Claude Duet is Anthropic’s proactive AI agent, opened as a research preview on October 21, 2026. It sends a morning brief, watches connected apps such as Gmail, QuickBooks and HubSpot, works on larger jobs in the background, and asks before it sends, pays, signs or deletes. Access starts by waitlist for US Pro and Max subscribers.

Anthropic says it built the app for professionals and small business owners. This guide covers what it does, who can get in, the limits that matter in daily use, how it stacks up against OpenAI’s Dots, and a setup plan that starts with one workflow.

Claude Duet, explained

Duet flips the usual chat pattern. The regular Claude app waits for you to open a conversation; Duet comes to you first with what needs attention, takes tasks by text or voice message, and keeps working after you close the app. It runs on Claude Opus 5.5 and lives in a standalone app on iOS and Android, plus claude.ai on the web (Anthropic).

Setup is light because Duet inherits the memory, skills and connected apps already in your Claude account. Everything sits in one ongoing thread, so a two-line reply and a week-long project start in the same place (Claude Duet product page). Anthropic names its audience directly: professionals and small business owners “with more on their plates than they can track.”

A working day with a proactive AI agent

The daily rhythm has three parts, according to Anthropic’s help center guide:

  • A morning brief. It scans email and calendar, picks what most needs you, and can start first steps such as drafting replies for review.
  • Quiet monitoring. It checks connected accounts at set intervals and messages you only when something needs a decision. Anything urgent that lands between checks waits for the next one.
  • Larger jobs on request. Describe a task and it picks the apps and steps. Results such as a research brief, financial model, slide deck or dashboard come back as files in the Make tab, with progress tracked in the Tasks tab.

Anthropic shared three early-tester stories. A real estate broker had a property showing flagged because it sat on her calendar for the wrong day. The owner of a South Florida kabob chain with about 110 employees used it for accounting, customer support and marketing, working through invoices store by store. A brand strategist got a client-call brief plus an offer to move the call, which fell right before her flight (Anthropic). These are vendor-selected examples, so treat them as a picture of intended use.

Claude Duet access, pricing and usage limits

You need three things to use Duet today: a Pro or Max plan, a US Claude account, and an invitation from the waitlist. Team and Enterprise plans are excluded for now, and invitations go out in waves tied to the email address you registered (Claude Help Center).

During the preview Duet costs nothing extra, but it draws on the same five-hour and weekly usage limits as your chats. Its morning brief and background check-ins count too, which is why it checks accounts every few hours instead of continuously. Anthropic gives each new preview user one full limit reset, usable until November 30, 2026.

Plan for that consumption. In McKinsey’s 2026 State of AI survey, one in five respondents said AI operating costs, token costs included, had already limited their organization’s AI use (McKinsey). A Pro plan that also covers your daily drafting and research will reach its limits sooner once an agent works in the background all day.

Approval rules and safety controls

Duet reads, drafts and prepares work on its own, then stops for approval before four kinds of action: sending a message, making a payment, signing a document, or deleting anything. It shows exactly what it is about to do and waits. Some actions are blocked even with your approval, including trading securities and moving money between accounts (Anthropic).

Anthropic launched Duet with a dedicated classifier that rejects requests to share the strictest tier of data, such as credit card numbers and government IDs. Health and personal finance details go out only on your explicit instruction. Credentials follow the same design as Claude Managed Agents, so Duet never sees passwords or sign-in tokens, and approved purchases through Link or Shop Pay use a one-time card (Claude Help Center).

The help center is candid about the gap that remains: email and web pages can carry hidden instructions, and “no defense is perfect” against prompt injection (Claude Help Center). That is the practical case for reading every approval request before you tap it. Our AI agent governance guide covers permission tiers and logging for teams that want written rules first.

