🔍 Read the full analysis: What Claude Duet’s Entry Reveals About The AI Agent Race on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Anthropic launched Claude Duet on October 21 as a waitlisted research preview for eligible US Pro and Max subscribers. Its cloud-based agent can prepare work across connected apps, but Anthropic says it requires approval for sensitive actions and blocks some financial actions outright; independent testing has not yet confirmed its performance or safeguards.
Anthropic launched Claude Duet on October 21, entering a growing contest to build AI agents that work in the background across users’ apps rather than waiting for prompts in a chat. The waitlisted research preview is available to US Pro and Max subscribers, with Anthropic describing it as a tool for professionals and small-business owners; its capabilities and safeguards have not yet been independently tested.
Duet is a separate app available on iOS, Android and claude.ai. Anthropic says it runs Claude on a cloud computer, sends a morning brief covering connected apps and accepts tasks by text or voice in an ongoing conversation. It can continue working while a user’s phone or computer is off, then message when it needs attention. The service is built on Claude Opus 5.5 and can use memory, skills and connected apps from the user’s Claude account.
Anthropic says Duet can handle tasks such as scheduling meetings and preparing dashboards, sales forecasts or small web apps. The company cites more than 6,000 connectors and plugins in the Claude Marketplace, with examples involving invoices in QuickBooks or Stripe, payroll in Gusto, agreements through DocuSign and storefront work in Shopify. Those are company-described features, not results from an independent evaluation.
Anthropic says Duet must show users the proposed action, its destination and its content before sending sensitive messages, making payments or deleting items. It says a separate classifier checks whether an action matches the user’s intent. Trading securities, transferring money between accounts and accessing another person’s account are blocked, even with approval. Anthropic also says Duet does not see passwords, browser cookies or sign-in tokens. These protections are the company’s account of the system’s design.
Infographic 01 · The permission ladder
What Claude Duet does alone, what it asks first, what it never does.
Anthropic’s three tiers for an agent that keeps working while you are away. Health or financial details go out only with your explicit consent.
Does it alone
no approval
- Reads connected apps: mail, calendar, Slack, invoices
- Sends a morning brief, then flags what needs you
- Drafts replies, reminders, decks, dashboards
- Prepares payroll or contracts for your review
- Checks accounts every few hours, not nonstop
Asks you first
approval card shows action, destination, content, recipients
- Sending a potentially sensitive message
- Paying (one-time card via Link, limited to merchant and amount)
- Signing a document
- Deleting anything
- Never on “don’t ask me again”: paying or sending money, deleting an order, creating an account
Never, even if you approve
blocked
- Trading securities
- Moving money between accounts
- Accessing someone else’s account
- Sharing sign-in credentials, full card or bank numbers, government ID numbers
- Seeing your passwords, cookies or sign-in tokens (kept off the model’s machines)
“Approval fatigue is a serious issue, because too many requests teach people to tap through them.”Anthropic, Claude Duet safety post
These are Anthropic’s stated rules for the research preview. No independent test has been published yet.
Sources: Anthropic, Introducing Claude Duet (21 Oct 2026); Anthropic, Duet safety post (23 Oct 2026); Claude Duet FAQ.
Powered by Thorsten Meyer AI
Agents Compete for the User’s Switchboard
Duet’s arrival adds to a shift from chatbots that answer questions toward agents that can take actions across connected services. For users, that could mean delegating routine work and receiving updates only when a decision is needed. It also gives an agent access to sensitive business and personal information, making permissions, approval flows and control over stored context central issues—not just the quality of its answers.
The source material places Duet alongside products from OpenAI, Meta, SpaceXAI and Google. It also points to a possible change in competition: if an assistant can route a task to different companies’ models, the company controlling the interface, connected apps and user context may shape the experience even when it does not supply every underlying model. That is an interpretation of the emerging market, not a settled finding about which company will prevail.
For organizations, the practical stakes include access to invoices, payroll, contracts and other records. A small pilot with limited permissions can reveal whether an agent saves time without granting it authority to perform consequential actions. The risks are concrete: an incorrect interpretation, missed update or hurried approval could have consequences when the agent is connected to real accounts.
