Claude Duet (Anthropic)

What Claude Duet (Anthropic) does

Anthropic's proactive 'chief of staff' agent, in research preview from 21 Oct 2026. It's a standalone app for iOS and Android, and on the web at claude.ai, separate from the main Claude app. Instead of waiting for a prompt, it sends a morning brief on what needs you across your connected apps. It checks those apps every few hours and messages you only when something needs attention. You hand it tasks by text or voice in one ongoing thread, with per-project channels. It works on its own cloud computer with its own browser, so your devices can be off, and returns finished work (docs, sheets, decks, dashboards, forecasts, prototypes). It uses your existing Claude memory, skills and connectors, including Gmail, Slack, QuickBooks, Stripe, Gusto, HubSpot, Docusign, Canva and Shopify from the 6,000+ in the Claude Marketplace. It asks before it sends a message, pays, signs or deletes anything, and some actions are blocked even with approval (trading securities, moving money between accounts). Purchases go through Stripe Link virtual cards or Shop Pay. Built on Claude Opus 5.5 and aimed at professionals and small-business owners.

Good for

• Once you're off the waitlist, connect only Gmail and Google Calendar for the first week and judge the morning brief: did it surface the right 3-5 things, and what did it miss or flag wrongly?
• Hand it one real admin task by voice (e.g. 'find unpaid invoices in QuickBooks/Stripe and draft reminders'), then review exactly what it shows you at the approval step before letting anything send
• Ask it to build a one-page daily dashboard from two connected tools and watch how much of your Pro/Max usage the background refreshes consume over a few days

Honest take

Some claims not independently checked

• Fact or hype verdict on Anthropic: Claude for Legal artifacts gallery: worth a look for the pattern rather than the legal content: start from a template, give clear business context, and ask for an interactive tracker with owners and status, then spin a line item into a document. That pattern works for any department. Source video
• Fact or hype verdict on Lydia Hallie (Anthropic): Maximizing the value of Claude Code, on token costs, caching and context (re-upload, video notes): worth watching for anyone with a Claude Code or API bill: concrete, official advice on cutting cost and keeping the agent sharp. Source video
• Fact or hype verdict on Jess Yan (Anthropic) on building long-running agents with Claude Managed Agents: re-upload of Peter Yang's podcast: worth the time for the ideas (harness, async self-recovering loops, evals mid-run, start with one agent), but watch the original on Peter Yang's channel, not this re-upload. Source video
• Verified: US-only, Pro/Max waitlist, iOS/Android plus web, shares plan usage limits, approve-before-acting on send/pay/sign/delete, credentials kept from the model (Claude Managed Agents design), auto-mode action classifier, a new sensitive-info classifier, pen-tested and covered by the bug bounty.
• Anthropic itself says it may misread threads, miscalibrate urgency, miss things that arrive between its check-ins (every few hours, not real time), is weaker on personal errands, and can't yet show you its browser. 1Password logins and a Shopify storefront editor are 'coming soon'.
• The example stories (Pro Homes, Shahs of Kabob, a brand strategist) and the polished demo business ('Marlo') are marketing; there are no independent reviews beyond early press first looks.
• Not available in the UK yet (no UK date announced).
• How it differs: the main Claude app (which absorbed Cowork in Sept 2026) waits for you and does background and scheduled jobs on request, while Duet comes to you first.
• Claude Code is for coding.
• It competes with OpenAI Dots, Meta Muse and Google Gemini Spark (always-on personal agents); ChatGPT agent is closer to on-demand task running.
• Agents that read your email and act for you carry prompt-injection and data-sharing risk, so keep approvals on.

What we couldn’t verify

• Fact vs hype: Officially announced by Anthropic on 21 Oct 2026.
• Third-party VentureBeast.Tech reported it as 'Claude Loop' in the Claude app from 17 Oct before the rename (not confirmed by Anthropic); its claim that check-ins run as Claude Code Routines is the agent's own account.

Our Hype or Fact? checks

• Partly factClaude Duet: Anthropic's proactive agent app 21 Oct

Compare

Personal AI agents compared: UK and EU availability, open source or hosted, bring your own model, approvals and price.

Pricing

Research preview by waitlist, included in Claude Pro ($20/mo, or $17/mo billed annually) and Max (from $100/mo) for US subscribers only; Team, Enterprise and Free plans are not included yet, and it isn't available in the UK yet. No separate fee: it shares your plan's usage limits, and its background checks and big tasks (decks, financial models) use them up faster.

Prices as listed by the maker when we last checked (2026-10-21); usually in US dollars. Check the website for current prices.

Latest update

• 2026-10-21: Research preview announced (waitlist; US Pro and Max; iOS, Android, claude.ai).
• 25 October 2026: Anthropic published Claude Duet's safety, privacy and security design (23 Oct): approval cards plus an intent check outside the model, hard limits on trading and moving money, passwords and sign-in tokens kept away from the model, one-time cards for purchases.

Is Claude Duet (Anthropic) worth it? Quick answers