Claude Duet Brings Anthropic Into the Always-On Agent Race, With a Work-First Pitch
Key points
- Anthropic announced Claude Duet as a research preview on a waitlist for US Pro and Max subscribers, built on Claude Opus 5.5 and available on iOS, Android and claude.ai 8.
- Duet sends a morning brief, checks connected apps every few hours, runs on its own cloud computer, and asks approval before sending messages, paying, signing or deleting, while trading securities and moving money between accounts are off-limits 8.
- The product targets professionals and small business owners, with more than 6,000 connectors and plugins including QuickBooks, Stripe, Gusto, Docusign and Shopify, and it counts against existing plan usage limits 8 4.
- It lands amid rivals' troubles: Meta says Muse is unprofitable because of high compute costs 3, and a security firm found Muse's shared chats were discoverable through DNS records 9.
A Proactive Agent Arrives
Anthropic has opened a research preview of Claude Duet, an app in which Claude acts as a proactive agent. It surfaces what needs attention, completes tasks, keeps working when the user steps away, and checks in before acting on the user’s behalf 8. It is built around Claude Opus 5.5 and runs as a standalone iOS and Android app and on claude.ai 2.
Access is limited. Duet is on a waitlist for Pro and Max subscribers in the US, with invitations going out in waves, and Team and Enterprise plans are not included yet 4. It carries no extra charge but draws on existing usage limits, including background work, so heavy users may hit their caps sooner 4. Mike Krieger, a co-lead of Anthropic’s Labs team, said the tool has the judgment to know when to be proactive and when to act for the user 1.
The launch puts Anthropic alongside Meta’s Muse and OpenAI‘s Dots. One outlet reported that SpaceXSI’s Grok Bot is also in the field 7.
How Duet Actually Works
Each morning Duet sends a short brief on the most important items across connected apps, then messages the user only when something needs attention. Tasks range from scheduling meetings to building dashboards, financial or sales forecasts and web apps 8. It carries over memory, skills and connectors from the user’s Claude account and takes requests by text or voice in one ongoing thread 8. Its work runs in a virtual machine with its own browser, so a laptop or phone need not be on 2.
Anthropic says Duet checks accounts every few hours because constant polling would burn through plan limits 8. Its connectors reach over 6,000 tools, including Gmail, Slack, QuickBooks, Stripe, Gusto, HubSpot and Shopify. It can also buy things through Link or Shop Pay, with approval 8 4. Finished work comes back as files such as a deck or financial model 4.
Anthropic’s own tester stories show the pitch. One user said Duet caught a property showing placed on the wrong day, and another said it noticed a client call fell right before her flight and offered to move it 8.
Guardrails and Known Limits
Before Duet sends a message, makes a payment, signs a document or deletes anything, it shows the planned action and waits for approval. Trading securities and moving money between accounts are blocked even with approval 8. Anthropic says Duet never directly sees passwords or sign-in tokens, and that red teams and an external firm penetration tested it 8.
A new dedicated classifier and safety system prompt is meant to reject requests to share the strictest tier of information, such as credit card numbers and government ID 8 2. Anthropic is candid that this is an early version that may misread a thread, miscalibrate urgency, or miss something that arrives between check-ins, and that it is weaker at personal errands such as booking tickets 2 6.
Trending Topics adds open questions: prompt injection risk, little real-time visibility into the cloud browser, and fast-draining usage limits for Pro customers 7.
The Competitive Backdrop
Anthropic timed the announcement for the day of Meta’s earnings report and used it to highlight a hire: Karan Ahuja, who led product management for Meta Superintelligence Labs until July, led Duet’s development 1. Meta’s Alexandr Wang answered with a meme about Duet’s resemblance to Muse 1. Apptopia estimated Muse at more than 14 million installs, and found Claude users who installed Muse were using Claude more than before 1.
Strategy differs. Muse targets consumers booking trips and shopping, while Duet leans on Anthropic’s business-heavy revenue base, with invoices, payroll and contracts as its showcase 7. Separately, Meta’s Mark Zuckerberg said Muse is unprofitable because it is so compute intensive, though he expects efficiency gains to make it highly profitable over several years 3.
In a related development, security firm Claroc reported that Muse’s shared conversations were published at public addresses built from chat titles, which passive-DNS tools could list. It found sensitive personal and business documents, and Meta said it would suspend the feature after Ars Technica asked about it 9. It is a reminder of the stakes as agents gain access to more personal data 5.
