are Anthropic’s own examples; the rest are ours.

Inbox and calendar

1. “Clear out anything in my inbox that I don’t need to read.”
2. “Check Stripe for unpaid invoices and draft reminder emails for me to approve.”
3. “Pull together a brief on this company before my 2 pm meeting.”
4. “Find every thread where someone is waiting on me for more than three days and draft a reply to each.”
5. “Look at next week’s calendar and tell me which meetings have no agenda or no clear owner.”

Running a small business

6. “Reconcile this month’s QuickBooks expenses against the receipts in my inbox and list what’s missing.”
7. “Prepare this month’s payroll summary for me to review. Don’t submit anything.”
8. “Get the contract in my Drive ready to send in Docusign, and hold it for my approval.”
9. “Build a cash-flow forecast for the next quarter from our invoices and expenses.”

Sales and clients

10. “Check HubSpot for deals with no activity in two weeks and draft a follow-up for each.”
11. “Send the people who replied to my proposal a Calendly link, once I approve the wording.”
12. “Turn last week’s client emails into a one-page status update for each account.”

Bigger builds

13. “Make a ten-slide deck from the board notes in my Drive.”
14. “Build a dashboard of this quarter’s sales by region from our spreadsheets.”
15. “Research three competitors’ pricing pages and give me a comparison brief, with sources.”

Notice what is missing: anything that sends, pays or signs without you. Keep it that way for the first month.

Your first week with Claude Duet

1. Before you accept the invite, check what Claude already remembers about you (Settings > Memory), and remove connectors you no longer use. Duet inherits all of it.
2. Days one and two: read, don’t delegate. Connect only Gmail and Google Calendar and read two morning briefs. Then tell it, in plain words, what counts as urgent for you and what never does. You are calibrating its judgement.
3. Day three: one recurring, low-risk job. Inbox clear-out drafts, or a brief before each external meeting. Time how long it saves you.
4. Day four: one big job. A deck, a forecast or a dashboard. If it’s a heavy day, this is where your one-off limit reset (valid until 30 November) earns its keep.
5. Day five: review. Go through the Tasks tab. Note every time it misjudged urgency or misread a thread, and tell it. Mark finished tasks done yourself. Then decide what it owns next week, exactly as you would with a new hire.

Outside the US? What to do in Sri Lanka, Singapore and elsewhere

During the preview, Duet requires a Claude account in the US, and Anthropic hasn’t given a date for other countries. Don’t try to work around it; access is tied to your account and the waitlist email. Here is what you can do now instead:

• Use Claude Cowork. It is on every paid Claude plan, including Team and Enterprise, and its help article lists no US-only restriction. Hand it multi-step jobs and set up scheduled tasks. It is the closest thing to Duet you can use today.
• Get your Claude house in order. Memory, skills and connectors are what Duet inherits. People who set them up well now will get the most from Duet on day one, whenever it reaches them.
• Look at the alternatives that already work here. ChatGPT Dots are open to Pro users in Sri Lanka and Singapore, Grok Bot comes with SuperGrok, and there are self-hosted options too. Our ultimate guide to AI agents compares them all, including open-source OpenClaw, Hermes Agent, OpenWorker and OpenBot.
• Build the habits first. The bottleneck for most teams isn’t access to agents. It’s knowing what to delegate, how to brief, and what to check. Our free AI Readiness Score shows where a person or team stands in about ten minutes.

Should you join the Claude Duet waitlist?

Join now if you are in the US, on Pro or Max, your work runs through Gmail and Google Calendar, and you spend real time on admin: invoices, scheduling, follow-ups, contracts, briefs. You lose nothing by joining, and the earlier waves are where the product gets shaped.

Wait if your company runs on Microsoft 365, you’re on a Team or Enterprise plan, you need to watch an agent work before you trust it, or you need it to make calls or texts. None of that is supported yet.

Whatever you decide, the bigger shift is already here. In ten weeks, every major AI lab shipped an agent that works while you don’t. The people who benefit won’t be the ones with the most agents. They’ll be the ones who manage them best.

Frequently asked questions

Sources

Anthropic

• Claude Duet product page, Anthropic
• Get started with Claude Duet, Claude Help Center
• Get started with Claude Cowork, Claude Help Center
• Claude Duet announcement, @claudeai on X

Launch coverage

• Anthropic takes on ChatGPT Dots and Meta Muse with Claude Duet, 9to5Mac
• Anthropic Launches Claude Duet, a Proactive AI Agent, MacRumors
• Claude Duet is Anthropic’s proactive AI agent answer to Muse and OpenAI’s Dots, 9to5Google
• Anthropic announces preview of Claude Duet, CNBC