Felix Laurent
Anthropic announced Claude Duet on October 21. It is a separate app where Claude works as a proactive AI agent. Claude Duet checks your connected apps every few hours, sends a morning brief and messages you when something needs your attention. It runs on Claude Opus 5.5 and opens as a research preview by waitlist for Pro and Max subscribers in the US. Duet lands three weeks after OpenAI's Dots and two months after SpaceXAI's Grok Bot. It is the third always-on agent from a frontier lab this fall.
What Anthropic announced with Claude Duet
Claude Duet comes as standalone apps for iOS and Android and on claude.ai. Anthropic says it built the product for professionals and small business owners who have more to track than they can handle. The key facts of the preview:
- Model: Claude Opus 5.5. You can't pick another one – Duet decides how much effort a task needs
- Context: memory, skills and connected apps carry over from your Claude account
- Interface: one ongoing thread, typed or by voice, for both quick replies and week-long projects
- Access: Pro and Max in the US, invitations in waves. Team and Enterprise are not supported yet
- Price: no extra charge during the preview, but Duet uses the same five-hour and weekly limits as the rest of Claude. Admitted users get one full limit reset
In short, Duet costs nothing extra, but every background check spends the same plan limits as a chat.
How Claude Duet works
Each morning Duet sends a short brief on the most important items across your connected apps. During the day it checks those accounts at set intervals – every few hours, not continuously. Anthropic says nonstop checking would burn through plan limits much faster. The help page warns that an urgent item may wait until the next check.
Tasks run in the background on a virtual machine with its own browser, so your laptop or phone can be off. Duet notifies you when a task is done or needs review.
It can use the 6,000+ connectors and plugins in Claude Marketplace. Anthropic's examples include missed Gmail and Slack messages, unpaid invoices in QuickBooks and Stripe, payroll in Gusto and agreements in DocuSign. Others are follow-ups from HubSpot, campaigns in Canva and store setup in Shopify. Purchases go through Stripe's Link with a virtual card, so the real card number isn't shared, or through Shop Pay.
Unconfirmed: Claude Duet agents with their own characters
An 8-second clip posted on X by @vishalsingh2972 shows screens that look like Duet. They don't match anything Anthropic has published: several agents in one app, each with its own pixel character. The agents screen lists 10 agents with 2 waiting on the user. Among them are Computer Use Bot, AI Research, Posting Buddy, SEO & AEO Desk, Etsy Bot and Design Buddy.
A "New agent" screen gives each one a name and a one-line job: "Drafts posts in my voice. Never posts without asking." It also has a character editor with shell color, hat, eyewear, extras and a "Surprise me" button.
We could not confirm where the clip comes from. The account is not a reliable source. The screens are not in Anthropic's 1:40 launch video, its X thread, the announcement or the help center. All of them describe Duet as one agent in one thread.
What stays with you
Duet reads, drafts and prepares work on its own. Before it sends a message, makes a payment, signs a document or deletes anything, it shows the exact action and waits for approval. Some actions stay blocked even with approval: trading securities, moving money between accounts and accessing someone else's account.
Anthropic describes four layers of protection:
- Credentials are handled like in Claude Managed Agents – Duet never sees passwords or sign-in tokens directly
- Actions go through the auto mode classifier used in Claude Code
- A new classifier blocks sharing of the strictest data tier, such as card numbers and government IDs. Health or personal finance data goes out only on your explicit instruction
- Red teams and an external firm ran penetration tests, and Duet is covered by Anthropic's bug bounty
Anthropic says it will publish more on Duet's safety design in the coming days.
Claude Duet vs OpenAI Dots and Grok Bot
On X, users framed Claude Duet as Anthropic's answer to OpenAI's Dots and SpaceXAI's Grok Bot. All three give the agent its own cloud computer and ask before consequential actions. The differences are in price, model and reach.
| Claude Duet | OpenAI Dots | Grok Bot | |
|---|---|---|---|
| Announced | Oct 21, 2026 | Sep 29, 2026 (DevDay) | Aug 11, 2026 |
| Model | Claude Opus 5.5 | GPT-6 Astra | Not disclosed, auto-routed |
| Access | Pro and Max, US only, waitlist | First dot included for Pro and Business Premium; Enterprise, Edu, Healthcare in beta | SuperGrok Heavy ($300/mo), Cursor Ultra ($200/mo), Cursor Premium Teams ($120/seat/mo) |
| Apps | 6,000+ connectors and plugins | 4,000+ apps via plugins | Not stated |
| Platforms | iOS, Android, claude.ai | ChatGPT app and web, Slack, Teams | macOS, Windows, Linux, iOS |
| Proactive mode | Checks every few hours, morning brief | Read-only background research when you're away | Learns over time when to act and when to ask |
Duet is the only one of the three that is consumer-first. It has no team or enterprise plans yet, and the pitch is built around a solo owner's inbox, invoices and calendar. Dots and Grok Bot both start from work teams: Slack and Teams for OpenAI, Cursor team seats for SpaceXAI.
What is still missing in the preview
Anthropic lists the preview's limits itself. Duet may misread a thread or misjudge what's urgent. Personal errands such as booking tickets are weaker than work tasks, and you can't watch or take over its browser. It can't make calls or send texts, and Microsoft 365 isn't available yet. For now, Duet lives in its own app, separate from the main Claude app.
