Anthropic launches Claude Duet, an agent that works in the background and messages you first

Anthropic has opened a research preview of Claude Duet, a new app in which its Claude model acts as a standing assistant that looks for work on its own, rather than waiting to be asked.

“Claude Duet is a new app where Claude works as a proactive agent,” the company said in its announcement on Wednesday, US time. It is aimed at professionals and small business owners who, in Anthropic’s words, have more tasks than they can track. The app is available through a waitlist, starting with Pro and Max subscribers in the United States, and is included in those plans at no extra charge.

What it does

Each morning Duet sends a short brief on the most important items across a user’s connected apps, such as email and calendars. Through the day it checks those apps periodically and messages the user only when something needs attention. Users can hand it tasks by typing or speaking in one running conversation, from scheduling meetings to building dashboards, sales forecasts, prototypes and simple web apps.

The agent runs in its own virtual machine in the cloud, with its own browser, so the user’s laptop or phone does not need to stay on. It carries over memory, skills and connected apps from the person’s Claude account, and works with more than 6,000 connectors and plugins from the Claude Marketplace, among them Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva and Shopify. Anthropic’s examples include chasing unpaid invoices and drafting job posts. It can also buy things for a business through Stripe’s Link wallet, which uses a virtual card, or check out with Shop Pay.

Guard rails

Before it sends a message, makes a payment, signs a document or deletes anything, Duet shows the action and waits for approval. Some actions, including trading securities and moving money between accounts, are off limits even with approval. Anthropic said the agent never directly sees passwords or sign-in tokens, that every message is screened by a separate classifier that rejects material such as card numbers and government IDs, and that the product has been through red-teaming and an external penetration test.

The company is frank about the limits of an early version. Duet may misread a thread or misjudge what is urgent, is less reliable at personal errands such as booking tickets, and uses up plan limits faster than an ordinary chat because its background checks count against them. Users cannot yet watch its browser.

The race for always-on agents

Duet arrives as rivals push the same idea. Meta’s Muse agents and OpenAI’s Dots are pitched as assistants that act for users, and in India, Bengaluru-based Emergent offers Wingman, which works across email and calendars through WhatsApp and Telegram.

Indian users will have to wait. Anthropic has given no date for wider access. The launch also lands a week after CERT-In warned that autonomous AI agents under testing had got into Indian companies’ systems without permission, a reminder of why the approval gates matter.