Anthropic Launches Claude Duet, a 6,000-Connector Proactive Agent on Opus 5.5

Anthropic opened a waitlisted research preview of Claude Duet on Oct. 21, a proactive agent built on Claude Opus 5.5 that briefs users each morning and keeps working inside their connected apps. The catch: Duet shares the subscriber’s existing plan limits, and its background checks count against them.

Article Brief

Key Takeaways

5 Points30s Read

• 01Research preview–Waitlist access for Claude Pro and Max subscribers in the US; standalone iOS and Android apps plus claude.ai.
• 02Opus 5.5–The launch confirms a new Opus-class model behind the agent, which decides what needs the user, what can wait, and when to ask first.
• 036,000+ connectors–TECHi counted the integrations Anthropic actually names: 13, every one a business tool — no consumer apps.
• 04Strict gates–Approval required before sending, paying, signing or deleting; securities trading and moving money are off-limits even with approval; it never sees passwords.
• 05Metered agent–Background checks every few hours and larger tasks draw from the subscriber’s existing plan limits.

Anthropic is now the third of the three frontier labs to ship a proactive agent within weeks of its rivals. OpenAI launched always-on “dots” agents last month that pursue user goals across Slack and Microsoft Teams, and Meta’s Muse agent drew millions of downloads, Reuters reported — and now Anthropic has answered with Claude Duet, announced Oct. 21 on the company’s claude.com site. Anthropic’s announcement makes the positioning explicit: Duet is “a new app where Claude works as a proactive agent,” built for professionals and small business owners with more on their plates than they can track. The formula is converging across the industry — a morning brief, background work, and a ping only when a decision is needed — so the fight is now about distribution and trust, not the concept.

The core interaction is a reversal of the chatbot. Instead of waiting for a prompt, Duet “comes to you with what needs your attention”: a short brief each morning on the most important items across connected apps, then check-ins through the day when something needs the user. Tasks can be handed over by typing or voice in one ongoing thread, from scheduling meetings to building dashboards with dozens of tools, creating financial or sales forecasts, prototyping, and building web apps. Duet carries over the subscriber’s memory, skills and connected apps from their Claude account, so setup is minimal. The work happens on a virtual machine with its own browser, so the agent keeps going when the user’s laptop or phone is off, and sends a notification when a task is complete or needs review.

The safety architecture is the most detailed part of the announcement, and reads as Anthropic’s differentiation. Duet draws on the company’s framework for trustworthy agents, the same one that guides Claude Code and Claude in Chrome; credential handling follows the Claude Managed Agents design, so the agent never directly sees passwords or sign-in tokens. An auto-mode classifier approves actions, prompt-injection defenses come from Anthropic’s browser-use research, and the company says red teams and an external firm penetration-tested Duet, which is also covered by its public bug bounty program. A new dedicated classifier and safety system prompt checks each message for sensitive-information leaks: requests to share the strictest tier of information — credit card numbers, government identification — are rejected outright, while health or personal-finance details go through only when the user has explicitly instructed the agent to send them.

Before it sends a message, makes a payment, signs a document or deletes anything, Duet shows the user exactly what it is about to do and waits for approval. A few actions are off-limits even with approval, notably trading securities and moving money between accounts. Users can disconnect any app Duet can access.

Commerce is built in. Duet can check Intuit QuickBooks or Stripe for unpaid invoices and draft reminders, prepare and run payroll in Gusto, and set up and manage a Shopify store. For purchases it uses Link, Stripe’s digital wallet: Duet confirms the item, Link asks the user to approve the spend, and payment goes through a virtual card so the real card number is never shared; it can also check out with Shopify’s Shop Pay. Anthropic says 1Password sign-in support is coming, with the user’s permission.

