AI Industry

Anthropic Launches Claude Duet, a Proactive AI Agent App

Oct 22, 2026·4 min read·By Rahish Saifi

An unbranded smartphone in a wooden stand on a bakery counter showing an AI assistant app with message and approval cards, beside a notepad and pen, coffee cup, card reader and croissants, as a baker stocks bread shelves behind

Anthropic on October 21, 2026 opened a research preview of Claude Duet, a standalone app where Claude acts as a proactive agent for professionals and small business owners. Instead of waiting for prompts, Duet surfaces what needs attention, keeps working when you step away, and asks before it acts on your behalf—positioning Anthropic against personal-agent rivals including OpenAI's Dots and Meta's Muse, as CNBC and Reuters frame the launch.

Key facts

Item	Detail
Announcement	October 21, 2026 — Anthropic blog post, product page, and coverage from CNBC and Reuters
What it is	Standalone proactive agent app for professionals and small business owners
Where	iOS, Android, and claude.ai (separate from the main Claude app during preview)
Access	Waitlisted research preview for US users on Pro and Max plans; not on Team or Enterprise yet
Model	Claude Opus 5.5 only — no model picker
Usage	Shares your plan's limits with the rest of Claude; background checks count toward them
Context	Carries over Claude memory, skills, and connected apps
Infrastructure	Cloud virtual machine with its own browser so work continues when your devices are off
Integrations	6,000+ connectors and plugins (e.g. Gmail, Slack, QuickBooks, HubSpot, Shopify)
Payments	Purchases via Stripe Link and Shopify Shop Pay with per-purchase approval and a one-time card (Anthropic)

What Claude Duet is

Anthropic built Claude Duet for professionals and small business owners who juggle more than they can track. You message it by typing or voice in one ongoing thread (or separate channels that share memory). Each morning it sends a short brief across connected apps; through the day it reaches out only when something needs you.

CNBC contrasts that workplace focus with Meta's more consumer-oriented Muse and OpenAI's Dots, which straddles enterprise and consumer markets. Anthropic says personal errands such as booking tickets are not “quite as strong” yet. During preview, Duet is included in Pro and Max at no extra cost and lives in its own app, apart from the main Claude client.

How it works

Duet inherits your Claude memory, skills, and connectors, then shifts the workflow: it monitors accounts, drafts work, and notifies you when a task finishes or needs a decision. It checks connected apps every few hours—not continuously—to avoid burning plan limits; urgent items can still land between checks.

Larger jobs return as editable files (briefs, models, decks, dashboards) in the Make tab; the Tasks tab shows active work and pending input. A cloud virtual machine with its own browser keeps tasks running after you close the app, and Duet can browse sites that lack marketplace connectors.

Reuters cites 6,000+ Claude Marketplace integrations, including Gmail and Slack. Anthropic highlights QuickBooks and Stripe invoice follow-ups, Gusto payroll prep, HubSpot email drafts, Docusign flows, Canva campaigns, and Shopify store tasks—plus purchases through Stripe Link and Shop Pay after you approve each charge.

Safeguards and limits

Duet must get your OK before it sends a message, makes a payment, signs a document, or deletes anything. Trading securities, moving money between accounts, and accessing someone else's account stay forbidden even with approval. You can disconnect apps at any time.

Preview limits matter: no model choice (Opus 5.5 only), no watching or taking over Duet's browser, no phone calls or texts to others (email from connected accounts is allowed with approval), Microsoft 365 not yet supported, and Slack only with limits. Anthropic also warns Duet can misread threads or misjudge urgency—review drafts carefully.

Safety work draws on Claude Code safeguards, managed-agent credential handling, browser-use defenses, and external penetration testing, plus a new classifier aimed at blocking leaks of card numbers and government IDs.

Who can get it

The waitlisted research preview is US-only for Pro and Max subscribers; Team and Enterprise are not eligible yet. Invitations ship in waves, and the email on your waitlist entry must match your Claude and mobile sign-in. When access opens, you install the standalone iOS or Android app or use claude.ai, following the invitation email on the phone tied to that address.

New admittees get one full usage-limit reset through November 30, 2026. Background briefs and check-ins count toward the same five-hour and weekly caps as regular Claude chat, so Anthropic spaces account checks to every few hours rather than running them nonstop.

Why it matters next to Dots and Muse

CNBC frames Claude Duet as a direct answer to OpenAI's always-on Dots and Meta's Muse. Reuters says Muse drew millions of downloads and that Dots agents pursue user goals across apps such as Slack and Microsoft Teams; CNBC reports Mark Zuckerberg said more than 10 million people in the U.S. and Canada have set up Muse personal agents. Anthropic's pitch is proactive judgment on top of Opus 5.5—briefs, background monitoring, and human approvals—rather than another passive chat window.

The preview is candid about gaps: no visible browser, possible urgency mistakes, and work-first strengths over personal tasks. CNBC adds that competing agent products from Meta and OpenAI have already shown traction, raising the bar for what “always on” assistance must deliver.

For US Pro and Max users already invested in Claude memory and connectors, Duet is Anthropic's step into that race—an agent that schedules meetings, builds forecasts, and assembles dashboards while you focus elsewhere, as long as you stay inside usage limits and keep approving the consequential steps.