AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Why AI’s Always-On Agent Race Is Turning Toward The Switchboard on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get audio and creator gear delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
As an affiliate, we earn on qualifying purchases.

TL;DR

Anthropic launched Claude Duet as a waitlisted research preview on October 21, joining a growing push toward agents that work in the background on cloud computers. The launch highlights a wider contest over who controls the routing layer connecting users, apps and AI models, but Duet’s capabilities and safeguards have not been independently tested.

Anthropic launched Claude Duet on October 21 as a waitlisted research preview, joining a growing race to sell AI agents that can keep working on a cloud computer when users are away. The product puts a new assistant into the same market as offerings from OpenAI, Meta and SpaceXAI, while the race is also shifting attention toward the service layer that routes tasks among models and connected apps.

Anthropic says Duet is a separate app available through iOS, Android and claude.ai, initially for US Pro and Max subscribers accepted from a waitlist. It is built on Claude Opus 5.5, according to the source material. Duet is designed to send a morning brief, take requests by text or voice, work in the background on a virtual computer, and contact the user when it needs a decision.

Anthropic’s examples include scheduling meetings, preparing dashboards and sales forecasts, and working with services used for invoices, payroll, contracts and online stores. The company says Duet can use more than 6,000 connectors and plugins in the Claude Marketplace. It also says the agent can prepare actions but must ask before sending sensitive messages, making payments or deleting items. Trading securities, moving money between accounts and accessing another person’s account are barred, even with user approval.

Those capabilities and protections are Anthropic’s descriptions, not the result of independent testing. The company says Duet shares Claude’s usage limits, checks connected accounts every few hours rather than continuously, and may misread a conversation or judge urgency incorrectly. The source material also says Duet cannot yet show users its browser, and that information it has already learned may remain in memory after an app is disconnected until that memory is reset.

At a glance
reportWhen: Claude Duet announced October 21, 2026;…
The developmentAnthropic launched Claude Duet, adding a waitlisted always-on agent to a field that includes products from OpenAI, Meta and SpaceXAI, while model routing raises questions about who controls the underlying service.
Thorsten Meyer AI
Vendor rules · research preview

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 Moves Beyond the Model

The commercial contest is not only about which model produces the best answer. Always-on agents need permission to read information across services, retain context and decide when to call a model or take an action. That makes the routing layer—or switchboard a point of control: it can determine which model handles a task, what information is available and how usage is measured.

The source material says SpaceXAI’s Grok Bot is intended to route tasks to outside models, including Claude Opus 5.5. It also cites an xAI security FAQ stating that the Grok Bot model choice cannot be restricted by a team model allowlist and that customers do not have a model picker. If those descriptions hold in practice, customers may use an assistant without choosing which model processes a particular request. That could make app access, memory and billing as consequential to buyers as model quality.

For businesses, an agent that can act across accounting, payroll or contract systems may save staff time, but a mistaken action or overshared data can carry real costs. A research preview with a waitlist offers little evidence yet about reliability at scale. Companies considering pilots will need to judge not just the agent’s outputs, but also its permissions, approval process, data handling and ability to stop or leave the service.

Products Worth Considering

Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black

Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

As an affiliate, we earn on qualifying purchases.

A Rapidly Growing Agent Field

The source article describes a sequence of product launches: SpaceXAI opened a Grok Bot beta on August 11, Meta launched Muse on September 8, OpenAI introduced Dots at its September 29 DevDay, and Google released Gemini Agent on October 8, according to TechTarget. Anthropic announced Duet on October 21. These dates and product descriptions come from the supplied source material; product availability and features may change.

The shared pitch is a move away from assistants that only respond inside a chat window. The products are presented as agents that can hold context, use a cloud computer and return with completed work or a request for input. Anthropic’s own safety post, published October 23 according to the source, describes approval prompts for consequential actions and acknowledges the risk of approval fatigue—users may stop scrutinizing requests if they receive too many.

Anthropic says Duet’s purchase flow uses Stripe Link with a one-time card restricted to a merchant and amount, and that the agent does not see passwords, browser cookies or sign-in tokens. It also says a classifier blocks sharing certain credentials and financial identifiers. These are company claims, and the source material does not provide independent verification or comparative test results for Duet and its rivals.

Reliability and Routing Remain Untested

No independent test of Claude Duet is cited in the source material, so its accuracy, speed and ability to follow approval rules in routine use remain unknown. It is also unclear how frequently users will encounter approval prompts, how well the system distinguishes routine from sensitive tasks, or how it will perform across its many connected apps.

The supplied material does not establish how Duet compares with Dots, Muse, Grok Bot or Gemini Agent in price, reliability, privacy or practical task completion. Nor does it provide details about how routed requests are logged, retained or handled by outside model providers. The exact commercial and technical terms may differ by product and change after launch. Claims about Duet’s limits and protections should be treated as Anthropic’s account until independently examined.

Pilots and Product Testing Ahead

Duet’s next step is wider access beyond its waitlisted US Pro and Max preview, though the source material gives no date for a broader rollout. Independent users and businesses will be able to test whether the agent’s background checks, memory controls and approval prompts work as described once they can use it at greater scale.

Companies weighing a trial can begin with a small, written pilot that uses read-mostly permissions, limits connected apps and sets a clear exit plan. Users can start with low-stakes tasks and keep approval requirements enabled for anything that sends, pays or deletes. Further disclosures about model routing, data retention, pricing and availability will help buyers compare the agents on more than their demonstrations.

Key Questions

What is Claude Duet?

Claude Duet is Anthropic’s agent app, announced October 21 as a waitlisted research preview. Anthropic says it can work on a cloud computer, provide morning briefs and handle tasks sent by text or voice.

Who can use Claude Duet now?

According to the source material, access is limited to US Pro and Max subscribers accepted from a waitlist. The source does not give a date for general availability.

Can Duet send messages or make payments without asking?

Anthropic says Duet must request approval before sending sensitive messages, paying or deleting items. The company says trading securities and moving money between accounts are prohibited even if a user approves.

What does the switchboard mean in the AI agent race?

It refers to the service layer that directs a request to a model or connected tool. Control of that layer can affect which model handles a task, what information it can use and how usage is charged.

Has Claude Duet’s performance been independently verified?

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

Source: ThorstenMeyerAI.com

You May Also Like

GPT-6 Astra Checked Claude’s Model Edit After The Tests Passed

OpenAI’s GPT-6 Astra reviewed a Claude-led model-editing experiment, finding reproducible test-rig and weight-conversion faults.

M. John Harrison Claims The 2026 Booker Prize With AI At The Edge Of Apocalypse

M. John Harrison won the 2026 Booker for The End of Everything, a near-future novel whose setting includes artificial intelligence.

Inside The AI Ballot Test That Asked Grok And GPT To Vote

A Convovo API test found GPT mostly chose Democrats and Grok mostly Republicans when forced to name candidates in 2026 races.