Sign in Subscribe
news

2026-10-21 Claude Duet - Anthropic's answer to Dots, in yet another app

Sebastien Dubois

Sebastien Dubois

— 5 min read

Canonical version: 2026-10-21 Claude Duet - Anthropic's answer to Dots, in yet another app.

On Wednesday (21 October), Anthropic announced Claude Duet, a "proactive agent" in research preview. It runs on Claude Opus 5.5, works in its own cloud computer, and keeps going while you're away.

The idea is simple. The Claude app waits for you to open a chat; Duet comes to you. Each morning it sends a short brief on the most important items across your connected apps, and it messages you when a task is done or when it needs a decision. Anthropic's tagline: "Get back to what needs you."

It's Anthropic's response to Meta Muse (8 September) and OpenAI's Dots (29 September). Thurrott's verdict: Anthropic was slow to enter.

What shipped

  • One ongoing thread. You message it "like a teammate you trust", by text or voice. No prompts, no setup. Quick requests and weeklong projects go in the same thread
  • A cloud computer of its own, with its own browser. Your laptop and phone can stay off
  • Check-ins every few hours. Duet looks at your connected accounts periodically (Engadget says that's to conserve usage limits) and you choose how often it messages you, and about what
  • Your Claude account comes along. Memory, skills and connected apps work from the first message
  • 6,000+ connectors and plugins from the Claude Marketplace: Gmail, Slack, Google Drive, Notion, QuickBooks, Stripe, Gusto, HubSpot, DocuSign, Shopify and more
  • Purchases with approval, through Stripe Link and Shopify Shop Pay. 1Password sign-in comes later

The target is "professionals and small business owners with more on their plates than they can track". Anthropic's examples are very small-business: finding unpaid invoices and drafting reminders, preparing payroll for review, getting contracts ready to send, building forecasts and dashboards. A real estate tester says it caught a misscheduled showing. A restaurant owner uses it for accounting, support and marketing across several locations.

Who gets it, and what it costs

  • Waitlist, US only, for Claude Pro and Max subscribers. Team and Enterprise aren't included yet
  • No extra price. But Duet shares your plan's usage limits with the rest of Claude, Claude Code included, and the background checks count too
  • A separate app on iOS and Android, plus claude.ai on the web, apart from the main Claude app during the preview

Sources disagree on Pro. Anthropic's page puts Pro and Max on the waitlist; Engadget says Pro ($20 per month) users get access at general availability.

Safety design

Duet shows you what it wants to do and waits before it sends a message, pays for something, signs a document or deletes anything. Trading securities and moving money between accounts are off-limits, even with your approval. It never sees your passwords or sign-in tokens (the design follows Claude Managed Agents), and actions go through the same auto mode classifier as Claude and Claude Code, plus Prompt injection defenses. Credit card and government ID numbers are blocked by a separate classifier. Health and personal-finance data only gets shared when you explicitly say so.

Anthropic also lists its own limits, which I appreciate: Duet "may misread a thread or miscalibrate what's urgent", and it can miss something that arrives between two check-ins. Personal errands like booking tickets are weaker, and you can't watch its browser yet.

How it compares

Duet is the last of the big labs to ship. Google's Spark came in May, then Grok Bot, Muse, Microsoft Autopilot and Dots between August and September.

  • vs Dots (OpenAI, 29 September): same idea, a managed agent with its own cloud computer, but from ChatGPT Pro ($100+) or Business Premium, and Pro isn't available in the EEA, Switzerland or the UK. You can open a dot's computer and take over; you can't watch Duet's browser yet. And you can't view or edit individual dot memories, while Duet reuses your Claude memory
  • vs Meta Muse (Meta, 8 September): the cheapest way in (free, then $20 or $100), in WhatsApp, iOS, Android, Mac and the web, US and Canada. A separate gatekeeper approves every outbound action, and its memory is Markdown files you can open. Six weeks of privacy incidents, though, from a 0-day in the Mac app to leaky share links
  • vs Grok Bot (xAI, run by Cursor, 11 August): team-oriented rosters of Bots that talk to each other, from $30 with The X Subscription. All of your Bots share one computer and act as you, which is how one CEO's bank balances ended up in his company Slack
  • vs Gemini Task (Google, 19 May, renamed from Spark in October): the deepest first-party data (Gmail, Drive, Photos, Chrome with saved passwords) from $19.99. Proactive mostly through schedules you define, and not available in the EEA
  • vs Microsoft Autopilot (Microsoft, 25 September, private preview): an enterprise agent inside Teams and Outlook, built on OpenClaw, with its own Entra identity per agent and usage-based pricing nobody has published yet
  • vs OpenClaw: self-hosted, MIT, any model, and it talks to you on WhatsApp, Telegram and a dozen other channels. Memory sits in Markdown files on your machine. Safety is your job though: sandboxing and approvals are off by default
  • vs Hermes Agent: also self-hosted and model-agnostic, with a learning loop that writes its own skills. It runs on a $5 VPS, but expect a real onboarding curve

Short version: Duet trades model choice, channels and memory ownership for zero setup and safety you don't have to configure. Among the hosted agents, it has the clearest hard limits on money and the smallest set of channels.

My take

I'm in Belgium, so I'll be reading about Duet for a while. US only.

The safety model looks right to me: passwords that never reach the model, a classifier on consequential actions, a short list of things it refuses even with approval. That's what OpenClaw users have to build themselves.

Two things hold me back. First, memory. Duet uses your Claude memory, which you can at least view and edit on claude.ai (better than Dots). But I keep my agents' memory in plain Markdown inside my vault, and I run my own agent system on top of Claude Code (AI Agent Memory, AI Agent Portability). Second, the budget. Duet's background checks draw from the same usage limits as Claude Code. With my agents running on Claude Code, that's a cost I'd watch closely.

And then the product story. In September, Anthropic merged Claude Cowork into the main Claude app because, as Cat Wu put it, people didn't want to choose which Claude product to use. About a month later, here's a proactive agent... in a separate app. Maybe that changes after the preview. For now, it's one more Claude to choose from.

The full breakdown (safety model, limitations, comparison tables with Dots, Muse, Grok Bot, Gemini Task, Microsoft Autopilot, OpenClaw and Hermes) is in my Claude Duet note.

That's it for today! ✨

References

About Sébastien

Ready to get to the next level?

Found this valuable? Share it with someone who needs it.

Join 6,000+ readers. Get practical systems for knowledge & AI. Free.

Subscribe ✨

Free: Knowledge System Checklist

A clear roadmap to building your own knowledge system. Subscribe and get it straight to your inbox.

6,000+ readers. No spam. Unsubscribe anytime.

Subscribe