Claude Duet vs OpenAI Dots

OpenAI announced Dots at DevDay on September 29, three weeks before Duet. Dots run on GPT-6 Astra with their own cloud computer, connect to more than 4,000 apps through plugins, and can be reached in ChatGPT, Slack or Teams. They are available on Pro and Business Premium plans, with Enterprise access once an admin approves, and one Dot is included per plan (9to5Google, quoting OpenAI).

The two products differ in four places that matter to a small business:

  • Reach. Duet is US-only on Pro and Max. Dots already cover Business Premium and, with admin approval, Enterprise workspaces.
  • Visibility. Dots let you watch their cloud computer while they work. Duet cannot yet show you its browser.
  • Where you talk to it. Duet lives in its own app and on claude.ai. Dots can be messaged in Slack and Teams once created.
  • Integrations. Anthropic cites more than 6,000 connectors and plugins in the Claude Marketplace. OpenAI cites more than 4,000 apps. Anthropic's help center lists Microsoft 365 as not yet available in Duet, and Slack works with limits.

Coverage of the Duet launch framed it as Anthropic’s answer to Dots and to Meta’s Muse agent (9to5Google). Neither company has published independent accuracy data, so pick based on which ecosystem already holds your email, files and team chat.

Setting up Claude Duet for one workflow

Start narrow. An agent that watches every inbox and ledger on day one produces a flood of approvals you will learn to skim, and skimming defeats the safeguard. These six steps keep the first month small enough to judge.

  1. Join the waitlist with the right email. Register on the Claude Duet product page using the email on your Claude account, since access is tied to that address.
  2. Pick one recurring workflow. Choose a repeatable job with a clear finish line, such as chasing unpaid invoices or preparing call briefs from your calendar.
  3. Connect only the apps that workflow needs. For invoice follow-up that might be QuickBooks or Stripe plus Gmail. Leave payroll, banking and file storage disconnected until the first workflow proves out.
  4. Tell it what to flag and how often. Duet accepts plain instructions on what you want to hear about. Set that on day one so the morning brief stays short.
  5. Read every approval and its source. Duet cites the email, file or spreadsheet behind each number or summary. Check that source before you approve anything that sends, pays or signs.
  6. Log hits and misses for 30 days. Note what it caught that you would have missed, what it misjudged, and how much plan usage it consumed. Expand to a second workflow only after that review.

Connecting the apps is quick. Deciding which workflow comes first, and what counts as success, takes more thought. Our digital marketing and technology team helps clients map that before any agent gets access.

Proactive AI agents in marketing work

Three of Anthropic’s launch examples are marketing jobs: drafting follow-up emails from HubSpot contact and deal history, turning a product launch into a social campaign in Canva, and drafting a LinkedIn job post for approval (Anthropic). In each case the agent does the drafting, and the task owner's job becomes working through an approval queue.

McKinsey’s 2026 survey points to why that matters. Nearly three-quarters of AI high performers, the roughly 6 percent of respondents who credit AI with at least 5 percent of EBIT, report fundamentally redesigning workflows because of AI. Just one-quarter of everyone else did (McKinsey). Duet gains you little if it simply drafts the same emails into the same backlog. Decide who reviews, how fast, and which tasks graduate to lighter review once the agent proves reliable. Our fractional CMO team runs that redesign alongside the agents themselves.

Claude Duet as a visitor to your website

Duet will also show up on your website. For sites without a connector, it works through its own browser in the cloud and, per Anthropic, “aims to follow the rules sites set for agents” (Claude Help Center). A buyer who asks Duet to compare vendors or pull a price is sending an agent to read your pages on their behalf.

Pages that state prices, service areas and next steps in plain HTML are easier for any agent to complete a task on. Our guide to optimizing your website for AI agents covers the technical checks, and agentic engine optimization covers being the source those agents cite.

Next steps for this quarter

Join the waitlist with your Claude account email, choose one workflow, and give it 30 days with every approval read in full. Expect the product to change during the preview; Anthropic says features can change during the research preview. If you want agents running across your marketing operations with named owners, look at Next Best Action or talk to our team.