Claude Duet reads your connected apps, sends a morning brief, drafts follow-ups from HubSpot history and asks your approval before sensitive sends. US Pro and Max, waitlist.
Most AI help in sales waits for you to open a tab and ask. Duet reverses the direction: it reads the calendar, the inbox and the CRM on a schedule, brings you the few things that moved, then drafts the reply and, for sensitive sends, holds it until you say send. For a rep that is the difference between a tool you remember to use and one that stands between you and a forgotten follow-up. The limits matter as much as the promise. It checks every few hours, not continuously, so a buyer's reply at 9:05 may surface at lunch. It is a US waitlist on individual plans, so your company cannot roll it out for you. And sensitive sends wait for your approval, but Anthropic says a routine note may not, and that balance is still being tuned, so read the approval settings before you connect a work inbox.
"Proactive agent" means the software starts the conversation. A normal chat assistant answers what you type. Duet looks at the apps you connected, decides what needs you, and messages you first. "Research preview" is Anthropic's label for software that is real but still being tuned and offered to a limited group. "Waitlist" means joining does not give access the same day.
"Connected apps" are the services you have linked to your Claude account, such as Gmail, your calendar, Slack or HubSpot. Duet reuses those links, so if you already set them up in Claude you do not set them up again. "A virtual machine with its own browser" means Duet does its work on Anthropic's computers, so it keeps going after you close your laptop.
"Waits for your approval" means a sensitive draft sits in front of you with a send button that only you press. Anthropic says routine notes may go out without an extra approval and that this balance is still being tuned, so do not assume every message waits for you.
The short version: Claude can now come to you each morning with what needs attention and a drafted reply, if you are a US Pro or Max subscriber who gets through the waitlist.
Anthropic published "Introducing Claude Duet" on October 21, 2026, opening "a research preview of Claude Duet, a new app where Claude works as a proactive agent." In Anthropic's words it "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." It runs "as a standalone app on iOS and Android, and on claude.ai," is built around Claude Opus 5.5, and carries over "your memory, skills, and connected apps from your Claude account."
Announced: October 21, 2026.
Availability: Research preview "available by waitlist today, starting with US Pro and Max subscribers."
Price: "included in Pro and Max plans and uses the same usage limits as the rest of Claude." No separate price stated.
Regions: United States first. Other regions not stated.
What it does each day, in Anthropic's words: "Each morning, it sends you a short brief on the most important items across your connected apps. Throughout the day, it keeps tabs on those apps and messages you only when something needs your attention." It has "a virtual machine with its own browser, so it doesn't require your laptop or phone to be on." Among the examples Anthropic lists: find "the Gmail and Slack messages you've missed and help you reply," and "Draft follow-up emails using the contact and deal history in HubSpot."
What it doesn't do: It does not take sensitive steps without you. The launch post says that 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." Anthropic's October 23 safety post is narrower: approval applies to actions "like sending a potentially sensitive message, paying for something, or deleting anything," a routine note "generally won't need additional approval," and how often it asks "is still changing." Your approval is also not a universal override: the launch post states "A few things are off-limits even with your approval," naming trading securities and moving money between accounts. It does not watch your accounts continuously: it "checks your connected accounts every few hours, because checking nonstop would use up your plan's limits much faster." Anthropic names only Pro and Max plans for this preview, so Team and Enterprise access is not stated, access outside the US waitlist is not stated, and the only CRM named is HubSpot.
SDR - The morning brief is aimed at your inbox problem: which of yesterday's sends got a reply, which meeting today has no prep, which thread went quiet. Join the waitlist on your own Pro plan if your company allows personal AI tools, then start with one connection, your calendar, before you connect mail. Approval covers sensitive sends, not necessarily every message, so decide what it may send before you connect work mail.
AE - The HubSpot example is the one to watch: a follow-up drafted from "the contact and deal history" rather than from what you remember. Before you trust it, compare three drafts against the deal notes you would have written yourself. If it invents a next step you never agreed, that is the limit of the preview, and you will catch it before the buyer does only if Duet holds that draft for approval, so check how it treats follow-ups before you rely on it.
Manager - This is an individual-plan preview, so your reps may arrive with it before you have a policy. Decide now whether reps may connect company mail and CRM to a personal AI subscription, and write it down this week. If the answer is no, say so before the waitlist clears. If yes, ask each rep to show you one morning brief so you learn what it surfaces and what it misses.
RevOps - Anthropic's HubSpot example is a read: draft emails from contact and deal history. The page does not say Duet writes anything back to the CRM, so assume it does not until you see it, and Anthropic names no Salesforce example at all. Check what a HubSpot connection exposes before anyone connects a shared inbox. The classifier Anthropic describes blocks card numbers and government IDs from being shared but lets other sensitive data through on explicit instruction, so your data-handling policy, not the tool, is the control.
Duet's morning brief is the proactive version of a job you can already schedule: a standing prompt that reads the day's calendar and open threads and returns what needs you first. Set it up once in Scheduled Prompts and you will know exactly what to ask Duet for when the waitlist clears.
Two minutes, once a week. What changed in AI, and what to run because of it.