Claude Duet’s AI Agent Entry Puts The Switchboard In Focus

🔍 Read the full analysis: Claude Duet’s AI Agent Entry Puts The Switchboard In Focus on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Anthropic launched Claude Duet on Oct. 21 as a waitlisted research preview, adding an AI agent designed to work across connected apps on its own cloud computer. The separate app is available to U.S. Pro and Max subscribers, according to the source report; Anthropic says Duet can prepare work proactively but must seek approval for sensitive actions such as sending messages, paying or deleting.

Duet is available through iOS, Android and claude.ai. Anthropic describes it as a service for professionals and small-business owners: it can send a morning brief based on connected apps, accept tasks by text or voice, and notify a user when something needs attention. Its work runs on a virtual machine with a browser, so the user’s own computer or phone does not have to remain on.

Anthropic says Duet carries over a user’s Claude account memory and skills and can use more than 6,000 connectors and plugins in the Claude Marketplace. Its examples include checking invoices in QuickBooks or Stripe, working with payroll in Gusto, handling agreements through DocuSign and managing a Shopify store. The source identifies the underlying model as Claude Opus 5.5; these product capabilities and figures are presented as Anthropic’s descriptions, not independent test results.

In a safety post dated Oct. 23, Anthropic said Duet shows the proposed action, destination and content, including recipients, before it sends a potentially sensitive message, pays or deletes anything. The user must approve. Anthropic also says some activities—including trading securities and moving money between accounts—are blocked, even if a user requests them. The source report says Anthropic’s purchase flow uses Stripe Link and a one-time card limited to a particular merchant and amount.

Control of the Agent Switchboard

Duet’s arrival adds Anthropic to a fast-moving market for agents that can act across services rather than only answer prompts. For users, the practical change is that an assistant may be able to read connected accounts, prepare work and return later with results. That convenience also makes permissions, approval design and data retention central questions, not secondary product details.

The source report argues that the competitive advantage may increasingly lie with the company operating the agent’s “switchboard”: the interface, its access to a user’s apps and memory, and the system that decides which model handles a task. It cites a statement by Elon Musk that Grok Bot would use models from other providers, including Claude Opus 5.5, and says xAI’s security FAQ indicates customers cannot restrict Grok Bot’s model choice. Those details suggest possible cross-company routing, but do not establish how Duet itself routes tasks or what model-selection controls its users will receive.

For businesses, an agent with access to invoices, payroll or contracts could affect sensitive workflows even when it cannot independently approve final actions. Anthropic’s stated safeguards may reduce some risks, but the preview’s actual reliability and the quality of its approval prompts have not been independently established in the material provided. A cautious pilot can help organizations test both utility and oversight before connecting more accounts.

A Crowded Agent Launch Cycle

The source report places Duet within a series of agent announcements. It says SpaceXAI opened the Grok Bot beta on Aug. 11, Meta launched Muse on Sept. 8, OpenAI introduced Dots at DevDay on Sept. 29, and Google released Gemini Agent on Oct. 8, citing TechTarget for the Google date. Anthropic’s Oct. 21 launch followed those developments. The products are described as sharing a broad approach: an assistant can continue working in a cloud environment and return with an update or request for help.

The details and access terms differ, and the supplied material does not provide enough comparable, independently verified information to rank the products. It also says prices change frequently and refers to launch figures without including a full comparison table. Duet’s status is a research preview, not a generally available release, and its waitlist limits who can assess it directly.

“for professionals and small business owners with more on their plates than they can track.”

— Anthropic, describing the intended audience for Claude Duet

What the Preview Has Not Proved

No independent testing is cited for Duet’s safety controls, performance or ability to complete the examples described. Anthropic itself says the early version may misread a conversation or misjudge urgency; personal errands such as ticket booking are weaker, and users cannot yet see the browser it uses. It checks connected accounts every few hours, according to the source, which can leave gaps between checks.

The report says background checks use the same Claude plan limits as other activity, while large tasks may require more model turns. Disconnecting an app prevents new access, but information Duet has already learned remains until its memory is reset. The supplied material does not specify how long the waitlist will last, when wider access may begin, or whether features and restrictions will change before a full release. It also does not establish how Duet’s model selection works or whether users can choose a model for individual tasks.

Testing Access and Permissions

Anthropic has not provided a broader release date in the source material. The next evidence will come from the preview itself: whether waitlisted subscribers can use Duet reliably, how often it requests approval, and how well its account checks identify time-sensitive items. Any changes to the stated restrictions, app connections or memory controls would also clarify how the product is developing.

The source report recommends that companies begin with small, written pilots, mostly read-only permissions and an exit plan. Users are advised to start with low-stakes tasks, connect only a few apps and retain approvals for actions that send, pay or delete. Those precautions are recommendations, not Anthropic requirements; they reflect the uncertainty around an early product whose claimed capabilities have not yet been independently assessed.

Key Questions

What is Claude Duet?

Claude Duet is Anthropic’s separate AI-agent app, launched as a waitlisted research preview on Oct. 21, 2026. Anthropic says it works across connected apps using a cloud computer and can provide briefs, carry out tasks and request attention when needed.

Who can use the preview?

The source report says access is limited to U.S. Pro and Max subscribers by waitlist. It does not give a date for general availability or say when the waitlist will open more broadly.

Can Duet send messages or make payments without approval?

Anthropic says Duet must show the action and wait for approval before it sends potentially sensitive messages, pays or deletes. The company says trading securities and moving money between accounts are blocked. These safeguards have not been independently tested in the material provided.

What limitations has Anthropic identified?

Anthropic says the preview may misread conversations or urgency, is weaker at personal errands, cannot yet show its browser and checks connected accounts every few hours. The source also says background activity counts toward Claude usage limits.

Does disconnecting an app erase what Duet already learned?

Not according to the source report. Disconnecting an app stops new access, but information Duet has already learned remains until the user resets its memory.

Source: ThorstenMeyerAI.com