Why Claude Duet’s Always-On AI Agent Push Centers On The Switchboard

🔍 Read the full analysis: Why Claude Duet’s Always-On AI Agent Push Centers On The Switchboard 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 October 21 as a waitlisted research preview, adding a cloud-based agent that can work across connected apps and contact users when it needs input. The service is available to U.S. Pro and Max subscribers, and puts Anthropic into a growing competition over assistants that keep working beyond a single chat session.

Duet is a separate app available through iOS, Android and claude.ai. Anthropic says it can send a morning briefing, accept tasks by text or voice, and perform work on a virtual computer with its own browser while the user’s devices are offline. The company describes intended users as professionals and small-business owners, with examples including tracking invoices, preparing payroll tasks, handling agreements and working on a Shopify storefront.

Anthropic says Duet carries over memory, skills and connected apps from a user’s Claude account, and can draw on more than 6,000 Marketplace connectors and plugins. Its examples include QuickBooks, Stripe, Gusto, DocuSign and Shopify. The company says purchases use Stripe Link and a one-time card limited to a specific merchant and amount, rather than exposing the user’s ordinary card number.

Anthropic’s October 23 safety post says Duet must show users the action, destination and content before it sends a potentially sensitive message, pays or deletes something. A separate classifier checks whether the proposed action matches the user’s intent. The company says trading securities, transferring money between accounts and accessing another person’s account are blocked, even with user approval. These are Anthropic’s descriptions of the product; the source material reports no independent test of the preview.

Control of the Agent Switchboard

Duet’s launch is part of a shift from chatbots that respond on demand to agents that can act across connected services and return later with results or questions. For users, that could reduce the work of checking multiple apps and moving information between them. It also raises the stakes of granting an assistant access to email, business records, calendars and payment tools.

The competitive question is not only which model produces the best answer. It is also which company controls the interface, app connections, user memory and routing decisions. The source material says SpaceXAI’s Grok Bot may route tasks to models from other companies, including Claude Opus 5.5, and cites xAI’s FAQ saying customers cannot restrict the model choice for Grok Bot. If that account is accurate, the agent operator—not necessarily the model maker—could decide which system handles a task. That makes the “switchboard” a useful way to describe a potential point of control, rather than a confirmed advantage held by any one company.

For businesses, the immediate issue is governance: what data an agent can read, which actions it can take, and how staff verify its work. Anthropic’s approval prompts offer one stated safeguard, but approvals do not eliminate the risk of mistakes or users becoming accustomed to approving requests without review.

A Crowded Agent Launch Race

The source account places Duet in a quick sequence of product launches. It says SpaceXAI opened a Grok Bot beta on August 11, 2026, Meta launched Muse on September 8, OpenAI introduced Dots at its September 29 DevDay, and Google released Gemini Agent on October 8. Anthropic announced Duet on October 21. These products are presented as variations on a similar idea: an assistant that operates in a cloud environment, retains context and can follow up when a task is complete or requires attention.

Duet is built on Claude Opus 5.5, according to the source material, and uses the same usage limits as the rest of Claude. Anthropic says it checks connected accounts every few hours rather than continuously, partly to limit usage. It also says the preview can misread message threads, judge urgency poorly, and handle personal errands such as booking tickets less effectively. Users cannot yet watch its browser activity, and information Duet has already learned may remain in memory after an app is disconnected until that memory is reset.

““Approval fatigue is a serious issue, because too many requests teach people to tap through them.””

— Anthropic, in its October 23 safety post

Preview Limits and Open Questions

Duet is still a research preview, and the material supplied does not include independent performance or security testing. It is not clear how reliably the system will handle complex tasks, how often it will make errors, or how Anthropic will respond when an action is approved but the agent has misunderstood the user’s intent.

The source also does not provide a full comparison of Duet’s price, capabilities or safeguards against competing products. Anthropic says app checks happen every few hours, but does not give a precise schedule for each service. It says disconnecting an app blocks future access while previously learned information remains until memory is reset; the material does not explain in detail how retained information is stored or removed. The scope of Duet’s availability beyond U.S. Pro and Max subscribers is also unspecified.

What Early Users Should Watch

The next test is how Duet performs as more waitlisted users gain access and whether Anthropic changes its limits, permissions or availability during the preview. Users and companies should watch for clearer information about the frequency of account checks, memory deletion, error reporting and the way approval prompts work in practice.

For now, the source material recommends that companies run small, documented pilots, begin with read-mostly permissions and keep an exit plan. Individual users can start with low-stakes tasks, connect only a few apps and retain approval requirements for actions that send messages, spend money or delete information. Those steps reduce exposure while leaving open the central question of whether proactive agents can be useful without asking people to hand over too much control.

Key Questions

What is Claude Duet?

Claude Duet is Anthropic’s separate app for a proactive AI agent that works on a cloud computer, uses connected apps and can contact users when it needs attention. Anthropic launched it as a research preview on October 21, 2026.

Who can use Claude Duet?

According to the source material, the waitlisted preview is for U.S. Claude Pro and Max subscribers. Wider availability has not been specified.

Can Duet send messages or spend money without approval?

Anthropic says Duet waits for approval before sending potentially sensitive messages, paying or deleting information. The company says some actions, including trading securities and moving money between accounts, are blocked even if a user approves them.

Why does the switchboard matter?

The switchboard refers to control over the agent’s interface, connected apps, user context and choice of model for a task. The source material says some agent products may route work to models from multiple providers, potentially giving the agent operator influence over which model handles a request.

Has Claude Duet been independently tested?

The supplied source material reports no independent test of Duet. Its descriptions of safeguards, capabilities and limits are attributed to Anthropic.

Source: ThorstenMeyerAI.com