Anthropic is joining the race to build AI agents that work before you ask. Its new Claude Duet app is designed to monitor connected work apps, send a short morning briefing, and keep tasks moving in the background while you step away from your phone or computer.
Duet opened as a research preview on October 21. It is available as a standalone app for iOS and Android and on claude.ai, with access initially limited to US-based Claude Pro and Max subscribers through a waitlist. Anthropic says the app is built around Claude Opus 5.5 and shares the same usage limits as the rest of a user’s Claude plan.
Watch Anthropic’s launch video
A chatbot that comes to you
Duet’s main change is the direction of the conversation. Instead of waiting in a chat window, it checks connected accounts every few hours and reaches out when something needs attention. Anthropic says it can prepare a morning summary, find missed email or Slack messages, draft replies, schedule meetings, check invoices, prepare forecasts, and build dashboards or web-app prototypes.
The agent carries over memory, skills, and app connections from a user’s existing Claude account. It runs on its own cloud-based virtual machine and browser, so background work can continue after the user closes the app. Anthropic says Duet can draw on more than 6,000 connectors and plugins in the Claude Marketplace, including tools such as Gmail, Slack, QuickBooks, HubSpot, Docusign, and Shopify.
That “always on” label needs a caveat: Duet does not continuously poll every service. Anthropic says it checks accounts every few hours to limit how quickly background activity consumes a subscription’s usage allowance. Those checks count against the same plan limits as ordinary Claude chats, and larger tasks use more of that allowance.
The user still approves consequential actions
Anthropic says Duet can read, draft, and prepare work on its own, but will show users what it intends to do and wait for approval before sending a message, making a payment, signing a document, or deleting something. Some actions, including trading securities or moving money between accounts, remain off-limits even with approval. Users can disconnect apps from their account.
The design also reflects the risks of giving an agent access to email, files, and websites. Those sources can contain malicious instructions aimed at the agent, a technique known as prompt injection. Anthropic says Duet uses defenses drawn from its browser-use research, handles credentials so the agent does not directly see passwords or sign-in tokens, and screens messages for sensitive information. The company also acknowledges that defenses are not perfect and says the early version may misread threads or misjudge urgency.
Those limitations matter because a proactive agent is making decisions about what deserves attention and where information should go. Anthropic says health or personal-finance information may be sent only when a user has explicitly instructed Duet to do so, while especially sensitive items such as payment-card numbers and government IDs should be blocked. The company plans to publish more detail about the app’s safety design as the preview develops.
Anthropic enters a crowded agent race
Claude Duet joins a growing category of assistants designed to persist between prompts. OpenAI launched its Dots agents in September, while Meta’s Muse and xAI’s Grok Bot are also competing for a place in users’ daily workflows. The products differ in access, integrations, and how much autonomy they offer, but they share a broader bet: that an assistant which can keep track of work and return with useful progress may be more valuable than one that only answers when opened.
For Anthropic, Duet extends Claude from a model and chat app into a persistent worker. The test is whether its judgment is good enough to be helpful without becoming noisy, draining plan limits, or taking actions a user did not intend. A waitlist preview is not proof that the product can manage that balance at scale; it is a way for Anthropic to learn where its proactive behavior works and where users still need tighter control.
For now, Duet is a US-only preview for Pro and Max subscribers, not a general release. Anthropic is entering the agent game, but the contest will be decided less by who can run in the background than by who can do so reliably, transparently, and within boundaries users understand.