Claude Duet’s terms at a glance

Detail	What Anthropic says
Access	Research preview by waitlist; US Pro and Max subscribers to start
Platforms	Standalone iOS and Android apps; claude.ai
Model	Claude Opus 5.5
Connectors	Over 6,000 connectors and plugins claimed in the Claude Marketplace; 13 named in the announcement
Metering	Shares the plan’s usage limits; background checks and larger tasks count toward them
Approval gates	Shows exactly what it is about to do and waits for approval before sending, paying, signing or deleting
Hard prohibitions	Trading securities and moving money between accounts are off-limits even with approval
Credentials	Never directly sees passwords or sign-in tokens
Commerce	Stripe Link virtual-card purchases and Shopify Shop Pay checkout with user approval; 1Password sign-in coming

TECHi’s audit of the announcement surfaces a deliberate narrowing. Anthropic claims more than 6,000 connectors and plugins in the Claude Marketplace, but names only 13 integrations across the whole announcement: Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva, Shopify, Link, Shop Pay and 1Password. Every one is a business tool. Not one consumer app — no airline, no food delivery, no music service — appears in the examples. That matches the company’s own caveat: Duet is built around work, and personal errands such as booking tickets are not yet its strong suit. This is a work agent with a work-app distribution strategy, and the connector count is a marketplace figure, not a measure of what Duet does on day one.

Duet’s internal testing is part of the announcement’s evidence. Boris Cherny, who leads Anthropic’s Claude Code team, talks through his mental to-do list with Duet on walks instead of saving the work for his desk; Chelsea Larsson, who leads model and product behavior design, hands it her to-do list before leaving to be with her kids and gets an evening report; and Karan Ahuja, who leads the Duet team, uses it to schedule user interviews, prototype, and even ship code. The company also gave small business owners early access, and its examples are chosen to make one point: the agent is supposed to catch things unprompted, not just execute orders.

The interaction model matters as much as the features. Duet keeps one ongoing thread for everything — a quick reply and a weeklong project start in the same place — and accepts typed or spoken input. It searches past conversations so users do not repeat themselves, and it inherits the memory, skills and connected apps from the user’s Claude account. What changes, in Anthropic’s framing, is the direction of initiative, with the agent approaching the user first. For a research preview, that is also the risk: an agent that messages the user on its own judgment is useful only if its judgment about what deserves attention is good, and Anthropic says it is still tuning exactly that based on tester feedback.

The economics are the part subscribers will feel first. During the research preview, Duet is included in Pro and Max plans at no additional charge, but it shares the plan’s usage limits with the rest of Claude — and it uses them differently. Background checks count toward those limits, and Anthropic says it checks connected accounts every few hours precisely because nonstop checking would burn through plan capacity faster. Larger tasks such as building a financial model or a deck require more model turns and background work. Anthropic’s own framing is that users may reach their limits sooner than with chat alone. The strongest counterargument to the launch’s promise is right there in the announcement: an always-on agent that taxes a fixed usage meter has an incentive problem, and Anthropic says it is still tuning when the agent should message the user.

There is also a sequencing question. Anthropic says it built Duet for professionals and small business owners — and the launch’s company-supplied examples are all small businesses: a real-estate broker whose agent caught a wrongly dated property showing (“exactly the kind of thing I want caught without having to ask,” broker Jess Provencher said), a restaurant owner who ran accounting and invoices through it, a brand strategist whose agent prepped a client brief and spotted a scheduling clash before a flight. Yet the research preview starts with individual Pro and Max subscribers in the US. The announcement’s stated rollout does not include Team or Enterprise plans. For a company that has spent 2026 pushing Claude into large enterprise deployments, the flagship agent debuts on consumer plans first — a notable choice, and one that leaves the business-customer pricing question open.

Anthropic is candid about the preview’s limits: Duet may misread a thread or miscalibrate what is urgent, it cannot yet show the user its browser, and it lives in its own app, separate from the main Claude app. What the company learns from early users will shape how the two work together.

What to watch

Three things decide whether Duet matters beyond launch week. First, the waitlist conversion: when general availability arrives, and whether Team and Enterprise plans get access with business pricing. Second, the metering: whether background checks prove cheap enough that subscribers leave the agent on, or whether plan caps force Anthropic to sell a separate agent tier. Third, the model: Opus 5.5 is now confirmed as the engine behind Duet, and its broader rollout will tell us whether Anthropic’s next flagship model ships first inside an agent rather than a chat window.