Claude Duet Explained: Anthropic’s Proactive AI Agent That Works Before You Ask (2026 Guide)

Table of Contents

Professional checking a mobile assistant app — Claude Duet is Anthropic's proactive AI agent for professionals on iOS and Android

What Is Claude Duet?

On October 21, 2026, Anthropic quietly flipped the script on how AI assistants work. The company introduced Claude Duet, a standalone AI agent app built to act as a proactive assistant for professionals and small-business owners. Instead of waiting for you to type a prompt, Duet watches your connected apps, flags what matters, and starts handling the work before you even think to ask.

That single idea is what separates Claude Duet from every chatbot you have used so far. Traditional assistants are reactive: you open the app, type a question, and get an answer. Claude Duet is designed to notice the overdue invoice sitting in your accounting software, the customer email that never got a reply, or the meeting scheduled on the wrong date — and then do something about it without being told. This is the core of what makes Claude Duet different from a chatbot.

Anthropic is rolling out Duet as a research preview for now. It is available to Pro and Max subscribers in the United States through a waitlist, and it runs as a separate mobile app for iOS and Android, plus through the Claude website. Under the hood, Duet is powered by Anthropic’s Opus 5.5 model, which the company says was chosen specifically for its judgment — knowing what deserves your attention and when to ask for your approval.

The timing matters. Just weeks earlier, OpenAI launched its always-on Dots agents, and Meta’s Muse AI agent had raced past 10 million downloads in its first 40 days. The era of AI that waits politely for instructions is ending. Duet is Anthropic’s answer: an agent that goes out and meets the work instead of sitting around waiting for it.

How Claude Duet Works

The core promise of Claude Duet is simple: you should not have to initiate every interaction with your AI. Here is how it actually delivers on that promise.

The Daily Briefing and Background Checks

Every morning, Duet produces a custom daily briefing pulled from your connected apps. It summarizes what happened overnight, what needs attention today, and what has already been handled in the background. Then, every few hours throughout the day, it checks those accounts again for new developments — an unpaid invoice that crossed a deadline, a customer message waiting for a reply, a calendar change that conflicts with another commitment.

An important detail from Anthropic: these checks happen every few hours, not continuously. More frequent monitoring would burn through your usage allowance faster, so the company deliberately made Claude Duet a periodic checker rather than a real-time supervisor. You can adjust which subjects it flags and how often it updates you, so the agent learns what actually matters to your work instead of pinging you about everything.

The Cloud Computer That Never Sleeps

Perhaps the most important part of the Duet architecture is that it runs on its own cloud computer, not on your device. Anthropic describes a virtual machine equipped with a browser where Duet figures out which applications and steps a task requires, performs the work remotely, and then notifies you when it finishes or needs a decision.

That means you can close your laptop, turn off your phone, and Claude Duet keeps working. If you asked it to research a vendor, assemble a report, and draft follow-up emails, the job continues while your devices sit in your bag. When the work is done — or when it needs your approval for something consequential — you get a notification in the persistent conversation.

You communicate with Claude Duet through text or voice in one continuing conversation. It also brings over your account memory, skills, and existing app connections from Claude, so it starts with context you have already given instead of forcing you to re-explain your business from scratch.

Small business owner working on a laptop — Claude Duet is built for professionals and small business owners who juggle many apps

Claude Duet and the Claude Marketplace

Anthropic says Duet can draw on more than 6,000 connectors and plugins available through the Claude Marketplace. That connector catalog is the engine room of the whole product: Duet plugs into tools like Gmail, Slack, QuickBooks, Stripe, HubSpot, Gusto, Docusign, Canva, Shopify, and LinkedIn, then reads across them to build a picture of what your business needs right now.

This is also where Anthropic draws a clear line. The breadth of integrations lets Duet reduce the manual handoffs between business applications — but it also means the permissions attached to each connection matter enormously. Access that is sufficient to summarize an invoice is very different from authority to initiate a payment, and Anthropic built Duet’s approval system around exactly that distinction.

Key Features of Claude Duet

Here is what Claude Duet brings to the table, based on Anthropic’s official announcement and the reporting that followed it.

