(VIDEO) Anthropic Launches Claude Duet, an Always-On AI Agent App That Works for You

Anthropic on Wednesday unveiled Claude Duet, a standalone artificial intelligence agent app designed to work in the background on behalf of professionals and small-business owners. The launch deepens a fast-moving race with OpenAI and Meta to build AI assistants that can complete tasks on their own rather than simply answer questions.

The San Francisco-based company said Claude Duet will monitor a user's connected apps, deliver a daily briefing and flag items that need attention without waiting to be prompted. It can also take on larger jobs, from scheduling meetings to building financial forecasts and web apps.

The app is launching as a research preview, available through a waitlist on iOS, Android and the web.

"We're opening a research preview of Claude Duet, a new app where Claude works as a proactive agent," the company said.

An assistant you message like a person

Anthropic is pitching Claude Duet as a shift away from the traditional chatbot model, in which users type a request and wait for a reply.

"Claude Duet is a research preview of Claude that you message the way you'd message a person," the company said.

Users can communicate with the agent by text or voice in a single ongoing conversation. Each morning, it sends a brief summarizing priorities from email, calendars and other connected apps. Throughout the day, it monitors those services and surfaces only what it determines needs human attention.

The agent runs on its own cloud-based computer, which means it can keep working after a user closes the app. It also has its own web browser, allowing it to use websites that do not have a direct integration.

In a post on X, Anthropic said the agent "keeps working when you step away, and checks with you before it acts on your behalf."

Claude Duet is powered by Claude Opus 5.5, Anthropic's high-end AI model.

Built for small businesses

Anthropic is aiming the product squarely at professionals and small-business owners, a market it sees as underserved by AI tools built either for large companies or for consumers.

The agent can connect to more than 6,000 apps and plugins through the Claude Marketplace, including Gmail and Slack. Anthropic described examples such as sorting email and team messages, checking Stripe or QuickBooks for unpaid invoices, preparing payroll for review, drafting job postings, routing contracts for signature and building marketing materials.

Completed work, such as research briefs, financial models, slide decks and dashboards, appears in a dedicated section of the app. A separate tab tracks ongoing tasks, and the agent posts updates when a task is finished or needs a decision. Only users can mark tasks as complete.

Claude Duet also carries over memory, skills and app connections from a user's existing Claude account, so it can draw on past conversations and preferences.

Purchases and safeguards

The agent can make purchases through Stripe's Link digital wallet and Shopify's Shop Pay. It uses a virtual card so that users' real payment details are not exposed.

Anthropic said the app is designed to ask for permission before sending messages, making payments, signing documents or deleting anything. Some high-risk actions, such as trading securities or transferring money between accounts, remain restricted even with user approval.

The company said the agent never directly sees passwords or sign-in tokens. Automated systems review proposed actions and are designed to prevent sensitive information, such as credit card numbers and government ID numbers, from being leaked. Users can disconnect any app at any time.

Limitations acknowledged

Anthropic was candid about the product's shortcomings. The company said the agent can misjudge urgency or misread context, and that it is currently less effective at personal tasks such as booking tickets.

Anthropic said it is still adjusting how often the agent checks accounts and contacts users, and that features are expected to change based on feedback from early testers.

Who can use it

Initial access is limited to Claude Pro and Max subscribers in the United States. Users are being admitted gradually from a waitlist.

"Claude Duet isn't available on Team or Enterprise plans yet," the company said, adding that it plans to expand access to those business plans later.

There is no separate charge during the preview. However, the agent's work, including background monitoring and larger tasks, counts toward users' existing plan limits.

A crowded race for AI agents

The launch puts Anthropic in more direct competition with its biggest rivals, which are racing to build AI agents capable of completing work autonomously.

OpenAI last month launched its own always-on agents, called Dots, which pursue user goals across apps such as Slack and Microsoft Teams. Meta's Muse AI agent has drawn millions of downloads and helped fuel investor interest in autonomous AI services, aided by a free tier and Meta's vast user base across Instagram, WhatsApp and Facebook.

The three products share common features, including cloud-based computers, background operation, app integrations and approval steps for sensitive actions. Anthropic is seeking to stand apart by focusing on professional and small-business workflows rather than personal errands, and by including the agent in its $20-a-month Pro plan. Access to some competing agents requires pricier subscription tiers.

Trust and reliability questions

The rise of autonomous agents has raised questions about reliability, security and how much access users should hand over to AI systems. To be useful, agents like Claude Duet require broad permissions to read email, manage calendars and act inside business software.

Some companies have also begun pushing back against AI agents. Yelp and Delta Air Lines have reportedly moved to block autonomous agents from several AI companies, citing concerns about data scraping and violations of their terms of service.

How well the agents handle changing app interfaces, unexpected errors and judgment calls about what deserves a user's attention is likely to determine whether they win the trust of businesses.

Anthropic said it will expand access as the preview progresses and plans to bring Claude Duet to Team and Enterprise customers in the future. The company has not said when the app will be widely available or whether it will eventually carry a separate price.

For now, the launch signals that the competition among leading AI companies is shifting from building smarter chatbots to building assistants that can take work off users' hands, a contest that is likely to define the next phase of the AI industry.