Claude Duet: Anthropic’s Proactive AI Agent Explained

Claude Duet is a research preview of a proactive AI agent from Anthropic, announced on October 21, 2026. Instead of waiting for you to open a chat window, Duet works like a chief of staff you text: it messages you a morning brief, keeps watch over the apps you connect, carries out tasks on its own computer in the cloud, and pings you only when something needs a decision. It is powered by Anthropic’s Opus 5.5 model and is rolling out by waitlist to Claude Pro and Max subscribers in the United States.

The launch puts Anthropic in direct competition with the other proactive-agent products announced this fall — OpenAI’s Dots and Meta’s Muse, as widely reported by the tech press. But Claude Duet’s differentiator is what it inherits: your existing Claude memory, skills, and more than 6,000 connectors and plugins are available from your very first message, all in a single thread.

This guide breaks down how Duet works, what it can and cannot do yet, how its pricing and usage limits behave, and how it differs from both the regular Claude app and its rivals.

Claude Duet at a Glance

What Is Claude Duet?

Claude Duet flips the normal direction of an AI assistant. The Claude app waits for you to open a chat and work on something alongside the model. Duet comes to you: every morning it sends a brief summarizing what needs your attention across your connected apps, drafts the next steps, and then goes quiet until something actually needs you.

You talk to it the way you would message a person — typed or spoken, no prompts or setup. Hand it a task (“set up the board meeting”, “build me a dashboard I can check every morning”) and it works on its own computer in the cloud, so it keeps making progress after you close the app. When the work is finished — or hits a decision only you can make — it messages you in the same single thread where all your history lives.

Because it runs on your Claude account, everything you have already set up carries over: your memory, your skills, and your connectors. A Tasks tab shows what Duet is working on and flags anything waiting on your input — and only you can mark a task as completed.

What Claude Duet Can Actually Do

Anthropic’s product page and support docs sketch the preview’s scope with unusually concrete examples. On the daily-operations side, Claude Duet can:

• Send a morning brief and flag what needs you in your inbox and calendar, with drafted replies for your approval
• Find unpaid invoices and draft the reminder messages
• Prepare payroll for your review and get contracts ready to send through DocuSign
• Assemble briefs, decks, and dashboards — one demo shows it refreshing a business-health dashboard at 5:30 every morning before the brief
• Pay fees and make purchases through Link and Shop Pay, asking before each purchase and using a one-time card that only works for that purchase

The demos Anthropic published are pitched squarely at small-business owners and busy professionals — tracking a restaurant opening checklist, rebuilding a Series A deck, chasing 27 wrong delivery orders — rather than at developers. That positioning matches how the press covered the launch.

Safety, Approvals, and Privacy

The name “Duet” points at the design philosophy: the agent takes the lead on execution, but consequential moves are a duet. Claude Duet is built to ask before it sends a message, makes a payment, signs a document, or deletes anything. If you cannot approve right away, it finishes the rest of the task and brings the approval back to you later. When it quotes a number or summary, it tells you which email, file, or spreadsheet it came from.

Approvals, logins, and safety checks run on your own cloud computer, per Anthropic, so sensitive details stay private — and when a website asks for a password, you type it into a separate form on your device rather than into the chat. You can see everything Duet has done and is doing, and steer it at any time.

Anthropic is also candid about prompt injection: Duet reads email and web pages written by other people, which can contain instructions meant to mislead it. Treating what it reads as information rather than instructions is a design goal — but, as the docs state, no defense is perfect, which is exactly why consequential actions require your approval.

Pricing and Usage Limits

During the research preview, Claude Duet is included with Claude Pro and Max plans at no extra cost. The catch is usage: Duet shares your plan’s existing limits — the five-hour session window and the weekly cap — at the same rate as regular chats, and background work like the morning brief and check-ins also counts. Heavy users may hit their limits noticeably sooner than with chat alone. Duet checks your connected accounts at set intervals specifically to keep that background work from burning through your quota.

One welcome detail: when you are admitted to the preview, Anthropic grants a single full limit reset (available until November 30, 2026) so your first day with Duet is not throttled by your earlier usage.

Claude Duet vs the Claude App vs Claude Code

Anthropic now ships three very different ways to work with Claude. Knowing which one you are reaching for matters more than ever:

Against its true rivals — OpenAI’s Dots and Meta’s Muse — the early differentiator is ecosystem depth: Duet starts with the connectors, plugins, and memory Claude users already configured, rather than asking you to wire up an agent from scratch. All three products are days or weeks old, so treat any “which agent wins” take as premature.

How to Get Claude Duet

What Is Not Ready Yet

Anthropic labels Claude Duet a research preview, and the support docs list the failure modes plainly: it may misread a thread, miscalibrate what is urgent, or miss something that arrives between check-ins. Review drafts and approval requests before confirming them — the agent suggests, you decide.

Also missing at launch: availability outside the US, Team and Enterprise plan support, and any pricing commitment for after the preview. If any of those are dealbreakers, watch this space rather than joining the waitlist today — and in the meantime, the classic Claude app plus connectors already covers a surprising amount of the same ground on demand.

Claude Duet FAQ

The Bottom Line

Claude Duet is the most aggressive rethinking of the assistant relationship Anthropic has shipped: an always-on agent with real reach into 6,000+ apps, gated by approval prompts on exactly the actions that matter. The research preview is US-only and usage-limited, so it is a preview to watch rather than a day-one migration — but for Pro and Max subscribers who live in their inbox, the waitlist is worth joining.