  • Proactive daily briefings: a morning summary of what matters, generated from your connected apps, with only genuinely important items surfaced.
  • Periodic background monitoring: Duet re-checks your accounts every few hours and notifies you when something needs attention, based on subjects and frequencies you control.
  • Persistent text and voice conversation: one continuous conversation that carries your account memory, skills, and connections, so context never resets.
  • Cloud-based task execution: a remote virtual machine with a browser that completes multi-step work — gathering information across apps, producing documents, preparing follow-ups — even while your devices are off.
  • 6,000+ app connectors: integrations through the Claude Marketplace covering email, calendars, accounting, CRM, payroll, e-signatures, design tools, and storefronts.
  • Agentic purchases: support for buying through Stripe’s Link digital wallet and Shopify’s Shop Pay, with spending authorization built into the flow.
  • Approval-gated sensitive actions: Duet prepares work independently but must ask before sending messages, making payments, signing documents, or deleting anything.
  • Sensitive-information classifier: a dedicated safety layer that evaluates messages to prevent inappropriate disclosure, blocking the most restricted categories like payment-card numbers and government-issued identifiers outright.
  • Credential isolation: designed to keep passwords and authentication tokens out of the model’s direct view, with a separate action-approval classifier and defenses against prompt injection.

What Claude Duet Can Actually Do for Your Business

Feature lists only tell you so much. What matters is what Claude Duet does inside a real working day. Anthropic’s launch materials sketch out several concrete workflows that show Claude Duet’s practical value.

Accounting and Invoice Chasing

Duet can examine your QuickBooks or Stripe records for outstanding invoices, identify which customers are overdue, and prepare reminder messages for you to approve. For a freelancer or small business owner, this is one of the most immediately valuable use cases: unpaid invoices are the silent cash-flow killer, and most people put off chasing them because it feels awkward and time-consuming. Duet does the uncomfortable part — it finds the overdue amounts and drafts the reminder — while you keep the final say on what gets sent.

Customer Outreach and Follow-Ups

Using customer information from HubSpot, Claude Duet can draft outreach messages and follow-ups after meetings or purchases. It can also flag a customer email that sat unanswered and prepare a response that fits the context of the conversation so far. The pattern is consistent across every workflow: Duet does the research and the drafting, you do the approving.

Agreements and Documents

Claude Duet can help assemble contracts and agreements for review through Docusign. It can also build financial forecasts and create web applications — bigger projects that require it to plan across multiple steps and tools. Anthropic’s announcement describes it preparing documents from information gathered across several systems: pulling the numbers from accounting, the customer details from the CRM, and the template from your document library.

Payroll and Hiring

For people operations, Anthropic describes Claude Duet preparing and running Gusto payroll using employee hours and adjustments, including bonuses and reimbursements. It can also draft recruitment listings and publish them to LinkedIn after you review and authorize the content. Running payroll is exactly the kind of task where the approval boundary matters most — and Anthropic is betting that a “prepare, then ask” model will earn trust where a fully autonomous agent would not.

Entrepreneur reviewing business workflows — Claude Duet handles invoice chasing, marketing, payroll, and customer follow-ups

Marketing and Your Online Store

Claude Duet can develop Canva campaigns and help configure or maintain a Shopify store. Anthropic says a more capable storefront design interface, including a preview before publication, will arrive later. The commerce angle extends to actual purchasing: through Stripe’s Link wallet, Duet checks the proposed item with you, Link requests spending authorization, and payment goes through a virtual card that keeps your real card number private. Shopify’s Shop Pay is supported as well.

Anthropic separately plans to let Claude Duet use credentials stored in 1Password with your permission — a feature that is forthcoming, not yet live. That capability, once shipped, would close one of the biggest practical gaps in agentic shopping: the agent needs to sign in as you to buy for you, and doing that securely has been the industry’s hardest problem.

Real-World Early Examples

Anthropic’s announcement includes two company-selected early users. Real estate broker Jess Provencher uses Duet to catch scheduling mistakes — in one case, it spotted a property showing entered on the wrong date. Restaurant owner Amin Ebrahimi, whose Shahs of Kabob business employs roughly 110 people, has used it for accounting, customer service, and marketing. His quote from the announcement: “It brings it all together. I don’t need to open up 100 tabs.”

