AIREITERClaude Duet is not a rumor, but it is not a general Claude release either. Anthropic opened a research preview on October 21, 2026, with access limited to users in the United States on Pro or Max plans who join a waitlist. The useful question is therefore not “what can Claude Duet do in a demo?” but whether its proactive behavior is worth giving up part of your existing Claude usage budget.
Is Claude Duet available now?
Claude Duet is available as a research preview, not as an open public launch. Anthropic says the preview is offered through a standalone iOS and Android app and through claude.ai; initial access is for US Pro and Max subscribers through a waitlist. The official announcement does not list a separate Claude Duet price. During the preview, it uses the same limits as the user’s existing Claude plan. (Anthropic’s Claude Duet announcement)
| Question | Current answer |
|---|---|
| Is Claude Duet real? | Yes. Anthropic published the product announcement and opened a research preview. |
| Is it generally available? | No. Access is waitlist-based and initially limited to the United States. |
| Who is eligible? | Claude Pro and Max users are the stated initial audience. |
| Where does it run? | A standalone iOS or Android app and claude.ai. |
| Is there a separate Duet fee? | Anthropic does not state one in the announcement. |
| Does it have a separate quota? | No. Preview usage shares the user’s existing Claude limits. |
That distinction matters for anyone searching for a Claude Duet download or review. Joining the list is a request for preview access, not a guarantee that the app will be immediately usable in every country or account.
The workflow Claude Duet changes
Claude Duet is designed to start useful work before the user writes a new prompt. It can prepare a short morning brief, inspect connected applications every few hours, and notify the user when it decides that something deserves attention. Anthropic describes the goal as helping professionals and small-business owners decide where to spend attention across many active systems.
From prompts to attention management
The stated foundation is Claude Opus 5.5. The product carries over the user’s Claude memory, skills, and connected applications, then changes the interaction pattern: Claude Duet can contact the user first instead of waiting for a question. Users can interact by typing or voice, and Anthropic says the same ongoing thread can contain a quick request or a project that runs for days. (Anthropic’s product description)
This is a meaningful change for operators who already live in email, chat, calendars, finance tools, and project systems. It is less compelling if you want a predictable assistant that only acts when explicitly called. Proactivity is the feature, but judgment is also the main failure mode: Anthropic says Duet may misread a thread or misjudge urgency while the preview is being tuned.
Background work and approvals
Claude Duet runs in a virtual machine with its own browser. The user’s phone or laptop does not have to remain powered on or connected while the agent works. After a short task description, Duet can choose applications and steps, then notify the user when the work is complete or needs review.
The approval boundary is more important than the autonomy claim. Duet can read information, draft content, and prepare work independently, but it must show the user what it plans to do before it sends a message, makes a payment, signs a document, or deletes information. Anthropic also says Duet cannot trade securities or move money between accounts, even with approval. Applications can be disconnected from the account. (Claude Duet safety and workflow details)
That makes Claude Duet closer to a supervised operations agent than an unattended employee. It can remove preparation work, but consequential actions still require a human checkpoint.
What Claude Duet can connect to
Anthropic says Claude Duet can use more than 6,000 connectors and plugins in the Claude Marketplace. The announcement names examples across communications, accounting, payroll, hiring, legal work, marketing, customer relationship management, and commerce. The table below separates the described workflow from the access status that Anthropic actually states.
| Connected tools | Described workflow | What to watch |
|---|---|---|
| Gmail and Slack | Find missed messages and help draft replies | Drafting is not the same as sending; messages require approval. |
| QuickBooks and Stripe | Check unpaid invoices and draft reminders | Background checks consume plan limits. |
| Gusto | Prepare payroll with hours, bonuses, and reimbursements | Payroll remains a consequential workflow. |
| Draft a job posting and publish after approval | Publishing is gated. | |
| Docusign | Prepare agreements for review and signature | Signing requires approval. |
| HubSpot | Draft follow-ups using contact and deal history | Data quality still controls the result. |
| Canva | Turn a product launch into a social campaign | The announcement does not provide a success rate. |
| Shopify | Set up, design, and manage a store | A richer visual editor is described as planned. |
Payments use an additional control layer. Anthropic says Claude Duet can purchase business items through Link, Stripe’s digital wallet, with the user approving the spending and Link paying through a virtual card. Shop Pay is also described as supported for Shopify checkout. These are official product claims, not independent tests of payment reliability.
The announcement also describes 1Password sign-in as a future capability. It should not be treated as part of the current preview feature set until Anthropic marks it available.
