🔍 Read the full analysis: Claude Duet And The Scramble To Coordinate Always-On AI Agents on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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 U.S. Pro and Max subscribers. The app can monitor connected services and prepare work in the background, but Anthropic says users must approve sensitive actions and some financial activity is blocked. Those capabilities and safeguards have not yet been independently tested.
Anthropic launched Claude Duet on October 21, adding a waitlisted research preview of an AI agent that works on a cloud computer, monitors connected apps and messages users when it needs direction. The product is available to U.S. Pro and Max subscribers, and its arrival adds Anthropic to a growing contest over assistants that keep working after a user leaves the chat.
Duet is a separate app available through iOS, Android and claude.ai. Anthropic says it can provide a morning brief, accept tasks by text or voice, and handle work such as scheduling, preparing a sales forecast or building a dashboard. Its virtual machine has a browser of its own, allowing work to continue while a user’s computer or phone is off. Anthropic says the agent uses the memory and connected apps associated with the user’s Claude account, including services accessed through more than 6,000 connectors and plugins in the Claude Marketplace.
The company’s examples include checking unpaid invoices through QuickBooks or Stripe, working with payroll in Gusto, handling agreements in DocuSign and updating a Shopify storefront. Anthropic says Duet can read, draft and prepare work independently, but it must ask for approval before actions such as sending a sensitive message, making a payment or deleting something. Its confirmation screen is intended to show the action, destination, content and recipients before the user approves it.
Anthropic also says some actions are blocked even with user approval: trading securities, moving money between accounts and accessing another person’s account. For purchases, the company says Duet uses Stripe’s Link wallet and a one-time card limited to a specified merchant and amount. These descriptions come from Anthropic’s product and safety materials; the source material reports no independent testing of the preview’s capabilities or safeguards.
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
Control of the Agent’s Switchboard
Duet’s launch is part of a broader shift from chatbots that respond on demand to agents that can monitor services and carry out multistep work in the background. That can save users time, but it also gives the agent ongoing access to business records, communications and personal context. Approval prompts, permission settings and the ability to inspect what an agent has done will matter as much as the quality of its answers.
The source material describes a second competition: control over the systems that route tasks among AI models. It reports that SpaceXAI’s Grok Bot may send work to models from other providers, including Claude Opus 5.5, while its security FAQ says users cannot restrict the model choice through a team allowlist and that there is no customer-facing model picker. If agents can select or hand work to outside models, the company controlling the interface and integrations may shape where user data goes and which services perform a task. The reported routing details and their practical effects require further verification.
For companies, a background agent could touch invoices, payroll, contracts and customer records without a worker keeping the app open. That makes limited permissions, written pilot rules and a way to stop or leave the service practical safeguards—not simply product preferences. For individual users, the risk is that frequent prompts become routine and approvals receive less attention.
Products Worth Considering
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
Jackery Explorer 2000 v2 Portable Power Station with Two 200W Solar Panels
Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Eucalyptus Green
As an affiliate, we earn on qualifying purchases.
A Crowded Field of Background Agents
The source material places Duet in a rapid sequence of launches: SpaceXAI opened a 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, according to TechTarget as cited by the source. Anthropic announced Duet on October 21. These dates and descriptions are drawn from the supplied report, rather than independently verified here.
The products share a broad premise: an assistant can hold context, operate through connected tools and return when a task is complete or requires a decision. Duet’s distinguishing pitch in the supplied material is its focus on professionals and small-business owners, with morning briefs and examples involving common business services. The preview’s waitlist and subscriber restrictions mean its availability is currently narrower than a general public release.
Preview Limits and Safeguards
Duet is a research preview, not an independently validated service, and the supplied material does not establish how accurately it handles real-world tasks or how its safety controls perform under varied conditions. Anthropic says the agent can misread conversations or misjudge urgency, is less capable with personal errands such as booking tickets, and cannot yet show users its browser. It checks connected accounts every few hours rather than continuously, which may leave gaps between checks.
Anthropic says background checks count toward the user’s shared Claude usage limits and that large tasks can take more model turns. The company also says disconnecting an app prevents new access, but information Duet has already learned remains until the user resets its memory. The source material does not spell out how long that retained information persists, how users can inspect it, or what happens to it after an account is closed.
Other details also remain unconfirmed outside Anthropic’s statements, including how consistently the approval screen identifies risk, how often users will be prompted and whether controls work as described across all connectors. The source offers launch-stage pricing comparisons but no complete, independently checked comparison of the competing products’ current features or prices.
What Users and Companies Can Test
For now, access depends on Anthropic’s waitlist and U.S. Pro or Max subscription. The supplied material gives no date for a wider release. As the preview develops, users and businesses can test whether Duet’s background monitoring, briefings and task execution are reliable enough for their needs—and whether its requests for approval are clear and manageable.
The source recommends that organizations begin with small, documented pilots, read-mostly permissions and a clear exit plan. Users are advised to start with low-stakes tasks, connect only a few apps, and keep approval requirements active for actions that send, pay or delete. Those steps do not verify Anthropic’s safeguards, but they limit the scope of an early trial while users observe how the agent handles information and prompts for decisions.
Key Questions
What is Claude Duet?
Claude Duet is Anthropic’s cloud-based AI agent preview. It runs through a separate app, works with connected services and can message users when a task needs attention.
Who can use Claude Duet?
According to the source material, Duet is waitlist-only for U.S. Pro and Max subscribers. Anthropic has not provided a broader release date in the supplied information.
Can Duet send messages or spend money without approval?
Anthropic says Duet must request approval before sensitive messages, payments and deletions. The company says securities trading and transfers between accounts are blocked, even if a user approves them.
Has Duet’s safety been independently tested?
The supplied material reports Anthropic’s own descriptions, not independent test results. Duet remains a research preview, and its real-world performance and safeguards have not been established by the information provided.
What should a business do before trying Duet?
The source advises starting with a small written pilot, read-mostly permissions and an exit plan. Users should connect only a few apps and keep approval requirements for actions that send, pay or delete.
Source: ThorstenMeyerAI.com