Treat these as what they are: company-selected accounts, not independent evaluations of accuracy or business impact. They are useful as illustrations of the product’s intended shape — catching small operational mistakes that accumulate through a workday — rather than proof of how reliably it performs.

Claude Duet Pricing: What It Really Costs

Here is the straightforward pricing picture, and then the part most coverage glosses over.

There is no separate price for Claude Duet. During the research preview, Anthropic includes Duet in its existing Pro and Max subscriptions. Published subscription pricing lists Pro at $20 per month (or $200 per year, which works out to about $17 per month) and Max starting at $100 per month. If you already pay for either plan, Duet costs you nothing extra on the subscription line — at least for now.

The real cost is usage. Duet shares your existing Claude usage allowance, and its automated background checks consume part of that allowance. A bigger assignment — building a financial model or a presentation — consumes additional work on top. So the question is not really “what does Duet cost?” but “how much useful output can you get from your allowance after the agent’s recurring monitoring takes its share?”

Anthropic does not provide task-level cost estimates, quantified productivity gains, or benchmark results in the Duet announcement. That is a genuine gap for anyone evaluating it seriously. Businesses testing the preview will need to measure three things themselves: the work Duet actually completes, the correction time staff spend fixing its mistakes, and how fast it consumes the allowance. Without those numbers, the $20-per-month entry price tells you less than it seems to.

Two more pricing notes worth knowing. First, the initial preview is limited to individual Pro and Max subscribers — Team and Enterprise plans do not have access yet, and there is no Duet API. Second, per VentureBeat’s reporting, buyers on individual plans should not assume they get the same governance controls listed under Enterprise pricing, such as audit logs and custom retention controls. If your business needs those, Duet’s current preview is probably not where you start.

Safety and Approval Controls

Anthropic drew an explicit boundary in the Duet announcement between preparing work and taking consequential action — and that boundary is the heart of the product’s safety story.

Claude Duet can inspect information and assemble drafts on its own. But before it sends a message on your behalf, makes a payment, signs a document, or deletes anything, it must present the intended action and ask for your approval. Some activities are prohibited even with authorization: securities transactions and transfers between financial accounts are off-limits entirely.

Behind that approval flow sits a stack of safeguards. A sensitive-information classifier, paired with safety instructions, evaluates messages to prevent inappropriate disclosure — blocking the most restricted information categories (payment-card numbers, government-issued identifiers) and permitting disclosure of other sensitive material, like health or personal financial information, only after an explicit instruction. Credential isolation is designed to keep passwords and authentication tokens out of the model’s direct view. An action-approval classifier sits in the loop, and there are defenses against prompt injection — malicious instructions hidden inside content the agent encounters while browsing.

Anthropic says internal security teams and an outside firm tested Duet’s infrastructure, and its public bug bounty program covers the product. As VentureBeat rightly notes, though, these are descriptions of safeguards, not evidence that the agent will avoid every unauthorized action or information leak. For anyone evaluating Duet, the honest framing is this: the safety architecture is unusually explicit for a product this new, but the preview is young, and caution is the right default.

This matters in context. The agentic web is already under strain — major websites are blocking AI agents over security and liability concerns, and Meta’s Muse ran into a privacy scandal when thousands of shared chats went public. Anthropic is launching Duet into an environment where agent trust is the single most contested question in tech.

Claude Duet vs Meta Muse vs OpenAI Dots

Duet enters a market that formed almost overnight. Its closest comparisons are Meta’s Muse and OpenAI’s Dots — both persistent, proactive AI agents that run on their own cloud computers and continue working after you close the app.

Business professional comparing documents — Claude Duet compared with Meta Muse and OpenAI Dots on price, privacy, and features

Meta Muse

Meta introduced Muse in September 2026 with an emphasis on everyday tasks, purchases, travel, and longer-term goals. Users reach it through its app or WhatsApp. Like Duet, it continues working after the app closes and seeks permission for sensitive actions.

The headline difference is scale. Sensor Tower estimates put Muse at 10.9 million downloads since its September debut — 10.2 million of them in the US — with 2.5 million daily active users by October 16. That early traction suggests real consumer appetite for AI that goes beyond answering questions to carrying out tasks. Muse also offers a free entry point, which Duet does not — Duet’s minimum ticket is a $20-per-month Pro subscription.

