Back to blog
Artificial intelligence

Claude Duet Explained: Anthropic's Proactive Work Agent

Claude Duet is Anthropic's proactive agent for small businesses: what it does, its limits, who can get it, and how it compares with Meta Muse and OpenAI Dots.

İlker Ulusoy 2026-11-03 7 min read

Claude Duet is Anthropic's new proactive agent, opened as a research preview on 21 October 2026. Instead of waiting for a prompt, Claude Duet watches the work apps you connect, sends you a short morning brief, keeps working on tasks while you are away, and asks before it acts on your behalf. Think of it as a careful assistant who prepares everything and then waits for your signature. This guide covers what it does, its limits, who can get it, and how it compares with Meta Muse and OpenAI Dots.

The short version

Claude Duet is a separate app for iOS, Android and claude.ai, built around Claude Opus 5.5. It is aimed at professionals and small business owners, runs on its own virtual machine, and needs your approval before it sends, pays, signs or deletes anything. At launch it is US-only and waitlisted.

What Claude Duet Does

According to Anthropic's Claude Duet announcement, Duet works in three ways:

  • It comes to you. A morning brief summarises the key items across your connected apps. During the day it checks your accounts every few hours and only messages you when something needs attention. You decide what you want to hear about and how often.
  • It does tasks you ask for. Anthropic lists scheduling meetings, building dashboards, financial or sales forecasts, prototypes and simple web apps.
  • It keeps working when you leave. Duet runs on its own virtual machine with its own browser, so your laptop or phone does not need to stay on. It notifies you when a task is finished or needs a review.

You talk to it in one ongoing thread, by typing or by voice. It carries over the memory, skills and connected apps you already set up in your Claude account.

Model and connectors

Duet is built around Claude Opus 5.5. It can use the more than 6,000 connectors and plugins in the Claude Marketplace. Anthropic names Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva and Shopify. For purchases it uses Link, Stripe's wallet, which pays with a virtual card so your real card number is not shared. Shop Pay is supported too. A Shopify storefront editor and 1Password sign-ins are listed as coming soon.

Safeguards and the Limits Anthropic States

Anthropic is unusually direct about what Duet should not do and where it can go wrong.

Approval gates

  • Before it sends a message, makes a payment, signs a document or deletes anything, it shows you the action and waits.
  • Some actions are blocked even with approval: trading securities and moving money between accounts.
  • Duet never sees your passwords or sign-in tokens directly; credential handling follows the same design as Claude Managed Agents.
  • A dedicated classifier rejects requests to share the most sensitive data, such as credit card numbers and government ID numbers. Health and personal finance details are shared only when you explicitly say so.
  • Anthropic says Duet was red-teamed, penetration-tested by an outside firm, and is covered by its public bug bounty on HackerOne.

Known weaknesses

  • It may misread a thread or get the urgency of something wrong.
  • It is built for work, so personal errands such as booking tickets are weaker.
  • It cannot yet show you its browser while it works.
  • It lives in its own app, separate from the main Claude app.

That last point on urgency matters most for a business. A proactive agent that flags the wrong things trains people to ignore it. Plan a few weeks of tuning what it should and should not report.

Access and Pricing: Where Sources Disagree

Duet has no separate price. It is included in existing Claude plans and uses the same usage limits as the rest of Claude. Background checks count toward those limits, and large tasks such as a financial model take more of them.

Which plans qualify is less clear. Anthropic's own announcement says Pro, Max and Ultra. Early coverage from CNBC and TechCrunch mentions only Pro and Max. We treat Anthropic's page as the primary source, but check the waitlist page before you plan around Ultra. The launch date also differs by a day in some reports, because TechCrunch published on 22 October; Anthropic's post is dated 21 October 2026.

For reference, Anthropic's pricing page listed Pro at $20 per month (or $17 per month billed yearly), Max at $100 and Ultra at $200 when we checked on 3 November 2026. Access is US-only for now, by waitlist.

Claude Duet vs Meta Muse vs OpenAI Dots

All three launched within about six weeks of each other. They share the same core idea: an agent with its own cloud computer that keeps working in the background and asks before sensitive actions. The differences are in focus, model and access.

