---
title: "Claude Duet for small business: what it does and what to connect"
description: "See if your US Pro or Max plan can join the Claude Duet waitlist, which apps it can reach, what it asks before sending or paying, and what to connect first."
url: "https://webleveling.com/blog/claude-duet-for-small-business/"
lang: "en"
published: "2026-10-23"
modified: "2026-10-23"
author: "Cal Hewitt"
tags: ["ai-automation","business-ai","small-business"]
image: "https://webleveling.com/og/blog-claude-duet-for-small-business.jpg"
alternates:
  es: "https://webleveling.com/es/blog/claude-duet-para-pequenas-empresas.md"
---

# Claude Duet for small business: what it does and what to connect

See if your US Pro or Max plan can join the Claude Duet waitlist, which apps it can reach, what it asks before sending or paying, and what to connect first.

![A small brass desk bell beside a folded paper note and a pencil on a light wooden desk.](https://webleveling.com/blog/claude-duet-for-small-business/claude-duet-approval-checkpoint-featured.jpg)

Claude Duet is Anthropic's new proactive AI agent, announced on October 21, 2026 as a research preview. It sends a morning brief, watches the apps you connect, and asks before it acts.

Right now it is a waitlisted preview for Claude Pro and Max subscribers in the United States, on iPhone, Android and claude.ai. It has no separate price. It runs on your plan's usage limits.

If you run a small business, the useful questions are what Duet can reach and what it can do while you are away. Start with one low-risk connection, such as email or calendar, and let it draft before it does anything else.

**Key Takeaways**

**Access**

Claude Duet is a research preview, by waitlist, for US Claude Pro and Max subscribers, with no separate price and no Team or Enterprise access yet.

**Approvals**

It asks before it sends, pays, signs or deletes, and it cannot trade securities or move money between accounts, but it can still misread a thread or miss something between checks.

**First test**

Connect one low-risk source, let it read and draft, compare the result with the original email or calendar entry, then add one permission at a time.

## Who can get Claude Duet now, and what does it cost?

Anthropic's [announcement](https://claude.com/resources/articles/claude-duet) and [product page](https://claude.com/product/claude-duet) describe Claude Duet as available in Pro and Max plans in the US, by waitlist. Team and Enterprise plans are not included yet. Anthropic says it is powered by Opus 5.5.

There is no separate Duet price. The preview shares your Claude plan's usage limits, so background checks count against them, and bigger jobs such as a financial model or a deck use more than a typical conversation. If you are outside the US or on a plan that is not Pro or Max, plan for a wait.

Anthropic's welcome email to preview users says Duet runs on iPhone, Android and claude.ai with nothing to install, and that your Claude memory, skills and connected apps carry over. Anthropic's [getting started article](https://support.claude.com/en/articles/17631495-getting-started-with-claude-duet) covers joining the waitlist and the first steps after access opens.

![The Claude Duet welcome screen with steps to install the app on Android or iPhone and an option to open it on the web.](https://webleveling.com/blog/claude-duet-for-small-business/claude-duet-welcome-screen.png)

The Claude Duet welcome screen after access opens: install on a phone or open the web app.

## What can Claude Duet actually do for a small business?

Duet is built to come to you instead of waiting for a prompt. Anthropic describes a morning brief of what needs your attention, ongoing checks on connected apps, and work that continues after you step away. It can read, draft and prepare work, and it can carry out actions you approve.

Anthropic says it works with more than 6,000 connectors and plugins. The list it names includes Gmail, Slack, Google Calendar, QuickBooks, Stripe, Gusto, LinkedIn, DocuSign, HubSpot, Canva and Shopify, with 1Password support planned.

Anthropic says Duet checks connected accounts every few hours, so something that arrives between checks waits until the next one. Treat it as a preview that prepares work for you, with a person still answering customers and keeping the books.

## What should you connect to Claude Duet first?

Connect one source, and make it a low-risk one. Gmail or Google Calendar suits a first test, because a morning brief, meeting preparation and draft follow-ups all come from information you can check yourself in a minute.

Before you connect anything, take ten minutes in Claude's settings and list every app already linked to your account. Because connected apps carry over to Duet, that list is what Duet can see on day one. Remove the ones you do not need. Anthropic's [connector guidance](https://support.anthropic.com/en/articles/11817150-connect-your-tools-to-unlock-a-smarter-more-capable-ai-companion) explains how connected tools work, and our post on [which accounts to connect to an AI assistant](https://webleveling.com/blog/which-accounts-to-connect-to-an-ai-assistant/) helps you sort the list.

