Anthropic launches Claude Duet proactive AI agent in US research preview

Anthropic introduced Claude Duet on Wednesday, 21 October, adding a personal agent to a field that already includes OpenAI's dots, Meta's Muse and Google's Gemini Spark. The product is in a research preview, and the company is letting people sign up through a waitlist. According to The New York Times, the start-up Instinct, which raised $1 billion in its latest funding round, also competes in the space.

What the agent does

Duet is built to start work before a user asks for it. Each day it produces a short briefing of action items pulled from connected apps, then checks those accounts periodically for developments that need a response. Users can decide how often it sends messages and which topics it flags. Anthropic also says the agent can take on bigger jobs, including building dashboards with dozens of tools, producing financial or sales forecasts, prototyping and building web apps. In its own words, the product works like this:

It comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away.

Duet runs on Opus 5.5, one of the models Anthropic offers through Claude, and it does not require an open Claude chat. It operates inside its own virtual machine with a browser, which lets it keep going in the background. Anthropic is releasing standalone Android and iOS apps alongside access on the web at claude.ai. Before it sends a payment, signs a document or deletes anything, Duet asks permission. Integrations with Stripe let it check unpaid invoices and draft payment reminders, while Link and Shop Pay let it make purchases once a user approves.

Access, pricing and limits

For now the preview is open only to US subscribers who join the waitlist, and invitations will go out in waves. Engadget reports that subscribers on the $20 per month Pro plan will receive access once Duet reaches general availability. OpenAI's dots, by comparison, are limited to its Pro tier, whose plans start at $100 per month. Duet is not available on Team or Enterprise plans yet, and Anthropic plans to expand to those subscriptions later. The feature costs nothing extra, but its background work counts toward existing usage limits, which Anthropic says could be reached "sooner than with chat alone." Anthropic explains the check schedule this way:

Duet checks your connected accounts every few hours, because checking nonstop would use up your plan's limits much faster.

Security and the rivals

Anthropic says Duet uses a new classifier and a dedicated system prompt meant to stop it leaking sensitive information. The company says the agent will never directly see passwords or sign-in tokens. The Verge reports that the classifier is designed to reject requests for the strictest tier of information, such as credit card numbers. The New York Times reports that Anthropic built Duet for work and says it is less capable at personal errands, while OpenAI and Meta pitch their agents for everyday tasks like booking reservations.

Early version caveats

The Verge quotes Anthropic describing Duet as an "early version" that "still might misread a thread or miscalibrate what's urgent." Anthropic adds that it is still testing when the agent should message users:

We're continuing to test its judgment about when to message you, and we'll keep tuning it based on what testers tell us.

VentureBeat notes that the initial release targets professionals and smaller businesses, and that the announcement does not extend Duet to Team or Enterprise subscriptions, does not announce a Duet API and does not set an enterprise rollout schedule. It also observes that the periodic checks do not amount to continuous, real-time supervision.