On the other hand, Muse’s brand took a hit days before Duet’s launch when roughly 6,800 shared chats went public — the privacy episode covered in our Meta Muse privacy leak report. For professionals who will connect an agent to their business apps, that timing could not have been worse for Meta.

OpenAI Dots

OpenAI’s Dots overlap particularly closely with Duet’s professional ambitions. OpenAI describes agents powered by GPT-6 Astra that investigate issues, build applications, and advance projects between conversations. They work through ChatGPT and messaging channels including Slack and Teams.

Dots has one advantage Duet currently lacks: an organizational access route. OpenAI’s access documentation includes Pro and Business Premium subscribers plus an administrator-enabled Enterprise beta. Duet, by contrast, is individual-Pro-and-Max-only in the preview, with no Team or Enterprise path announced. On price, Duet’s $20-per-month entry point undercuts the $100 minimum individual Pro subscription level reported for Dots — though both bundle their first agent into a broader subscription, so the real comparison depends on what else you get from each plan.

Where Claude Duet Stands Apart

Anthropic’s pitch is that Duet is the professional’s agent: QuickBooks, Stripe, Gusto, HubSpot, Docusign — the vocabulary of running a business rather than running errands. Muse leans consumer (travel, shopping, daily goals). Dots straddles both but is anchored in the ChatGPT ecosystem. Duet is the only one of the three that launched US-only and subscriber-gated through a waitlist — a narrower, more cautious rollout that fits Anthropic’s safety-first brand, even if it means slower adoption.

The other differences worth noting: Instinct (from Spear Street Technology) competes on delegation through familiar communication — users text or call an assistant that can operate a phone and computer for errands like groceries, transport, and bookings. Tolan (from Portola) is more character-driven, emphasizing conversation and memory over business workflows. Neither is a demonstrated substitute for Duet’s business-app workflows.

The honest bottom line from VentureBeat’s analysis still holds: Duet’s launch makes serious comparisons possible, but provides no head-to-head evidence that it completes the same assignments more accurately or cheaply than its rivals. That evidence will come from independent testing over the coming months.

Honest Limitations You Should Know

No product this new is without rough edges, and Anthropic has been unusually candid about several of them. Here is what to watch.

  • It can misunderstand you. Anthropic acknowledges that the preview can misread conversations and assign the wrong urgency to an issue. A proactive agent that acts on a misunderstanding is worse than no agent at all — so keep approvals switched on and double-check anything consequential.
  • You cannot see its browser. Duet cannot yet display the browser in its cloud computer to you, which limits visibility into part of its work. For trust, this is a real gap: you are approving actions you cannot fully watch being prepared.
  • Background checks are periodic, not continuous. Checks every few hours mean Duet is not supervising your business in real time. Time-sensitive workflows — anything where a 3-hour delay costs you money — should not rely on Duet alone.
  • Background work eats your allowance. The automated checks consume part of your usage limits. Heavy users may find their allowance evaporating on monitoring rather than on the tasks they care about.
  • No Team or Enterprise path yet. Businesses that need audit logs, custom retention controls, or SSO have no Duet route right now. Those governance features sit under Enterprise pricing for Claude generally — do not assume the individual preview carries them.
  • US-only waitlist. The research preview is limited to the United States, with no confirmed timeline for international expansion. Readers outside the US will have to wait.
  • No public benchmarks or cost data. Anthropic disclosed no task-level performance numbers or quantified productivity gains. Every productivity claim right now is marketing until independent testing verifies it.

Who Should Use Claude Duet (and Who Should Skip It)

Claude Duet is a strong fit for US-based freelancers, consultants, and small-business owners who already pay for Claude Pro or Max and juggle the same operational drudgery every week: invoicing, follow-ups, scheduling, payroll, and basic marketing. If your work lives in QuickBooks, Stripe, HubSpot, Gmail, and Slack, Duet’s connector-first design was built for you. It is also worth a look for operations managers at small companies who want to test whether a proactive agent can absorb the attention tax of running a business.

Claude Duet is probably not for you if you are outside the United States (you cannot get in yet), you need enterprise governance like audit logs and SSO (no Team/Enterprise path), you want a free-tier experiment (there is none — $20/month minimum), or your workflows are time-critical down to minutes rather than hours (Duet checks every few hours). And if you are hoping for an agent you can set loose without supervision — send the emails, make the payments, sign the contracts — that is explicitly not what this product is. The approval boundary is the product.