A sensible order for connecting apps

| Order | Source | Why it fits that spot |
| --- | --- | --- |
| First | Gmail or Google Calendar, read and draft only | Easy to compare with the original email or event |
| Second | Documents or storage, read only | Adds background for drafts without changing records |
| Later | CRM, accounting, payments | Touches customers or money, so add after the first results hold up |
| Hold | Signing, payments, deletion, customer-data workflows | Keep out of the first test entirely |

Add one permission at a time after that. If a result looks wrong, you will know which change caused it.

![A closed laptop beside a small stack of unopened envelopes and a pencil on a worn wooden desk.](https://webleveling.com/blog/claude-duet-for-small-business/claude-duet-one-source-first.jpg)

Begin with one source you can check by hand.

## What does Claude Duet ask your approval for?

Anthropic says Duet asks before it sends messages, makes payments, signs documents or deletes anything. It also cannot trade securities or move money between accounts, even if you say yes.

Approval is a checkpoint, not a guarantee. Anthropic lists the limits it expects: Duet may misread a thread, miscalibrate how urgent something is, and miss something that arrives between its checks. Read each approval request as a decision you are making, not a box to click.

Anthropic's guidance on [custom integrations](https://support.anthropic.com/en/articles/11175166-about-custom-integrations-using-remote-mcp) tells users to review tool approval requests, turn off tools they do not need, and use "Allow always" only for tools they trust. The [OWASP guidance on excessive agency](https://owasp.github.io/www-project-top-10-for-large-language-model-applications/2_0_vulns/LLM06_ExcessiveAgency.html) describes the risk of an agent holding more permissions than the job needs, including instructions hidden in email or documents that an agent reads.

## How does Claude Duet compare with OpenAI Dots and Meta Muse?

All three are agents that work across your apps, and all three were announced within about six weeks of each other. Our posts on [Meta Muse for a small business](https://webleveling.com/blog/meta-muse-for-small-business/) and [OpenAI Dots for a small business](https://webleveling.com/blog/openai-dots-agent-small-business/) cover each rival in full.

Claude Duet, OpenAI Dots and Meta Muse, from each company's own pages

|  | Claude Duet | OpenAI Dots | Meta Muse |
| --- | --- | --- | --- |
| Announced | October 21, 2026 | September 29, 2026 | September 8, 2026 |
| Available to | US Claude Pro and Max subscribers, by waitlist | Pro, Business Premium and Enterprise plans in eligible markets | Muse app and WhatsApp |
| Where it runs | iPhone, Android, claude.ai | ChatGPT, Slack, Teams | Muse app and WhatsApp |
| Connects to | More than 6,000 connectors and plugins | More than 4,000 apps through plugins | Connected apps, browsing, forms, email, travel, purchases through Link |
| Approval | Asks before sending, paying, signing or deleting | Custom Rules let you allow, block or require approval for supported actions | Asks before sending an email or making a purchase |

The table shows differences in access and reach, not which agent is better. Pick by the plan you already have. A Claude Pro or Max subscriber in the US can join the Duet waitlist. A business already on a qualifying ChatGPT plan can read [OpenAI's Dots announcement](https://openai.com/index/introducing-dots/) and its [getting started article](https://help.openai.com/en/articles/20001530-getting-started-with-your-dot). Muse is described in [Meta's announcement](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/).

## What are good first tasks for Claude Duet, and what should wait?

Anthropic's welcome email suggests handing off a to-do, connecting email and calendar for a morning brief, and asking for a forecast, deck or dashboard. All of those produce something you can read and correct before it reaches anyone else.

Good first tasks share a trait: you can check the answer against a record you already trust. A morning brief can be compared with your inbox and calendar. A draft follow-up can be read before it goes out. A forecast can be checked against your own numbers.

Keep payments, signing, deletion and customer-data workflows out of the first test. Those are the steps where a misread thread costs money or trust, and Duet asks for your approval on them for a reason. Our [AI use policy guide](https://webleveling.com/blog/ai-use-policy-for-a-small-business/) helps you write down what staff may connect, and our post on [checking an AI answer before you act on it](https://webleveling.com/blog/check-an-ai-answer-before-you-act-on-it/) gives you a habit to pair with Duet's drafts.

![A paper calendar page with a few pencil marks beside a small stack of blank index cards on a wooden table.](https://webleveling.com/blog/claude-duet-for-small-business/claude-duet-check-the-draft.jpg)

Compare every draft with the original before you trust the next one.

## When does outside help with Claude Duet make sense?

You can do the first test yourself. Join the waitlist, connect Gmail, ask for a morning summary and compare it with your inbox. That takes no outside help.

Help earns its place when Duet will touch more than one system or more than your own inbox. That means a permission review across accounting, CRM, email, calendar and storage, written approval rules for what Duet may do alone and what needs a person, protection against instructions hidden in emails or documents, training for staff, and a workflow tested before it reaches customers or money.