Companies mentioned: Anthropic • Shopify (SHOP $174.29 ▲0.5%) • Meta (META $697.52 ▲0.5%) • OpenAI
Primary sources
Anthropic answers Meta and OpenAI with a new personal agent, Claude Duet (businessinsider.com) – Business Insider reports that Anthropic announced Claude Duet, a proactive background agent in research preview with a waitlist, built on Opus 5.5. It notes the launch came on Meta's earnings day, highlights Karan Ahuja's move from Meta Superintelligence Labs to lead Duet, and quotes Mike Krieger on the tool's judgment. It also covers Alexandr Wang's meme, Anthropic's possible IPO, and Apptopia's estimate of more than 14 million Muse installs. Business Insider.
Further sources
Anthropic takes on Muse with Claude Duet, its new AI agent (theverge.com) – The Verge describes Duet as an agent for professionals and small business owners, available on iOS, Android and the web, built around Opus 5.5. It covers the morning brief, task types, memory and connected apps, a virtual machine with a browser, approval before big steps, and the new classifier for strict-tier information. It also relays Anthropic's admission that the early version may misjudge urgency and is weaker at personal errands, and notes the US Pro or Max requirement. The Verge.Meta CEO Mark Zuckerberg says Muse and Reality Labs are highly unprofitable — for now (businessinsider.com) – Business Insider reports on Meta's third-quarter earnings call, where Zuckerberg said Muse is unprofitable because it is compute intensive but could become highly profitable if efficiency keeps improving. Meta plans a small transaction fee on purchases Muse drives. Muse launched only in the US and Canada, partly due to compute needs. Reality Labs is expected to narrow losses to about $15 billion in 2027 from $18 to $19 billion this year. Business Insider.
Anthropic takes on ChatGPT Dots and Meta Muse with Claude Duet (9to5mac.com) – 9to5Mac details the rollout: Duet is a research preview for US Pro and Max subscribers via waitlist, at no extra cost but counting toward usage limits, and not yet on Team or Enterprise. It runs on its own cloud computer, reuses Claude's memories, skills and connectors, integrates with Stripe, Link and Shop Pay, and has a Tasks tab and a Make tab. Only users can mark tasks finished. 9to5Mac.
A user told Meta’s Muse to sell his keyboard. What happened next was alarming (fastcompany.com) – Fast Company's piece covers a user who asked Meta's Muse to sell his keyboard and found the result alarming, framed around the privacy tradeoff of giving agents more personal information. Only a short excerpt was available, so details are limited. Fast Company.
Claude Duet (claude.com) – Claude's product page presents Duet as a proactive agent powered by Opus 5.5, for Pro and Max plans in the US. It stresses judgment, a single thread with memory, approval before sensitive actions, private cloud-side logins, and examples such as unpaid invoices and payroll prep. It states that background checks count toward usage limits and warns Duet may misread a thread or miss something between check-ins. claude.com.
Claude Duet: Anthropic’s New A.I. Agent Takes On Dots, Muse and Grok Bot (trendingtopics.eu) – Trending Topics offers a business-focused analysis. It covers access limits (US only, Europe excluded), the 6,000-plus connectors, blocked securities trading and transfers, and the competitive field including Dots, Muse and SpaceXSI's Grok Bot. It also ties Duet to Anthropic's enterprise-heavy revenue and lists risks: prompt injection, overeager agents, limited visibility into the cloud browser, and usage limits, citing early tester mishaps. Trending Topics.Claude Duet (claude.com) – Anthropic's official announcement introduces Duet as a research preview for professionals and small business owners, built around Opus 5.5. It describes the morning brief, checks every few hours, a virtual machine with a browser, and approvals with off-limits actions. It lists safety work (trustworthy-agents framework, credential handling, auto mode classifier, red teaming, bug bounty, and a new classifier for sensitive information), tester stories from Anthropic staff and small business owners, and the 6,000-plus connectors and Link payments. claude.com, Anthropic.
GhostShares: A Shared Muse Chat Exposed Prep for a Confidential Call With a Frontier AI Lab Leader (blog.claroc.io) – Claroc's research team reports that Muse's shared conversations were published at public links whose hostnames contain the chat title and a random ID, which get recorded in public DNS. Using passive-DNS data, they found more than 6,800 share hostnames and opened a sample revealing sensitive personal and business documents. They also found over 75 urlscan.io scans. Meta told Ars Technica it would suspend the feature, and Claroc confirmed shares stopped opening. Claroc blog, October 18, 2026, updated October 20.
Powered by News Ranker