Claude DuetMeta MuseOpenAI Dots
Announced21 Oct 2026 (research preview)8 Sep 2026; small business version 29 Sep 202629 Sep 2026 (DevDay)
ModelClaude Opus 5.5Muse SparkGPT-6 Astra
Main focusProfessionals and small businessesConsumers, plus small businessesBusiness and personal tasks
Where it runsOwn virtual machine and browserMuse Secure VM with a separate Sentinel agent that approves outbound actionsOwn cloud computer
Apps6,000+ connectors and pluginsMeta apps plus Shopify, Slack, Notion, QuickBooks and others4,000+ apps via ChatGPT plugins
AccessUS, waitlist; Pro/Max (Anthropic also lists Ultra)US first, free tier with paid plans for more useChatGPT Pro, Business Premium, some Enterprise; not in Europe or the UK at launch
ApprovalsBefore sending, paying, signing, deletingBefore sensitive actions such as email or purchasesUser-set rules for when it acts alone and when it must ask

Sources for the table: Meta's Muse announcement, TechCrunch on Muse for small businesses, TechCrunch on the Dots launch, and NBC News on Dots availability. We go deeper in our guides to Meta Muse and OpenAI Dots.

What a Business Should Check Before Using Claude Duet

Duet is a research preview. That label is a signal to test it on real but low-risk work before you depend on it.

  1. 1Whose account is it? The launch post names individual plans. If an employee connects company Gmail, Slack and QuickBooks to a personal subscription, company data flows through an account the business does not control. Decide this first.
  2. 2Connect the minimum. Start with one or two apps and read-only access where the connector allows it. You can disconnect any app later.
  3. 3Write down your approval rules. Duet already asks before sending or paying. Add your own: who may approve a payment, and up to what amount.
  4. 4Watch the usage limits. Background checks eat into the same plan limits as normal chats. Heavy users may hit limits faster than expected.
  5. 5Tune the noise. Track for two weeks how many alerts were useful. Adjust what Duet reports before you roll it out wider.
  6. 6Check the region. At launch Duet is US-only. Teams in the EU or Türkiye cannot use it yet and will also need a data protection review when it arrives.

If you want the bigger picture first, our explainer on proactive AI agents covers how this category works, and our earlier piece on Claude Cowork background agents shows where Anthropic's agent products came from.

How Halmob Helps

Halmob's agent platform consultancy helps businesses choose, set up and roll out Claude Duet, Meta Muse, OpenAI Dots, Hermes Agent, OpenClaw and Pi Durable. We map which apps an agent should see, set approval rules, and run a short pilot before a wider rollout. When an off-the-shelf app is not enough, we build proactive agents that watch your own systems and act on their own: follow up a lead, flag a late invoice, open a ticket.

Read more about our agent platform services or contact us to plan a pilot.

Related Articles

Artificial intelligence

Proactive AI Agents Explained: What They Are, How to Start

Proactive AI agents act before you ask. Learn how they differ from chatbots and automations, what Muse, Dots and Claude Duet do, and how to start safely.

2026-11-03 · 7 min read
Artificial intelligence

Durable AI Agents: Checkpoints, Retries and Resume Explained

Durable AI agents keep their state, retry failed steps and resume after a crash. See how it works and compare Temporal, Cloudflare, Pi Durable and others.

2026-11-03 · 8 min read
Artificial intelligence

Pi Durable Explained: Crash-Safe AI Agents in TypeScript

Pi Durable is Earendil's experimental library for crash-safe AI agents. Learn how checkpoints, tool replay rules and Cloudflare, Temporal and Rivet fit in.

2026-11-03 · 8 min read
Artificial intelligence

AI Software Factory Explained: From Spec to Pull Request

What an AI software factory is, which coding agents cover each stage from spec to pull request, where humans stay in charge, and the failure modes to plan for.

2026-11-03 · 7 min read
Artificial intelligence

AI Marketing Factory Explained: One Brief, Many Languages

How an AI marketing factory turns one brief into on-brand posts, articles and ads in several languages, plus the brand risks and EU AI Act rules to plan for.

2026-11-03 · 7 min read
Artificial intelligence

Meta Muse Explained: Personal AI Agent Guide for Business

Meta Muse is Meta's personal AI agent. Learn how it works, what the $20 and $100 plans include, the Amazon block, and what small businesses should check.

2026-11-03 · 7 min read