How to Get Access to Claude Duet

Access is currently gated three ways: geography, plan, and waitlist.

  1. Be in the United States. The research preview is US-only for now.
  2. Subscribe to Claude Pro or Max. Pro costs $20 per month ($200 per year), and Max starts at $100 per month. Duet is included with the subscription during the preview — there is no separate Duet charge announced.
  3. Join the waitlist. Anthropic is onboarding preview users gradually. Sign up through the Duet app or the Claude website and wait for your invitation.

Team and Enterprise subscribers cannot access Duet yet, and Anthropic has not announced a timeline for expanding to those plans. If you are evaluating this for a company rather than for yourself, the honest move is to test on an individual Pro plan first — bounded tasks with checkable results, like briefing preparation or invoice-follow-up drafts — before building anything consequential on it.

Final Verdict: Is Claude Duet Worth It?

Claude Duet is the most thoughtfully constrained AI agent launched so far. That is both its strength and its limitation. Anthropic refused to build a fully autonomous business agent — and in a month when websites are actively blocking AI agents and agent privacy scandals are making headlines, that restraint looks wise.

What Duet does well is notice. The daily briefings, the periodic checks, the cross-app awareness — this is the unglamorous, high-value work that no chatbot can do, because no chatbot is watching when you are not typing. For a small-business owner, catching one wrong-date property showing or one overdue invoice per week could pay for the $20 subscription several times over.

What remains unproven is reliability at scale. No independent benchmarks, no task-level cost data, an invisible browser you cannot audit in real time, and an admission that it sometimes misunderstands urgency — these are preview-stage rough edges that would be disqualifying in a finished product. The usage-allowance question is equally real: background monitoring consumes the same allowance as your actual work, and Anthropic has not given users the tools to budget that trade-off precisely.

Our take: if you are a US-based Pro or Max subscriber, joining the Claude Duet waitlist is a low-risk experiment with a genuinely new interaction model. If you are buying it today as the operational backbone of your business — hold off. Let the preview mature, let independent testing publish real accuracy numbers, and let Claude Duet’s governance features catch up. The proactive-agent era is real, but Claude Duet is version 0.9 of it, and the best move is to watch closely rather than commit fully.

The space is moving fast. The agent-infrastructure layer underneath products like Duet — cheaper search APIs built for agents, like Exa 3.0’s new $4 tier, and voice-agent stacks like Voxtral 3 and Eleven Realtime — is evolving week by week. Duet is the first serious professional agent from the most safety-conscious lab in the industry. It deserves your attention, even if it does not yet deserve your full trust.

Frequently Asked Questions

What is Claude Duet?

Claude Duet is Anthropic’s proactive AI agent, launched October 21, 2026. It monitors your connected apps, sends daily briefings, and completes multi-step work on its own cloud computer — asking for your approval before sensitive actions. It is available in research preview to US-based Claude Pro and Max subscribers through a waitlist.

How much does Claude Duet cost?

There is no separate charge for Duet during the preview — it is included with Claude Pro ($20/month or $200/year) and Max (from $100/month). However, its background monitoring and tasks consume part of your existing usage allowance.

What model powers Claude Duet?

Duet runs on Claude Opus 5.5, which Anthropic says was chosen for its judgment in deciding what needs your attention and when to ask for approval.

Is Claude Duet available outside the US?

Not yet. The research preview is limited to the United States, with no confirmed timeline for international expansion.

How is Claude Duet different from Meta’s Muse or OpenAI’s Dots?

All three are proactive agents with cloud computers, but Duet targets professionals and small businesses with business-app connectors (QuickBooks, Stripe, Gusto, HubSpot), launched with a narrower US-only, subscriber-gated rollout, and bundles into Claude’s $20/month Pro plan — undercutting Dots’ reported $100 entry while lacking Muse’s free tier.

Can Claude Duet make purchases or send emails without asking?

No. Claude Duet must present the intended action and request approval before sending messages, making payments, signing documents, or deleting anything. Securities transactions and inter-account transfers are prohibited even with authorization.

Leave a Comment

Your email address will not be published. Required fields are marked *