Anthropic's Claude Duet: what its approval rules teach small businesses about AI agents

On October 21, 2026, Anthropic opened a waitlisted research preview of Claude Duet, a proactive AI agent it says is built for professionals and small business owners, according to Anthropic’s announcement and The Verge. For now it is limited to Pro and Max subscribers in the United States. The most useful part for any business owner is not the app itself but the rule Anthropic set for when the agent must stop and ask: before it sends a message, makes a payment, signs a document, or deletes anything.

This article covers what Claude Duet does, who can use it today, and the approval rules you can copy for any AI agent in your business.

What is Claude Duet?

Anthropic describes Claude Duet as an agent that “comes to you with what needs your attention, completes tasks you ask of it, keeps working even when you step away, and checks with you before it acts on your behalf.” CNBC quotes the same line from Anthropic’s blog post.

In practice, that means three things, according to Anthropic and The Verge:

• A morning brief. Each morning it sends a short summary of the most important items across your connected apps, then messages you during the day only when something needs your attention.
• Tasks of different sizes. Anthropic’s examples range from scheduling meetings to building dashboards and creating financial or sales forecasts, which CNBC also lists.
• Work that continues without you. It runs on its own virtual machine with its own browser, so your laptop or phone does not have to stay on.

CNBC and The Verge both framed the launch as a challenge to Meta’s Muse and OpenAI’s dots, two agents we covered earlier this month: Muse for small business and dots and always-on agents.

Who can use Claude Duet right now?

Access is narrow during the preview:

• It is a research preview with a waitlist, open to Pro and Max subscribers in the United States, according to Anthropic and The Verge.
• It is not available on Team or Enterprise plans yet, according to Anthropic’s help center and product page.
• It runs as a standalone app on iOS and Android and on the web, according to Anthropic and CNBC.

If your business is in Canada, or your team works on a shared business plan, this is one to watch rather than one to try this week.

What does Claude Duet do on its own, and what needs your approval?

Anthropic says the agent reads, drafts, and prepares work on its own. But before it sends a message, makes a payment, signs a document, or deletes anything, “it shows you exactly what it’s about to do and waits for your approval.” The Verge describes the same design: before any big step, like sending a payment or deleting something, the agent messages you for approval.

A few actions are off limits even with your approval, including trading securities and moving money between accounts, according to Anthropic’s announcement and its help center.

Anthropic also says a new classifier checks the messages the agent drafts and is designed to reject requests to share the most sensitive information, such as credit card numbers and government identification. That is in Anthropic’s announcement, and The Verge reports it too.

What does Anthropic say Claude Duet still gets wrong?

Anthropic is upfront that this is an early version. It says it “might misread a thread or miscalibrate what’s urgent,” a line quoted by both The Verge and CNBC.

It also checks your connected accounts at intervals rather than nonstop. Anthropic’s announcement says it checks every few hours, and the help center notes that something urgent can arrive between checks.

What does this mean for your business?

Even if you never use Claude Duet, its rules make a practical checklist for any AI agent you put to work:

1. Name the actions that always need a person. Sending to customers, paying, signing, and deleting are a good starting list. A dental clinic’s agent can draft appointment reminders while staff approve anything that discusses treatment. A property manager’s agent can prepare a lease renewal and wait for approval before it goes out for signature. A landscaping company’s agent can find unpaid invoices and draft reminders that the office manager approves.
2. Decide what is off limits entirely. Some actions should not be available to an agent at all, even with a click of approval. Moving money between bank accounts or changing an employee’s payroll details are common examples.
3. Keep approval requests few and clear. If an agent asks for approval too often, people start tapping yes without reading. A handful of clear requests a day is safer than dozens of vague ones.
4. Keep urgent alerts on a direct path. An agent that checks in every few hours should not be the only way you hear about a burst pipe call to a plumbing company or a same-day cancellation at a clinic.
5. Keep payment and ID numbers out of what the agent can send. Set this as a rule, then test it before go-live.

If email is where you would start, our guide to giving an AI agent email access safely walks through the setup in more detail.

Frequently asked questions

Is Claude Duet available in Canada?

Not during the research preview. Anthropic’s help center says it is available in the US only, and The Verge reports that you need to be a Pro or Max subscriber in the US to use it.

Does Claude Duet need approval before it sends an email?

Yes, according to Anthropic. The announcement says it waits for your approval before it sends a message, and the help center says it can send emails from your connected accounts with your approval.

Can Claude Duet move money?

Only in a limited way. Anthropic says it can pay for purchases you approve using a one-time or virtual card from a digital wallet, so your real card number is not shared, but moving money between accounts is off limits even with your approval. Both points are in the announcement and the help center.

Should my business wait for tools like this, or set up an AI agent now?

It depends on the workflow. If one task eats hours every week, such as invoice follow-ups at an accounting firm or booking confirmations at a restaurant, an agent can help now, provided the approval rules above are in place. If you mostly want a smarter inbox, waiting for these previews to mature is reasonable.

Where does your business stand?

If you are not sure which of your workflows are ready for an AI agent, or which ones should always keep a person in the loop, the free AI readiness quiz takes about two minutes and gives you a score with recommendations.