Short answer: Claude Duet is Anthropic's new proactive AI agent: instead of waiting for you to open a chat, it briefs you every morning, takes tasks by message, works on its own cloud computer, and asks before it does anything sensitive. It runs on Opus 5.5, ships with 6,000 connectors and plugins in the Claude Marketplace, and is a US-only research preview on a waitlist for Claude Pro and Max subscribers. The pattern matters more than the product: Meta Muse, OpenAI Dots, and now Claude Duet have all converged on the same agent design - proactive, approval-gated, connector-driven. If you build software or operations for a business, that convergence is your spec sheet.

What is Claude Duet?

Per Anthropic's announcement (Oct 21), Duet is "a new app where Claude works as a proactive agent" - one you hand work to through messages. The differences from the Claude app people already use:

  • It comes to you. A morning brief flags what needs attention in your inbox and calendar. The regular Claude app waits for you to open a chat; Duet reaches out when something is done or needs a decision.
  • It works on its own computer. Tasks run on a dedicated cloud computer - it can build a dashboard, a financial model, or a deck and return the finished file for review.
  • One thread, full memory. It uses your existing Claude account: your memory, skills, and connectors are there from the first message, by text or voice.
  • 6,000 connectors and plugins in the Claude Marketplace at launch.

The approval gate is the product

The design choice to study: Duet asks before sensitive operations like sending a message or paying. Approvals, logins, and safety checks run on your own cloud computer, so credentials stay out of the model's reach. You can see what it has done and is doing, and steer it.

This is the same trust architecture Meta built into Muse (every action needs an explicit OK) and it is becoming the default expectation for any agent that touches real accounts. Chat interfaces answered questions. Agents do errands - and errands need an approval step and an audit trail.

What Anthropic thinks you will use it for

Their own examples are worth reading carefully, because they are not consumer novelties. They are back-office work:

  • Finding unpaid invoices and drafting the reminders.
  • Preparing payroll for your review.
  • Catching scheduling problems before they cost you - a showing booked on the wrong day, a call landing right before a flight.
  • Morning briefs, drafted replies, briefs/decks/dashboards on request.

That is a list of tasks small and mid-size businesses currently hand to office staff, VAs, or agencies. Anthropic is packaging them into its Pro and Max subscription tiers. Team and Enterprise plans are explicitly not included yet - this is aimed at founders and operators first.

Availability and limits

Duet is a research preview, US only, via waitlist, for Claude Pro and Max subscribers. Usage shares your plan's existing limits, and its background checks count toward them - heavy always-on use will eat the same allowance as your conversations. No Team or Enterprise version, no international date.

The three-way agent race, one week in

  • Meta Muse (September): 10.2 million US downloads in its first 40 days - one of only 12 apps ever to hit that mark, per Sensor Tower. We broke down the launch numbers yesterday.
  • OpenAI Dots: OpenAI's consumer agent entry.
  • Claude Duet (Oct 21): Anthropic's answer, leading with judgment and approval UX rather than distribution.

Meta wins on distribution (3.64 billion daily active people across its apps, September 2026 average, per Meta's Q3 2026 results). Anthropic cannot match that, so it is competing on trust design and work output quality. For buyers, the vendor race is less important than the convergence: proactive + approval-gated + connected to your real apps is now the reference design.

What to do with this if you run a delivery, logistics, or service business

  • Steal the pattern, skip the waitlist. Approval-gated agents with an audit trail can be built on your own accounts today - that is exactly how we build dispatch, support, and back-office agents. You do not need Duet access to get Duet's architecture.
  • Watch the jobs list. If Anthropic's invoice-chasing agent works, your customers will soon expect the same from you: proactive updates, not status pages.
  • Expect the expectation. Within a year, "the agent checks in with me before acting" will be table stakes in ops software. Building that UX now is cheaper than retrofitting it.

If you are mapping which of your back-office or dispatch jobs fit this pattern, we do that as a fixed-scope engagement. Book a 30-minute call: https://calendly.com/matrytech-info/30min

Straight answers

Is Claude Duet free?
No. It needs a Claude Pro or Max subscription, is US-only, and is waitlist-only during the research preview. Its background work shares your plan's usage limits.

How is Duet different from the Claude app?
Same account, memory, and connectors. The Claude app waits for you; Duet briefs you each morning, takes tasks by message, and checks in when work finishes or needs approval.

Duet vs Meta Muse?
Muse is winning on Meta's distribution (10.2M US downloads in 40 days). Duet competes on judgment and approval UX. Both use the same trust pattern: the agent proposes, you approve.

Can a business outside the US use it?
Not yet. No international availability or Team/Enterprise plan has been announced.

Sources: Anthropic - Introducing Claude Duet (claude.com/resources/articles/claude-duet, Oct 21, 2026); The Verge, "Anthropic takes on Muse with Claude Duet" (Oct 21, 2026); Sensor Tower data via TechCrunch (Oct 20, 2026) for the Muse numbers. Capability claims are the vendors' own, attributed as such.