🔍 Read the full analysis: Why The Switchboard Matters In The Race For 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 as a waitlisted research preview on October 21, joining a growing race to build AI agents that work in the background. The source argues that control of connected apps, user memory and model routing—the “switchboard”—may matter as much as the underlying models; Duet’s capabilities and safeguards have not yet been independently tested.
Anthropic launched Claude Duet as a waitlisted research preview on October 21, adding a background-working AI agent to a contest that includes products from OpenAI, Meta and SpaceXAI. The launch puts a practical question at the center of the race: beyond which model answers best, who controls the agent’s access to apps, its memory and the choice of model doing the work?
Duet is a separate app available through iOS, Android and claude.ai, according to Anthropic. The company says it runs Claude on a cloud computer, can send a morning brief based on connected apps, and accepts tasks by text or voice in an ongoing thread. Users may be notified when something needs their attention, while the virtual machine can continue work when their own devices are off. Anthropic says the preview is built on Claude Opus 5.5 and is limited to US Pro and Max subscribers selected from a waitlist.
Anthropic describes tasks such as scheduling meetings, preparing dashboards and sales forecasts, and building small web apps. It says Duet can carry over memory, skills and connected apps from a user’s Claude account, and access more than 6,000 connectors and plugins in the Claude Marketplace. Examples cited by the company include checking invoices in QuickBooks or Stripe, payroll in Gusto, agreements through DocuSign and storefront work in Shopify. These are vendor descriptions; the supplied material reports no independent testing of Duet’s performance.
In a safety post dated October 23, Anthropic said Duet pauses for approval before sending sensitive messages, paying or deleting items, showing the proposed action, destination and content. The company says some activities—including securities trading and moving money between accounts—are prohibited even with approval. It also says Duet does not see passwords, browser cookies or sign-in tokens. Anthropic lists limitations, including checks of connected accounts only every few hours, possible errors in judging urgency, and weaker performance on personal errands. Background work counts against Claude usage limits.
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 May Outlast Model Leadership
The competition is not just about which model produces the strongest answer. An agent that can work across a user’s email, calendar, payments and business software may become the main access point for routine work. Its operator could shape which information the system sees, what it remembers, when it interrupts the user and which model receives a task.
The source describes that layer as the “switchboard”: control of app connections, memory and model routing. That framing is an interpretation, not a demonstrated market outcome, but it highlights a real decision for customers. If agents can send tasks to models from multiple providers, the service directing the task may retain the user relationship even as the underlying models change. The source says Grok Bot is intended to route work to outside models, including Claude Opus 5.5, and cites xAI’s FAQ as saying customers cannot restrict Grok Bot’s model choice through a team allowlist. The material does not establish how broadly that routing is used in practice.
For companies, the issue reaches beyond answer quality. Agents with persistent access can touch sensitive records and initiate consequential actions, so buyers will need to weigh permissions, approval design, data handling and the ability to leave alongside productivity claims. For individuals, the convenience of a morning brief or delegated task comes with choices about how many accounts to connect and what the agent is allowed to do.
Popular right now
Vitamix Propel Series 750 Professional-Grade Blender, 4 Programs, Black
Dyson Cyclone V10 Origin Cordless Vacuum
Ninja DualZone Smart Air Fryer, 10 Qt, 6-in-1, DZ950
As an affiliate, we earn on qualifying purchases.
A Crowded Launch Calendar
The source places Duet in a rapid sequence of product announcements. It says 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, citing TechTarget for the Google date. Anthropic announced Duet on October 21. The products are presented as sharing a broad aim: moving beyond a user-initiated chat toward agents that can operate in a cloud environment and return with results or requests for help.
That common description does not mean the services have the same functions, access rules or availability. Duet is a research preview with limited access, and the source gives launch-period descriptions rather than independent comparisons. Prices and capabilities can change, and the material does not provide a consistent benchmark across the products. Any claim that one agent is more capable or safer than another would need evidence beyond the announcements summarized here.
The source also points to a shift in how model providers may compete. On October 6, Elon Musk wrote that SpaceX would use the best backend model for a given task, naming Claude Opus 5.5, MidJourney and Suno among examples. The cited xAI FAQ says Grok Bot’s model choice cannot be restricted through its team model allowlist and that there is no customer-facing model picker. These statements suggest a service may route work behind its interface, but they do not by themselves show how routing decisions are made for every user or task.
““Approval fatigue is a serious issue, because too many requests teach people to tap through them.””
— Anthropic, in its October 23 safety post
Independent Tests Still Missing
Duet’s real-world accuracy, reliability and safety have not been independently established in the supplied material. Anthropic’s descriptions of its approval checks, data protections and connector support are company claims, and the source does not report a third-party audit or a hands-on test. It is also unclear how often users will encounter approval requests, how well the system handles complex or ambiguous tasks, and how its performance compares with rival agents under the same conditions.
Several operational details remain open. Anthropic says Duet checks connected accounts every few hours and that learned information remains in memory after an app is disconnected until memory is reset. The source does not specify how long other task data is retained, how routing among models works across competing services, or how customers can inspect and control every model handoff. The preview’s waitlist also leaves its wider availability and final feature set unsettled.
Watch for Access and Control
The next test is whether Anthropic expands Duet beyond its US Pro and Max waitlist and whether independent users can verify its advertised safeguards and task performance. Comparisons with Dots, Muse, Grok Bot and Gemini Agent will be more useful when their access, pricing and capabilities can be assessed on a consistent basis rather than from launch descriptions alone.
For organizations considering a pilot, the source recommends keeping trials small, documenting permissions and setting an exit plan. Users can limit early trials to low-stakes work, connect only a few apps and retain approval requirements for actions that send messages, spend money or delete information. Those steps do not settle broader questions about data retention or model routing, but they limit exposure while products and policies are still developing.
Key Questions
What is Claude Duet?
Claude Duet is Anthropic’s cloud-based agent, announced October 21, 2026, as a waitlisted research preview. Anthropic says users can assign tasks by text or voice while the agent works in a virtual computer environment.
Who can use Duet now?
According to Anthropic’s launch details in the source, access is limited to US Pro and Max subscribers selected from a waitlist. Broader availability has not been specified.
Can Duet send messages or make payments without approval?
Anthropic says Duet asks for approval before sensitive messages, payments and deletions. The company says securities trading and transfers between accounts are blocked, even if a user approves them. These safeguards have not been independently tested in the supplied material.
Why does model routing matter?
Routing determines which model handles a task, while the agent service may control the user’s app connections and ongoing interaction. The source argues that this control layer, or “switchboard,” could matter alongside the models themselves; its long-term competitive effect remains uncertain.
Source: ThorstenMeyerAI.com
