Claude Duet scans email and calendar, drafts replies, and asks before acting. Here's how Anthropic's new agent app works and who can try it.
What is Claude Duet? #
Claude Duet is Anthropic’s new agent app that acts less like a chatbot and more like an assistant working in the background on your behalf. Instead of waiting for a prompt, Duet checks your connected apps on its own, scans email and calendar every morning, flags what needs attention, and drafts replies for you to review. It’s currently in research preview, limited to the US, and only available on Pro, Max, and Ultra paid plans through a waitlist.
TL;DR #
- Claude Duet runs proactively , scanning your email and calendar each morning without you needing to prompt it first, then surfacing what matters and drafting responses.
- It follows up automatically on conversation threads that have gone quiet during the day, acting more like a persistent assistant than a one-off chat session.
- Sensitive actions require your approval , meaning Duet asks before sending a message, making a payment, signing a document, or deleting anything.
- Access is currently restricted , limited to US users on paid Claude plans (Pro, Max, Ultra) and gated behind a research preview waitlist, with no free-tier access yet.
- It positions Anthropic against similar proactive agents , sitting in the same category as Meta’s Muse and OpenAI’s dots, which take a comparable approach to AI that acts on your behalf across connected apps.
- Duet is separate from Claude Code and chat , representing a third mode of using Claude that’s built specifically around ongoing, autonomous task management rather than single conversations.
made for the enterprise.
How does Claude Duet actually work? #
Duet connects to the apps you give it access to, most notably email and calendar, and then operates on a recurring cycle rather than a single request-response pattern. Each morning it reviews what’s come in, identifies items that need your attention, and starts drafting replies so you’re not starting from a blank page. Throughout the day, it keeps tabs on open threads and nudges things forward when conversations stall, which is the kind of task that normally falls through the cracks in a busy inbox.
The interaction model is conversational. You message Duet the way you’d message a colleague, by typing or talking, and it translates that into actions inside the tools it’s connected to. That’s a different posture from Claude’s chat tab, where you’re actively driving every exchange, or Claude Code, where you’re directing work inside a project folder. Duet is designed to initiate.
What guardrails does Duet have? #
Anthropic built in a confirmation step for anything consequential. Before Duet sends a message, makes a payment, signs a document, or deletes something, it checks with you first. That approval gate is the core safety mechanism for a tool that otherwise behaves autonomously: it reads your communications and acts on your calendar without waiting for instructions, so the irreversible or sensitive actions are the ones pulled out for explicit sign-off.
This matters because the entire value proposition of a proactive agent depends on trust. An assistant that drafts a reply for review is low risk. An assistant that can quietly send money or delete files without checking in is a very different product. By drawing the line at message-sending, payments, document signing, and deletion, Anthropic is keeping the proactive behavior (reading, flagging, drafting) separate from the irreversible behavior (acting on the world).
How does Claude Duet compare to Meta’s Muse and OpenAI’s dots? #
Duet sits in the same emerging category as Meta’s Muse and OpenAI’s dots: AI tools designed to proactively manage tasks across connected apps rather than simply respond to prompts. All three represent a shift in how the major AI labs think about assistants, moving from “ask a question, get an answer” toward “give it access, let it work continuously.”
The specifics of how Muse and dots implement proactive behavior differ by company, but the underlying bet is the same across Anthropic, Meta, and OpenAI: users don’t want to manually feed an AI every task. They want it watching the parts of their digital life that generate recurring, low-stakes decisions (inbox triage, scheduling, follow-ups) and handling the busywork automatically. Duet is Anthropic’s specific answer to that bet, built on top of its existing Claude models rather than as a standalone product line.
Who can access Claude Duet right now? #
Access is deliberately narrow at this stage. Duet is in research preview, which means Anthropic is testing it with a limited group before a broader rollout. Three restrictions currently apply:
- It’s available only to users in the US.
- It requires a paid Claude subscription: Pro, Max, or Ultra. The free plan doesn’t include it.
- Even eligible users need to join a waitlist rather than getting instant access.
regulatory compliance and security.
- Name
- Maria Alvarez
- Date of birth
- SSN
- Diagnosis
- Balance due
- $412.80
This is a common pattern for agent features that touch sensitive data like email and calendar. Anthropic has rolled out several capabilities this way recently, starting in beta or research preview on paid tiers before expanding. Docs, slides, and design tools followed a similar path, launching in limited form before reaching every plan including free. There’s no confirmed timeline for when Duet will expand beyond the research preview, but the pattern suggests broader availability is the eventual goal rather than a permanent restriction.
Is Claude Duet worth signing up for? #
For anyone already paying for Claude Pro, Max, or Ultra and dealing with a high volume of email and scheduling, Duet addresses a real pain point: the daily grind of triaging messages and keeping threads moving. The proactive, “check every morning” model is a genuine departure from chat-based AI, where the burden is always on the user to remember to ask. The tradeoff is access friction. A US-only, waitlist-gated research preview means most people curious about it can’t try it yet, and even those who can need to clear the paid-plan bar first. If you’re not already on Pro or above, Duet alone probably isn’t a strong enough reason to upgrade until it leaves research preview and the feature set stabilizes. If you’re already paying and your inbox is the bottleneck, joining the waitlist costs nothing and the downside is limited since Duet asks before taking any action that can’t be undone.
Frequently Asked Questions #
What does Claude Duet do differently from regular Claude chat?
Regular Claude chat responds to what you type. Duet works proactively in the background, checking your email and calendar on its own schedule, flagging items that need attention, drafting replies, and following up on stalled conversations without you having to initiate each step.
Does Claude Duet require a paid subscription?
Yes. Duet is currently available only to users on Pro, Max, or Ultra paid plans. It’s not available on the free tier during the research preview.
Can Claude Duet send emails or make payments without asking first?
No. Anthropic built Duet to ask for approval before sending a message, making a payment, signing a document, or deleting anything. Proactive actions like drafting and flagging happen automatically, but anything irreversible requires your confirmation.
How is Claude Duet different from Claude Code?
Claude Code works inside a folder on your computer, reading and editing files and running commands for coding or file-based work. Duet operates across connected apps like email and calendar, focused on ongoing task management rather than project files.
Is Claude Duet available outside the US?
Not currently. The research preview is limited to US users, and there’s no announced timeline for international availability.