That is setup and risk reduction, and it is the work our [AI consulting](https://webleveling.com/services/ai-consulting/) team does. The aim is a setup you can read, review and switch off, not an agent that runs the business unattended.

![A single key on a ring resting beside a small padlock and a blank paper tag on a clean desk.](https://webleveling.com/blog/claude-duet-for-small-business/claude-duet-permissions-you-can-revoke.jpg)

Every permission you grant should be one you can take back.

**Quiz: Are you ready to try Claude Duet?**

1. Who can join the Claude Duet research preview waitlist right now?
   - Any Claude user in any country
   - Claude Pro and Max subscribers in the United States **(correct answer)**
   - Claude Team and Enterprise customers only
   - Anyone with a Gmail account
2. What should the first Claude Duet test look like?
   - Connect accounting and payments so it can pay invoices
   - Connect every app so it has full background
   - One low-risk source, read access and drafts, checked against the original **(correct answer)**
   - Let it send replies to customers without review
3. Which of these does Anthropic say Claude Duet asks about before it acts?
   - Sending messages, making payments, signing documents and deleting anything **(correct answer)**
   - Only deleting files
   - Only trading securities
   - Nothing, it always acts on its own

## Frequently Asked Questions About Claude Duet

**What is Claude Duet?**

Claude Duet is Anthropic's proactive agent. It sends a morning brief, monitors the apps you connect, prepares work while you are away and asks you before it acts. It launched as a research preview on October 21, 2026.

**How much does Claude Duet cost?**

There is no separate Duet price. The preview uses your Claude Pro or Max plan, and background checks and larger tasks draw on that plan's usage limits.

**Can Claude Duet send email or pay invoices without asking?**

No. Anthropic says it asks before sending messages, making payments, signing documents or deleting anything. It cannot trade securities or move money between accounts.

**What should I connect to Claude Duet first?**

One low-risk source, such as Gmail or Google Calendar, with read access and drafts. Compare what it produces with the original, then add one permission at a time.

**Is Claude Duet available outside the United States or on Team plans?**

The research preview is for Claude Pro and Max subscribers in the US, by waitlist. Team and Enterprise plans are not included yet.

**Can Claude Duet make mistakes?**

Yes. Anthropic says it may misread a thread, miscalibrate urgency and miss something that arrives between its checks. Review what it prepares before anything goes to a customer.

## Final Thoughts

Claude Duet is a research preview for US Pro and Max subscribers, by waitlist, with no separate price and usage that shares your plan's limits. It reads, drafts and prepares work across the apps you connect, and it asks before it sends, pays, signs or deletes. Its stated limits are real: it can misread a thread, misjudge urgency and miss what arrives between checks.

Used with care, it can take routine preparation off your desk: the morning brief, the meeting notes, the draft follow-up. The owners who get the most from a tool like this are the ones who start small, check each result against the original and widen access slowly.

If you want help setting that up, [Web Leveling](https://webleveling.com/) can review your connected apps, write the approval rules and test the workflow before it touches customers or money. Our [AI automation](https://webleveling.com/services/ai-automation/) work is built around setup you can audit and switch off. We work with small and medium businesses across the country and overseas. [Tell us which accounts you want Duet to reach](https://webleveling.com/contact/), and we will help you plan the first test.

## Terms

Claude Duet words in this post

Tap a term to see what it means.

**Research preview.** An early release that a company offers to a limited group while it tests and changes the product.

**Proactive agent.** An AI tool that checks your apps and brings you work on its own, instead of waiting for you to ask.

**Connector.** A link that lets an AI tool read from or act in another app, such as Gmail or QuickBooks.

**Usage limits.** The cap on how much you can use Claude in a period, which background checks and larger tasks count against.

**Excessive agency.** The risk that an AI tool holds more permissions than its job needs and can cause harm if it is misled.

**Prompt injection.** Hidden instructions placed in an email or document that try to steer an AI tool that reads it.

- [Goldman Sachs small business AI: who qualifies and how to apply](https://webleveling.com/blog/goldman-sachs-small-business-ai-institute/): Check the 10,000 Small Businesses screen: an owner or co-owner, at least two years operating, $75,000 in revenue in one of the past two fiscal years.
- [Gemini tools for small business: what launched and how to test](https://webleveling.com/blog/google-gemini-tools-for-small-business/): Check whether your account has the Gemini tools for small business, what Google has said about cost, and a one-hour test you can run this week.
- [Usually, You Do Not Need the Newest AI Model](https://webleveling.com/blog/do-you-need-the-newest-ai-model/): Keep your current AI model if it handles your key tasks accurately and at acceptable cost. Before switching, test the newer model on five real prompts.