Claude Duet: What It Does and What to Delegate First

Claude Duet is worth joining the waitlist for if you already use Claude Pro or Max in the US and want routine work prepared for your review. Built on Anthropic's Claude Opus 5.5, it can bring you a morning brief and keep working while you are away. Start with meeting preparation and draft replies, with a person checking the result before anything consequential happens. Anthropic's product page.

Verified against Anthropic's live product, announcement and support pages on 22 October 2026. This is a documentation-based assessment; we have not had access to Duet.

What Is Claude Duet, and How Does It Work?

Claude Duet turns a message into ongoing work on a cloud computer. It is a proactive agent, meaning it can take steps toward a task and bring something to your attention without waiting for another prompt. Anthropic announced the research preview on 21 October 2026. Duet has its own browser and virtual machine, a computer hosted remotely, so your laptop does not have to stay open. Anthropic's announcement.

Think of it as an assistant who prepares the morning folder before you arrive. The useful handoff is specific: gather the material, identify what needs a decision, and bring back something you can check. A vague request to “run the business” gives you no clear standard for deciding whether the work is right.

Duet uses the memory, skills and connected apps already associated with your Claude account. A connector is the link that lets Claude access another service. That continuity reduces setup, but it also makes an access review worthwhile before you delegate business work. Duet setup guide.

What Should a Small Business Hand It First?

Give Duet work where a mistaken draft is easy to catch and cheap to correct. Anthropic lists morning briefs, inbox and calendar flags, reply drafts, meeting scheduling, briefs, decks and dashboards. It also describes finding unpaid invoices and preparing payroll for review. These are documented capabilities; their reliability still needs evaluation in your workflow. What Duet can do.

The distinction is the consequence of getting the result wrong. A missed detail in an internal meeting brief can be corrected before the call. A mistaken payment or an email promising a customer a refund has already changed the business once it goes out.

For a founder, start with a brief for the next investor or customer meeting. Ask for the unresolved questions and the source behind each material claim. The payoff to look for is less preparation work without more correction work.

For an operator, an invoice reminder draft is a useful later trial. Anthropic names Intuit QuickBooks and Stripe for invoice checks, and Gusto for payroll work. Keep payroll preparation behind a knowledgeable review: hours, reimbursements and employee details need checking against the underlying records before approval. Documented business integrations.

For a builder or CTO, choose an internal brief before a workflow that touches production systems. Treat the trial as a way to establish which handoffs your organization can review well. Access to a personal agent does not itself establish permission to connect company data.

Who Can Join the Claude Duet Waitlist?

The research preview is for US Claude Pro and Max subscribers, with an invitation required. Team and Enterprise are not included yet. Duet is available through standalone iOS and Android apps and on the web at claude.ai. Current availability.

Pro is $20 when billed monthly and Max starts at $100 per month before tax; Duet is included at no extra charge during the preview, with no separate Duet price announced. Plan prices and preview inclusion.

To join, use the product page's waitlist form, enter your existing Claude account email, choose your plan and phone type, and submit. Invitations arrive in waves, with no promised access date. Waitlist instructions.

Join now if you are already eligible, have a recurring administrative task, and can check the output. The case is strongest when preparation is the bottleneck and you remain available to make decisions.

Wait on adoption if you need a guaranteed response time or your workflow depends on Microsoft 365: Anthropic's Duet support page says that connector is not available yet. Do not infer readiness from a logo on a marketing page. Keep your current workflow if you need Team or Enterprise access; a personal waitlist place is not a company rollout. Preview limitations.

What Does Duet Ask Before Doing?

Anthropic says Duet is designed to wait for approval before sending messages, making payments, signing documents or deleting anything. Preparation can happen before that decision. Approval behavior.

Make that boundary explicit in the task itself. A useful first request could be:

Prepare a brief for my next customer meeting from the relevant email thread and calendar entry. Separate confirmed facts from assumptions, link to the source material, and flag anything you could not find. Draft any follow-up for my review. Do not send messages, make payments, sign documents or delete anything.

This suggested instruction defines both the deliverable and what must come back to you.

Review the proposed action, not just the summary above it. For an invoice reminder, open the invoice and verify that payment has not arrived elsewhere. For a meeting invitation, check the attendees and time zone. A clear approval screen cannot compensate for a reviewer approving the wrong thing.

How Background Work Uses Your Claude Limits

Duet draws on the same plan allowance as your other Claude work. Morning briefs and background checks count, and bigger deliverables can require more model turns. Anthropic says it checks connected accounts every few hours because nonstop checking would consume limits faster. How Duet uses the plan.

Paid Claude plans have a session allowance that resets every five hours and a weekly allowance. These are usage limits, not a promised number of finished Duet tasks. Anthropic's usage-limit explanation.

For a Pro subscriber who already reaches the limit during ordinary work, background activity creates a practical tradeoff. Start with a narrow task and observe the remaining allowance before expanding to recurring briefs, decks and dashboards together. A Max subscriber should use the same method; a larger allowance still needs to cover the work they already do.

Do not plan around a made-up conversion such as “one dashboard equals a certain number of messages.” Judge the trial by whether the completed work is useful, how much checking it needs, and whether it crowds out higher-priority Claude work.

What's Overhyped: Proactive Does Not Mean Continuously Watching

The dangerous assumption is that silence means nothing needs attention. Anthropic says Duet can misread a thread, misjudge urgency, or miss an item between check-ins. Stated limitations.

Imagine a client changes tomorrow's meeting location after a check. Even a perfectly accurate earlier brief cannot contain a message that arrived later. Keep calendar alerts and direct customer escalation paths in place while you evaluate Duet. Give someone responsibility for checking deadline-sensitive work independently.

Approval and detection solve different problems. Approval gives you a chance to stop the wrong outgoing action. It does not reveal an incoming item that the agent never surfaced. The trial therefore needs two review questions: was the prepared work correct, and what important work did it miss?

You also cannot currently watch or take over Duet's browser, according to the support guide. That makes tasks requiring live supervision a poor first handoff. Browser limitation.

Where Duet Fits Beside Claude Cowork and Claude Code

Claude Cowork, Anthropic's task workspace for files and connected tools, already handles assigned and scheduled work and can continue after your laptop closes. Use it when the deliverable is known, such as a recurring report. See the Cowork product page and our Claude Cowork guide. Duet's distinctive emphasis is bringing you priorities and follow-ups through an ongoing assistant relationship.

Claude Code is Anthropic's coding agent: it writes code, runs tests and opens pull requests. Keep a repository change in that development workflow. Claude Code's capabilities and our Claude Cowork vs Claude Code comparison explain the division. Choose by the work and the review process, rather than assuming every task belongs in the newest app.

What About OpenAI Dots and Meta Muse?

OpenAI Dots and Meta Muse are comparable at the level their makers document: agents that keep advancing work on their own computers. OpenAI describes Dots using a cloud computer, browser and connected apps, with rules for what may proceed independently. Meta describes Muse continuing after the app closes and checking before actions such as email or purchases. That establishes overlapping capability, not equal reliability, access or approval behavior. For a founder already organized around Claude, the sensible starting point is the smallest useful handoff inside that existing account. If Dots is also under consideration, the separate Is OpenAI Dots Free? guide covers its access and allowance question.

Your First Week Checklist

Start the week you receive access with preparation work, approvals on, and only the connections the task needs. Use the week to decide which work earns a continuing handoff.

For more practical guides to delegating business work with AI, join the newsletter.