Safety is a boundary, not a benchmark
Claude Duet’s safety story is specific about controls but not about measured outcomes. Anthropic says Duet uses its framework for trustworthy agents, credential handling based on Claude Managed Agents, an auto mode classifier, and defenses against hidden instructions and prompt injection. Anthropic also says the product received internal red-team testing, external penetration testing, and coverage under its public bug bounty program. (Anthropic’s safety description)
The credential model is designed so Claude Duet does not directly see the user’s passwords or sign-in tokens. Anthropic also describes a dedicated classifier and safety system prompt for sensitive information. Requests involving credit-card numbers or government identification are intended to be rejected; health and personal-finance information may be sent only when the user explicitly instructs Duet to do so.
Those controls reduce the blast radius of an incorrect action, but they do not prove that the agent will make good decisions. The announcement provides no public rate for false approvals, prompt-injection resistance, task completion, or data-handling errors. A buyer should therefore evaluate the workflow by asking what Duet can prepare safely, not by assuming that a red-team claim equals production reliability.
Current limitations and evidence gaps
Claude Duet is an early research preview, and the limitations are material to a purchase decision. Anthropic says the agent can misread a conversation, misjudge urgency, and perform less well on personal errands such as booking tickets. The product is currently separate from the main Claude app, cannot yet show the user its browser, and is optimized for work-related tasks.
The usage model is another constraint. Duet checks connected accounts every few hours rather than continuously because more frequent monitoring would consume plan limits quickly. Larger assignments such as a financial model or presentation deck require more model turns and background work. A user who enables many integrations may therefore feel the quota cost before seeing the time savings.
The public evidence also has a clear ceiling. Anthropic supplies examples from its own team and early small-business users, but the announcement does not publish a sample size, independent study, error rate, task-success rate, or comparison against another agent. One quoted user, Amin Ebrahimi of Shahs of Kabob, described the cross-application experience this way:
“It brings it all together.” — Amin Ebrahimi, owner of Shahs of Kabob, quoted in Anthropic’s Claude Duet announcement.
That is useful context about the intended value, not a controlled performance result. The right interpretation is that Claude Duet has a credible product direction and publicly stated safeguards, while its reliability remains unproven outside the preview evidence Anthropic chose to publish.
Who should join the preview?
Claude Duet is worth joining if your work is fragmented across several business applications and the main cost is discovering what needs attention. A small-business operator who repeatedly checks inboxes, support messages, invoices, calendars, and CRM records is a stronger fit than someone who only needs occasional drafting.
It is also a reasonable fit for supervised multi-step projects. Duet can prepare dashboards, forecasts, prototypes, job posts, agreements, and marketing assets in the background, then stop before consequential actions. The approval gates make it more suitable for review-heavy work than for fully autonomous execution.
The recommendation is different for high-risk finance or personal administration. Duet cannot trade securities or move money between accounts, and its preview behavior can misjudge urgency. Do not join expecting an autonomous bookkeeper, personal travel agent, or replacement for approval controls.
| Your situation | Decision |
|---|---|
| You manage several business apps and lose time triaging them | Join the waitlist if you are in the US on Pro or Max. |
| You want background drafting with a human approval step | Good preview use case. |
| You need reliable autonomous payments or account transfers | Poor fit; those actions are restricted or gated. |
| You are outside the US or lack Pro/Max | You cannot rely on current preview access. |
| You want independent benchmark data before trying it | Wait; Anthropic has not published those metrics. |
Claude Duet FAQ
Is Claude Duet free?
Anthropic does not list a separate Claude Duet price in the launch announcement. During the research preview, access is tied to Claude Pro or Max and uses the plan’s existing usage limits.
Is Claude Duet available outside the United States?
The initial preview is for users in the United States. The announcement does not promise a global launch date.
Can Claude Duet send emails automatically?
It can prepare messages, but it must show the user what it plans to do and receive approval before sending a message.
Can Claude Duet move money or trade stocks?
No. Anthropic says trading securities and moving money between accounts are prohibited actions.
Does Claude Duet run while my computer is off?
Yes, the described architecture uses a virtual machine and its own browser, so the user’s laptop or phone does not need to stay powered on or connected.
Does Claude Duet use my normal Claude quota?
Yes. Background checks and larger tasks consume the existing plan limits, and Anthropic says the product checks connected accounts every few hours rather than continuously.
Is 1Password sign-in available now?
No current availability is stated. Anthropic describes sign-in using saved 1Password logins as a future capability.
Claude Duet is a sensible preview for US Pro and Max users who want proactive triage across work tools and can tolerate shared quota consumption. The unresolved trade-off is straightforward: the more often Duet watches and the more applications it can use, the more valuable its attention management may become, but the more usage capacity and permission oversight the workflow requires. That is a reason to join the waitlist selectively, not a reason to treat the research preview as a finished autonomous assistant.