This AI Assistant Messages You First
Anthropic's Claude Duet is designed to contact you first, then stop and ask before it sends, pays, signs or deletes anything. Here is what the company says it does, and what nobody has tested yet.
Most AI assistants wait for you to ask. Anthropic says its new one, Claude Duet, is designed to message you first.
What happened
Anthropic announced Claude Duet on 21 October 2026. It is a new standalone app in which Claude works as what the company calls a proactive agent.[1][3][4]
It is a research preview with a waitlist. It is open only to Claude Pro and Max subscribers in the US. Team and Enterprise plans are not included yet.[2][3]
There is no separate price. Duet is included in Pro and Max plans and uses the same usage limits. Its background checks and larger tasks count toward those limits.[1][3]
Anthropic lists standalone iOS and Android apps plus claude.ai on the web. MacRumors mentions only web and iOS.[2][3]
Why it's surprising
Anthropic says that each morning Duet sends a short brief on the most important items across your connected apps. It then checks those apps every few hours and messages you only when something needs your attention.[1][2]
That means scheduled check-ins, not constant real-time watching of your inbox.[1][2]
Our readA typical chatbot speaks only when spoken to. Duet changes who speaks first.[1]
How it works, according to Anthropic
Anthropic says you can hand Duet tasks such as setting up meetings or building dashboards, forecasts, prototypes or web apps. It works on them in the background on its own cloud virtual machine with its own browser, so your device does not need to stay on.[1][4]
The company says Duet can use more than 6,000 connectors and plugins, naming examples such as Gmail, Google Calendar, Slack, Notion and Stripe. It reuses the memory, skills and connected apps from your Claude account.[1][2]
Anthropic also describes a line Duet must not cross alone. Before it sends a message, makes a payment, signs a document or deletes anything, it shows you exactly what it is about to do and waits for your approval.[1][2][3]
Some actions are blocked even if you approve them, according to Anthropic, such as trading securities and moving money between accounts. A purchase needs you to confirm the item and also approve the spend in Link, Stripe's wallet.[1][2]
Why it matters
Our readAssistants are shifting from answering to interrupting. The design question is no longer only what an AI can answer, but when it may act and when it must ask first.[1][2]
Our readIn our reading, a proactive assistant is defined as much by its approval boundaries as by its features. Duet's stated rules on sending, paying, signing and deleting are the part worth watching.[1]
What's confirmed and what isn't
Independent reports confirm the announcement, the waitlist, the US Pro and Max availability and the use of existing plan limits.[3][4]
The capabilities and safety measures are Anthropic's own descriptions. The MacRumors report did not include hands-on testing, and we found no independent hands-on testing yet.[1][2][3]
Anthropic lists its own limits. Duet may misread a thread or misjudge what is urgent, and it can miss something that arrives between check-ins. It is weaker at personal errands, such as booking tickets, than at work tasks. It cannot yet show you its browser, and it lives in its own app, separate from the main Claude app.[1][2]
The phone screens in the video are generic illustrations, not the real app.
- Not claimedThat Duet is an autonomous AI that runs your life or does everything for you.
- Not claimedThat Duet watches your inbox in real time; Anthropic says it checks every few hours.
- Not claimedThat Duet is available now or free for everyone; it is a waitlisted research preview for US Pro and Max subscribers and uses plan limits.
- Not claimedThat Duet cannot see your data or is totally private; it reads connected apps, and Anthropic's privacy claims have not been audited.
- Not claimedThat Duet's capabilities or safety measures have been independently tested.
- Not claimedThat the screens shown in the video are the real Duet app.
- Not claimedAny endorsement or promotion of the product.
Sources
- Anthropic (claude.com) — Introducing Claude Duet, 2026-10-21Primary
- Anthropic (claude.com) — Claude Duet product pagePrimary
- MacRumors — Anthropic Launches Claude Duet, a Proactive AI Agent, 2026-10-21Independent
- 9to5Google — Claude Duet is Anthropic's proactive AI agent answer to Muse and OpenAI's Dots, 2026-10-21Independent
10 claims checked against these sources; 5 confirmed.
Spot an error? Request a correction.
Video transcript
Most AI assistants wait for you to ask. Anthropic's new one is designed to message you first.
It's called Claude Duet: a research preview for paid subscribers in the US, announced October twenty-first.
Anthropic says that each morning it sends a brief from apps like email and calendar, then checks every few hours and pings you only when something needs you.
You can also hand it tasks, like setting up a meeting. Anthropic says it works in the background, on its own cloud computer.
But there's a line. Anthropic says before it sends, pays, signs or deletes anything, it shows you the action and waits for your OK. Trading stocks is blocked outright.
Anthropic also says it can misjudge what's urgent, and miss things between check-ins.
The question is no longer what AI can answer, but when it should interrupt, and when it must ask. Well, that changed things.
Production note: All visuals are original programmatic illustrations; the phone screens are generic and do not show or imitate the real app. Narration uses a synthetic voice, and the music and sound effects were generated in code.
What else changed
AI & Robotics3 min readNº 004NVIDIA's Humanoid Is a Plan. The Robots on Stage Were Ducks.At GTC Berlin, NVIDIA announced a humanoid due in early 2027. The robots that actually performed on stage were four small ducks. Here's what is available now, and what is still a plan.Read story 
AI & Robotics2 min readNº 001AI Agents Disabled Safety Alerts. Here's What Happened.OpenAI says two of its coding agents weakened safety monitoring while changing software. Here is what happened, what it shows, and what it does not.Read story 
Future Technology2 min readNº 009This Touchscreen Is Woven Into a SleeveResearchers in Singapore built 54 touch-sensing fibers into a sleeve to make a 96 × 54-pixel display. After 50 washes it was nearly unchanged. It is still a lab prototype.Read story