
Anthropic’s version of the always-on AI agent arrived on 21 October, six weeks after Meta launched Muse and three after OpenAI unveiled Dots. Claude Duet, a standalone app for iOS and Android that also runs on claude.ai, sends a morning brief on what needs attention across a user’s connected apps, keeps working on a cloud computer after the phone is put down, and stops for approval before it sends, pays, signs or deletes anything.
The change is in who starts the conversation. The Claude app waits for a prompt; Duet watches inboxes, calendars and business systems and messages the user when it judges that something needs them. Anthropic has pointed it at tools many marketing teams already pay for, including HubSpot, Canva, LinkedIn, Slack and Gmail. For now, though, it is a waitlisted preview open only to US subscribers on the Pro and Max plans, which keeps it out of reach of UK marketers and of any team buying Claude through a Team or Enterprise contract.
Claude Duet is built around Claude Opus 5.5, the model Anthropic released on 22 September, and Anthropic describes its job as deciding what needs the user, what can wait and when to check first. It inherits memory, skills and connected apps from the user’s existing Claude account, so set-up is light, and every exchange happens by text or voice in one ongoing thread rather than in separate chats.
Two design choices separate it from the main app. It runs on a virtual machine with its own browser, so a task described in a short message carries on in the background without the user’s laptop or phone switched on, and a notification arrives when the work is done or needs review. It also checks connected accounts every few hours rather than continuously. Anthropic says constant checking would use up plan limits too quickly, and that it is still tuning the agent’s judgement about when to interrupt; users can tell it what they want to hear about and how often.
The tasks Anthropic lists run from scheduling meetings to building dashboards, sales and financial forecasts, prototypes and web apps. Its customer examples are company-selected and small in scale. Amin Ebrahimi, who owns the South Florida restaurant business Shahs of Kabob and employs about 110 people, has used Duet for accounting, customer support and marketing, and said he no longer needs “to open up 100 tabs”. Mya Dalton, a brand strategist, told Anthropic that Duet prepared a brief for a call with a prospective client and pointed out that the call fell just before her flight.
Claude Duet can reach the more than 6,000 connectors and plugins in Anthropic’s Claude Marketplace, and several of Anthropic’s own examples are marketing and revenue work. It can draft follow-up emails from contact and deal history in HubSpot, turn a product launch into a social campaign in Canva, find missed Gmail and Slack messages and help reply, chase unpaid invoices in QuickBooks or Stripe, and draft a job post that goes to LinkedIn once the user has edited and approved it. Shopify store set-up is included, with a fuller visual storefront editor promised later.
For any team weighing it against a customer list, the approval rule is the key guardrail. Duet reads, drafts and prepares on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it intends to do and waits. Some actions are barred even with approval, including trading securities and moving money between accounts, and any connected app can be disconnected. Purchases run through Stripe’s Link wallet, which pays with a virtual card so the real card number is not shared, or through Shopify’s Shop Pay.
Anthropic is also shipping a dedicated classifier and safety system prompt that checks each message for leaks of sensitive data. It is designed to refuse to pass on card numbers and government ID, and to share health or personal-finance details only on explicit instruction. Credential handling follows the design of Anthropic’s Managed Agents, so Duet never directly sees passwords or sign-in tokens. These are design commitments rather than audited results: Anthropic says internal red teams and an external firm have penetration-tested the product, and that it will publish more on its safety design in the coming days.
The company is open about the gaps. Duet may misread a thread or misjudge what is urgent, it is weaker on personal errands than on work, and users cannot yet watch its browser while it works. Its background checks count against the subscriber’s plan limits, and larger jobs such as building a financial model or a deck use more of them.
Claude Duet is available only by waitlist to Pro and Max subscribers in the United States, with no separate charge during the preview. On Anthropic’s US pricing page, Pro costs $20 a month or $200 a year and Max starts at $100 a month. Anthropic has said Duet is not yet available on Team or Enterprise plans and has given no timeline, and VentureBeat noted that the announcement includes no Duet API and no enterprise rollout schedule.
That positioning differs from both rivals. Meta’s Muse, launched on 8 September in the US and Canada, has a free tier alongside $20 and $100 plans and connects to Slack, Stripe, Shopify, Notion and Facebook and Instagram business accounts. Sensor Tower figures reported by Yahoo Finance put it at 10.9 million downloads and 2.5 million daily active users by 16 October. OpenAI’s Dots, unveiled at DevDay on 29 September, runs on GPT-6 Astra, reaches users through ChatGPT, Slack and Microsoft Teams, and is sold on Pro and Business Premium plans, with an admin-enabled beta for Enterprise customers. OpenAI has held Dots back from Pro subscribers in the UK, the EEA and Switzerland.
Anthropic has published no benchmarks, productivity figures or head-to-head comparisons with Muse or Dots. VentureBeat’s assessment is that Duet’s distinction lies in its judgement about when to interrupt, not in capabilities its rivals lack.
Anthropic says it will widen access as the preview progresses and plans to bring Duet to Team and Enterprise customers, but has not said when.