Anthropic's Claude Duet Is an AI Assistant That Works While You Sleep — Here's What It Actually Does

I've been watching AI assistants for years, and every few months someone announces "the assistant that finally does your work for you." Most of them fizzle. This one is worth a closer look anyway: on October 21, 2026, Anthropic launched Claude Duet, a proactive AI agent that doesn't wait for you to type a question. It watches your connected apps, sends you a morning brief, and keeps working in the cloud after you've closed your laptop.

It's in research preview, it's US-only, there's a waitlist, and Anthropic itself admits it can misread emails and misjudge what's urgent. So no, it's not magic. But Duet is the clearest sign yet of where the AI industry is heading: from chatbots you talk to, to agents that message you. Here's what Claude Duet is, what it can actually do, where the guardrails are, and why you should care even if you can't get it yet.

What is Claude Duet?

Regular Claude waits for you to open a chat and ask something. Claude Duet is a separate, standalone app where Claude acts as a proactive agent — it comes to you with what needs your attention, completes tasks you assign, keeps working when you step away, and checks with you before it acts on your behalf.

In practice, that means three things, according to Anthropic's announcement:

• A morning brief. Each day starts with a short summary of what matters across your connected accounts: flagged emails, calendar items, unpaid invoices, that sort of thing.
• Background monitoring. Duet checks your connected apps every few hours (not continuously, Anthropic says nonstop checking would burn through your usage limits too fast) and pings you only when something needs a decision.
• Its own computer in the cloud. Duet runs inside a dedicated cloud virtual machine with its own browser. A task keeps progressing after you close your laptop or turn off your phone; you get a notification when the work is done or needs your review.

The whole thing is built around Claude Opus 5.5, Anthropic's top model, and it inherits the memory, skills, and connected apps from your existing Claude account — so there's no rebuilding your setup from scratch. Conversations happen in one ongoing thread rather than scattered across fresh chats.

Photo: JESHOOTS / Wikimedia (en.wikiversity) / CC0 — illustrative; no freely-licensed photos of Anthropic's offices or the Duet app itself exist on Wikimedia projects, so images in this article are illustrative.

How Duet actually works under the hood

The connection layer is the Claude Marketplace, which Anthropic says offers more than 6,000 connectors and plugins. The named examples include Gmail, Slack, Google Calendar, Google Drive, QuickBooks, Stripe, Shopify, HubSpot, Gusto, DocuSign, Notion, Canva, and Microsoft 365. That's the plumbing that lets Duet read your inbox, watch your store, or pull numbers from your accounting software.

When you hand it a job — by typing or by voice — Duet figures out which apps and steps the task needs, does the work in its cloud VM, and comes back when it's finished or stuck. Anthropic lists concrete jobs: finding missed Slack and Gmail messages and drafting replies, checking QuickBooks or Stripe for unpaid invoices, running payroll in Gusto with hours and bonuses filled in, preparing a DocuSign agreement for signature, building dashboards and financial forecasts, even prototyping and building web apps.

One detail I found genuinely interesting: Duet can also spend money on your behalf. It can buy things through Stripe's Link digital wallet or Shopify's Shop Pay, paying with a one-time virtual card so your real card number is never shared. Anthropic says Duet confirms the item with you first, then the payment provider asks you to approve the spend. Anthropic also says sign-in support for sites saved in 1Password is "coming soon" — it's not live yet.

The safety guardrails (and why they matter)

Here's where Anthropic is clearly trying to get ahead of the obvious fear: an AI with the keys to your business apps is an AI that can send the wrong email, pay the wrong bill, or delete the wrong file. Duet's answer is a set of hard approval gates:

• It asks first before sending a message, making a payment, signing a document, or deleting anything.
• Some actions are blocked outright, even if you approve: trading securities and moving money between accounts are fully prohibited.
• It never sees your passwords or logins when you connect accounts.
• A dedicated classifier and safety system prompt checks each message and guards against sensitive information leaking out. Anthropic says details about health or personal finances are shared only when you explicitly instruct Duet to send them.

That's a sensible design. The real risk is approval fatigue — ask someone to approve forty things a day and they'll start clicking "yes" without reading, which is how mistakes happen. The every-few-hours cadence is at least an attempt to keep the interruption volume down.

Photo: server racks at NERSC (National Energy Research Scientific Computing Center) / Wikimedia Commons / CC0 (public domain) — illustrative of the cloud infrastructure behind AI agents like Duet.

What it can't do yet — Anthropic's own honesty

Credit where it's due: Anthropic isn't pretending this is finished. The company says the research preview version can misread a conversation, misjudge what's urgent, and miss items between account checks. It also admits Duet is aimed at work — professionals and small business owners with more on their plates than they can track — and is weaker at personal tasks like buying tickets.

A few more honest limitations worth knowing:

• You can't watch it work. Users cannot yet view Duet's browser, so there's limited visibility into how it performs certain tasks. That's a big deal for trust — you're approving outcomes without seeing the process.
• Background work costs you. Duet shares your plan's usage limits, and its background checks count against them. Heavy monitoring means less allowance left for your own chats.
• It's US-only and waitlisted. Only Pro and Max subscribers in the United States can join the waitlist for now; Team and Enterprise plans aren't included yet, and there's no published timeline for anywhere else.

These aren't footnotes — they're the difference between a demo and a tool you'd actually trust with your business. I'll be watching how many of them get fixed before general release.

The agent race: Duet vs Dots vs Muse

Claude Duet doesn't exist in a vacuum. The AI labs are now openly racing to own the "agent that works while you're away" slot:

• Meta's Muse: a proactive agent that reaches out before you ask. (Its free right now. Go to Muse.AI and join, and use coupon code "S8KP4Y" to get 1 Billion free tokens that don't expire)
• OpenAI's Dots: the same idea, proactive and persistent.
• Claude Duet: Anthropic's entry, pitched specifically at professionals and small business owners.

Each has its own VM-with-a-browser setup and its own version of "check with you first." The real competition isn't the feature list — it's who becomes the default entry point for your work context. Whoever owns the agent that knows your inbox, calendar, and invoices owns the layer between you and everything you do. That's why Duet's business-first framing — invoices, payroll, contracts — is a deliberate differentiator, not a limitation.

Who can get it, and what does it cost?

Right now: US-based Claude Pro and Max subscribers, via a waitlist, in standalone iOS and Android apps and on the web at claude.ai. There is no separate price — during the research preview, Duet is included in Pro and Max plans at no extra cost, and it draws from your existing usage limits. Team and Enterprise accounts aren't included yet, and Anthropic hasn't published a date for widening access, so anyone outside the US should treat availability as unconfirmed for now.

My take

Duet's announcement comes with all the usual demo-video polish — the usage examples in Anthropic's launch materials are the company's own curated scenarios, not independent testing, so treat the smooth stories with the usual grain of salt.

But the direction is real. The chatbot era — AI sitting quietly until summoned — is ending, and the company that gets the trust model right (meaningful approvals, visibility into what the agent did, limits that don't surprise you) wins. Duet's approval gates and its honesty about misreading urgency are a decent start, but "you can't watch its browser" is a limitation that has to die before general release.

If you run a small business and you're in the US with a Pro or Max plan, the waitlist is worth joining — if only to see where this goes. For everyone else: watch this space. The agent race just got its third serious contender, and the boring, unglamorous work of admin is about to become the most contested battlefield in AI.

Watch: Claude Duet explained in detail

"Claude Duet: Anthropic's AI Agent That Works While You're Away" — watch here: https://www.youtube.com/watch?v=LzgcUsopapQ

Let us know if you have any cool suggestions