Anthropic has opened a research preview of Claude Duet, a new app the company positions not as another chatbot but as an agent that works on your behalf while you are away. The launch was announced on October 21, and the app is built on Claude Opus 5.5, available as dedicated iOS and Android applications as well as through the claude.ai website.
The pitch is straightforward: Duet compiles a daily brief of what matters across your connected apps, flags items that need attention, and continues executing delegated tasks in the background even after your phone or computer is switched off. To make that possible, each Duet session runs in its own virtual machine with a browser environment, so the work happens on cloud hardware rather than on the device in your hand.
Anthropic is aiming the product at small businesses and professionals. In its examples, Duet digs through Gmail and Slack for messages that slipped through, drafts replies, chases unpaid invoices in QuickBooks or Stripe, builds financial or sales forecasts, assembles dashboards, and can even develop small web applications. In one early-tester anecdote, a brand strategist said Duet spotted an upcoming client meeting from her calendar, prepared briefing material, and proposed rescheduling because the meeting collided with her flight.
On the integration side, Duet can call on more than 6,000 connectors and plugins from the Claude Marketplace, covering Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, HubSpot, Canva, Shopify and DocuSign, among others. It can also complete online payments through Link or Shop Pay after user confirmation, using virtual card numbers so real card details are never exposed.
The safety design is where Anthropic most wants to differentiate. The company says Duet never touches user passwords or login tokens, and that any action involving sending a message, making a payment, signing a document or deleting content requires explicit user approval. A dedicated classifier and safety system prompt inspects outgoing messages and refuses requests to share the most sensitive categories of data, such as credit card numbers and government ID numbers; health and personal finance information only passes when the user explicitly instructs it to.
For now the rollout is deliberately narrow: a waitlist for subscribers to the Pro and Max plans in the United States. Duet shares the usage allowance of the existing Claude subscription, and background work draws down that allowance too, which is why Anthropic has it check in every few hours rather than polling continuously. The company also lists the current limitations candidly: Duet may misread conversations, misjudge how urgent something is, handle personal errands less well than work tasks, and cannot yet show users what its browser is doing.
Context matters here. Meta's Muse takes tasks over WhatsApp, xAI's Grok Bot can split work across multiple agents, and OpenAI's Dots reaches users through ChatGPT, Slack and Teams. All of them promise cloud computers that keep working while you are gone. Background operation is no longer the selling point; whether the task actually gets done, and how much supervision it needs, is where this category will be decided.
The deeper shift is that the evaluation of AI products is moving from how well a model answers to whether it can close a loop: connectors, permissions, payments, signatures, audit trails and exception handling, all engineered rather than improvised. Whoever makes delegated work boring and reliable wins the category. Duet is Anthropic's entry ticket into that race, and it is buying in later, and with more visible guardrails, than most of its rivals.
Comments (0)
Log in to join the discussion
Log InNo comments yet