Claude Duet: What It Does, Who Can Use It and Its Limits

Claude Duet is Anthropic's new assistant for work that continues between conversations. Announced on October 21, 2026, it can work across connected apps, prepare tasks in the background and bring decisions back to you. The practical question is whether that preparation saves more attention than checking it takes. Anthropic's announcement describes the release as a research preview.

This guide explains the preview as documented on October 23, 2026, then offers a small trial you can use to judge it. It is research-based analysis; Techotick has not tested Duet hands-on.

Who can use Claude Duet?

The preview is limited to invited US Claude Pro and Max subscribers. Team and Enterprise plans are excluded for now. Duet is available through a separate iOS or Android app and through claude.ai on the web. Anthropic's product page sets out those eligibility rules.

For Australian readers, the current US-only preview means a local Claude subscription does not establish Duet eligibility. Neither the product page nor the setup guide gives an Australian launch date. Do not buy a subscription solely on the assumption that Duet access will follow.

To register, use the official product page's waitlist and the email associated with your Claude account. Invitations arrive in waves, with no guaranteed date. Anthropic also flags mismatched Claude and phone-store account emails as a possible access problem. Follow the invitation's installation instructions when it arrives. The setup guide is the source to check if access fails.

What does it actually do?

Duet carries over Claude account memory, skills and connections. Its own cloud computer lets work continue after you close the app. You can hand it a task by text or voice rather than keeping a laptop running beside you. Anthropic says it checks connected accounts every few hours, so an always-available agent is not the same thing as continuous monitoring. Anthropic describes how the background checks work.

The advertised work includes morning briefs, email drafts, invoice reminders, prepared payroll, contracts and dashboards. Treat that list as possible workflows, not evidence that every connector can complete every step in your account. The product examples show the intended scope.

A useful first assignment has a small output you can inspect: a meeting brief with links to its evidence, for example. An open-ended request to manage an entire inbox makes it harder to tell what was missed and how much supervision the result needed.

Included access still uses your Claude allowance

The setup guide confirms there is no separate Duet fee during the preview. Its background work shares your existing plan allowance. Morning checks and larger assignments compete with your other Claude work for that capacity. Anthropic's usage FAQ explicitly counts background checks toward the limits.

Claude's paid plans have five-hour session limits and weekly limits. Consumption varies with task complexity, features, model and effort; usage across Claude surfaces shares the allowance. That makes a fixed number of completed tasks a poor way to estimate the allowance. Anthropic explains the distinction here.

Evaluate Duet against your busiest working day. A morning brief may be a poor trade if it leaves too little capacity for the work you value most. Record whether you hit limits earlier and whether the prepared work was worth that change. The right comparison is useful work delivered within the capacity you need for the rest of your day.

What to check before approving an action

Anthropic says Duet seeks approval before sending messages, making payments, signing documents or deleting material. Securities trading and transfers between accounts are off-limits even with approval. Those are the boundaries described in the launch announcement; they are not a guarantee that the work presented for approval is correct.

Check the recipient, attachment, amount and requested change before accepting an action. A well-written draft can still use the wrong customer's information. For a first trial, give it work where you can compare the proposed output with a short, known source and decline it without disrupting a business process.

For a broader approach to choosing connected accounts, see Techotick's guide to AI-agent access and privacy.

Check these preview limitations before connecting apps

The help guide says Duet uses Opus 5.5 without a model selector. You cannot watch or take over its browser, and its access does not extend to every project or local file. It also cannot make phone calls or send SMS messages to other people. Some Slack capabilities are limited, and Microsoft 365 is not yet available. Read the current limitations before designing a workflow around an integration.

That last point is easy to miss: the product page displays a Microsoft / Office 365 logo. For an access decision, follow the more specific limitation in the setup guide. A logo is not enough evidence that your proposed workflow is supported today.

Periodic checks also make Duet unsuitable as your only alert for a deadline that cannot wait. Keep a direct notification for that event, and judge the agent on the preparation it adds around it.

A practical first trial

Techotick's suggested trial is to choose one recurring job and compare its result with work you already understand. Use the checks below to decide whether to keep or expand the assignment.

Trial task	Check before relying on it
A brief for one meeting	Compare dates, people and open questions against the linked messages. Count important omissions as well as correct facts.
A draft response to one customer	Verify the recipient and commitments. Record how much rewriting is needed before you would send it.
A short list of outstanding invoices	Match each entry to the source record and check for recent payments before considering a reminder.

Keep a simple log: preparation time saved, review time spent, corrections needed and any effect on your normal Claude usage. Net time saved is preparation avoided minus review and repair time. Do not turn one successful task into a percentage productivity claim; repeat the same task type before expanding its scope.

For eligible subscribers, a small, repeatable trial is a sensible way to assess Duet. If your work depends on an unavailable integration or immediate alerts, wait for the specific limitation to change. The useful outcome is reliable preparation that reduces your workload after review.

Editorial disclosure: This article was researched and drafted with AI assistance, then revised and checked against the linked primary sources. The cover is an AI-generated conceptual illustration, not a Claude Duet screenshot. This article contains no affiliate links or paid product placement.