Products Worth Considering
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER)
Dell UltraSharp 32 U3226QER 4K Thunderbolt Hub Monitor
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
A Crowded Launch Calendar
The source account describes a rapid series of agent launches: SpaceXAI opened the Grok Bot beta on August 11, Meta launched Muse on September 8, OpenAI introduced Dots at DevDay on September 29, and Google released Gemini Agent on October 8. Anthropic announced Duet on October 21. These services are presented as sharing a broad premise: an assistant can work in a cloud environment, retain context and return with a result or a request for input.
One development in the source material points to the growing role of model routing. On October 6, Elon Musk posted that SpaceX would use the best backend model for each task, including Claude Opus 5.5 and other services. The cited xAI security FAQ says Grok Bot’s model selection cannot be restricted through the team model allowlist and that customers do not have a model picker. This suggests that users may not always know which model handles a task, although the specific routing behavior and disclosures can vary by product.
Anthropic’s own description also makes clear that Duet remains an early product. It may misread a conversation or judge urgency incorrectly, is weaker at personal errands such as booking tickets and cannot yet show users its browser. It checks connected accounts every few hours, according to the company, rather than continuously, and background work draws on usage limits shared with Claude.
“for professionals and small business owners with more on their plates than they can track.”
— Anthropic, describing Claude Duet
Performance and Safeguards Await Testing
No independent test is provided for Duet’s accuracy, the effectiveness of its approval checks or how consistently it observes the stated restrictions. The launch information comes from Anthropic and the cited reporting; readers should treat capability and safety descriptions as company statements until outside testing or further evidence is available.
It is also unclear how often users will encounter missed updates, mistaken urgency judgments or approval fatigue in ordinary use. Anthropic says account checks happen every few hours and that learned information remains in memory after an app is disconnected until users reset that memory. The source does not establish how Duet performs across different tasks, what user controls exist for every type of stored information, or how its feature set may change during the preview.
The broader competitive picture is unsettled, too. The launch sequence does not establish how the products compare in reliability, pricing or adoption, and the source gives no consistent independent benchmark across them. Model routing may also make it harder for customers to determine which model handled a task, but the extent of that issue depends on each service’s design and disclosures.
Watch the Preview and Pilot Results
The next useful evidence will come from Duet’s waitlisted preview: how it performs on real workflows, how clearly it presents proposed actions and whether its safeguards work as Anthropic describes. The source material does not give a timetable for wider availability or a date for an independent evaluation.
Companies considering agent software can begin with small, written pilots, read-mostly access and a clear exit plan, as the source recommends. Users can start with low-stakes tasks, connect only the apps they need and keep approval requirements in place for anything that sends, pays or deletes. These steps do not establish that an agent is risk-free; they limit the consequences while its performance is being evaluated.
For the competing products, readers should watch for updates to access, permissions, model-selection controls, pricing and usage limits. Until more evidence is available, Duet’s launch confirms that Anthropic has joined the agent race—not which approach is safest, most capable or likely to lead it.
Key Questions
What is Claude Duet?
Claude Duet is Anthropic’s cloud-based AI agent, offered as a waitlisted research preview. Anthropic says it can work across connected apps and send users a brief or a message when it needs input.
Who can use Claude Duet?
According to the launch information, access is limited to US Pro and Max subscribers through a waitlist. Anthropic has not provided a wider-release date in the source material.
Can Duet send messages or make payments without approval?
Anthropic says it asks for approval before sensitive messages, payments and deletions. The company says trading securities and moving money between accounts are blocked outright. These are Anthropic’s stated safeguards and have not been independently tested in the source material.
What are Duet’s stated limitations?
Anthropic says the early version may misread conversations or urgency, is weaker at personal errands such as booking tickets and cannot yet show its browser. It checks connected accounts every few hours, and background work uses Claude’s shared usage limits.
What does Claude Duet’s launch say about the AI agent race?
It adds Anthropic to a crowded field that includes products from OpenAI, Meta, SpaceXAI and Google. The source suggests that control of the interface, connected apps and user context may matter alongside the underlying AI model, but it does not establish which company will lead.
Source: ThorstenMeyerAI.com
