Claude Duet Proactive Agent Research Preview Explained: Anthropic's Agent That Works While You Step Away
Image: Anthropic — the Claude Duet app (official product image)
📌 KEY HIGHLIGHTS
- What it is: a proactive agent, not a chatbot you open — Duet checks your connected apps, surfaces what needs attention, and completes tasks you hand it by text or voice in one ongoing thread.
- Who gets it: research preview from 21 October 2026, by waitlist, starting with Claude Pro and Max subscribers in the United States. Team and Enterprise plans are not included yet. It is a standalone app on iOS and Android, and on claude.ai.
- How it runs: designed around Claude Opus 5.5, on its own cloud virtual machine with a browser — your phone or laptop does not need to stay on. It carries over your Claude memory, skills and connected apps.
- The approval rule: Duet reads, drafts and prepares work on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it is about to do and waits. Trading securities and moving money between accounts are off-limits even with approval.
- The hidden cost: Duet is included in Pro and Max at no extra charge, but its background checks count toward your normal plan usage limits — Anthropic warns you may reach them sooner than with chat alone.
What happened?
Every major AI assistant so far has waited for you to open it and type. On 21 October 2026, Anthropic flipped that direction. Claude Duet is a new, separate app in which Claude works as a proactive agent: in Anthropic's words, you "message it the way you'd message a person", it works in the apps you have connected, lets you know when something needs you, and is designed to ask before it acts on your behalf. Anthropic is pitching it at professionals and small business owners with more on their plates than they can track, and it arrives in the same agent race as OpenAI's Dots and Meta's Muse — assistants that come to you instead of waiting to be asked.
Be clear about the status, because the headlines blur it: this is a research preview, not a launch. Access is by waitlist, invitations go out in waves, and only US Pro and Max subscribers can join for now. Anthropic itself describes this early version as one that can still misread a thread or misjudge what is urgent — treat everything below as what the company says Duet does, not as independently tested performance. No independent benchmark of Duet's work quality exists yet.
Claude duet proactive agent research preview explained — how it works, in a table
| Part of Duet | What Anthropic says it does | The catch to know |
|---|---|---|
| Morning brief | Each morning it looks across your email and calendar, picks what most needs attention, and can start first steps such as drafting replies for review | It decides what is "important" — and can misjudge urgency, by Anthropic's own admission |
| Check-ins through the day | It checks connected accounts every few hours and messages you only when something needs a decision | Not nonstop — Anthropic says constant checking would use up plan limits much faster; background checks still count toward your limits |
| Background tasks | Runs on a cloud virtual machine with its own browser, so work continues with your devices off; bigger jobs (research brief, financial model, deck, dashboard) arrive as files in a Make tab | You cannot yet watch its browser while it works, and only you can mark a task finished |
| Connectors | More than 6,000 connectors and plugins from the Claude Marketplace — Gmail, Slack, QuickBooks and Stripe invoice checks, Gusto payroll prep, Docusign agreements, HubSpot follow-ups, Shopify stores | Microsoft 365 is not supported yet, and it cannot make phone calls or send text messages |
| Approval gate | It prepares everything, then waits for your OK before sending, paying, signing or deleting; purchases run through Link (virtual card, real number never shared) or Shop Pay | Securities trading and moving money between accounts are banned outright; a dedicated classifier also refuses to share card numbers or government ID |
The design is best understood as the enterprise agent pattern moving into a consumer app. The same approval-first logic is why businesses are buying agent platforms at the scale covered in our OpenAI GPT-6 models on Google Cloud Model Garden explained post, and why legal teams are adopting the agentic tools in our Legora $550 million Series E legal AI explainer.
Image: Unsplash (free Unsplash licence) — Illustrative image: a mobile app in use, representing Duet's phone-first assistant experience
Why it matters for USA readers
For US readers, Duet is the first proactive agent from a frontier lab that is explicitly built around permission, at exactly the moment rivals are racing the other way — Meta's Muse and OpenAI's Dots both promise agents that act first. Anthropic's bet is that the assistant people actually trust with their inbox, invoices and calendar is the one that shows its work and waits. The safety design supports that pitch: Duet never directly sees your passwords or sign-in tokens, a classifier checks each message for the strictest tier of sensitive data, and Anthropic says the app was red-teamed and externally penetration-tested, with a public bug bounty on HackerOne. The trade-off is equally American and equally practical: the preview is US-only, Pro/Max-only, and its always-on checking quietly eats the usage limits you already pay for.
Should you care?
✅ Care if you run your work through email, calendar and a stack of business apps — ⏳ everyone else should watch the preview, not chase it. If you are a US Pro or Max subscriber drowning in follow-ups, joining the waitlist costs nothing and the approval gate makes early testing low-risk: Duet can draft, prepare and organise, but nothing leaves your accounts without your tap. Do not move important work into it yet — it is an early preview that can misread threads, has no Microsoft 365 support, and burns your plan limits in the background. Outside the US, or on Free, Team or Enterprise plans, there is nothing to do but wait: Anthropic says Team and Enterprise come later, with no date given. For the research direction behind cheaper, sharper agent building blocks, see our NVIDIA Nemotron 3.5 open models for AI agents explained post.
FAQs
What is Claude Duet and how does it work?
Claude Duet is Anthropic's proactive AI agent, opened as a research preview on 21 October 2026. Instead of waiting for you to start a chat, it sends a morning brief across your connected email and calendar, checks those apps every few hours, takes tasks by text or voice, and keeps working on a cloud computer with its own browser even when your devices are off. It is designed around Claude Opus 5.5 and carries over your existing Claude memory, skills and connectors.
Who can get Claude Duet right now?
Only Claude Pro and Max subscribers in the United States, and only through a waitlist — invitations are sent in waves, so joining does not guarantee immediate access. Use the email address on your Claude account when you join, because access is tied to it. Team and Enterprise plans are not eligible yet, though Anthropic says it plans to add them later.
Does Claude Duet ask before it sends, pays or signs?
Yes — that is its central design rule. Duet reads, drafts and prepares work on its own, but before it sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it is about to do and waits for your approval. Some actions are off-limits even with approval, including trading securities and moving money between accounts, and a dedicated safety classifier refuses requests to share credit card numbers or government identification.
How much does Claude Duet cost?
There is no separate price during the research preview: Duet is included in Claude Pro and Max plans at no additional cost. The real cost is usage — its background briefs, check-ins and tasks count toward the same plan usage limits as regular Claude chat, and Anthropic warns that background work can make you reach those limits sooner than with chat alone.
TAKEAWAY
Claude Duet is the clearest sign yet that AI assistants are becoming agents that come to you — and Anthropic's answer to the trust problem is simple: it prepares everything, then asks before it acts.
Follow USA Tech on TechByCountry for the next hourly update.
Sources: Unite.AI — Anthropic Opens Claude Duet Research Preview to US Pro and Max Users (21 Oct 2026) • 9to5Mac — Anthropic takes on ChatGPT Dots and Meta Muse with Claude Duet (21 Oct 2026) • Digital Trends — Anthropic unveils Claude Duet, its answer to AI agents like Dots (21 Oct 2026). Checked 22 Oct 2026.
Comments
Post a Comment