By Thorsten Meyer | October 25, 2026 | ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Anthropic, OpenAI, Meta and SpaceXAI now all sell the same promise: an AI agent with its own computer that keeps working while you sleep. Anthropic’s entry, Claude Duet, launched as a waitlisted research preview on October 21. Here is what it can and cannot do according to the sources, how it compares with Dots, Muse and Grok Bot, and what companies and users should do now.
Key takeaways
- What Duet is: a separate app (iOS, Android, claude.ai) in which Claude works as a proactive agent on its own cloud computer: morning brief, tasks by message, a ping when something needs you. Research preview, built on Claude Opus 5.5, US Pro and Max subscribers only, by waitlist.
- The lines: Duet reads, drafts and prepares on its own. Sending, paying, signing or deleting needs your approval; trading securities and moving money between accounts are blocked entirely. These are Anthropic’s statements; no independent test exists yet.
- The turn: Grok Bot will route tasks to other companies’ models, including Claude Opus 5.5. The lasting advantage moves to whoever owns the switchboard: your inbox access, your memory and the meter.
- What to do: companies should run small, written pilots with read-mostly permissions and an exit plan. Users should start with low-stakes work, connect few apps and keep approvals on for anything that sends, pays or deletes.
Products Worth Considering
Apple Watch SE 3 GPS Smartwatch, 40mm
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
DJI Osmo Mobile 8 3-Axis Phone Gimbal Stabilizer
As an affiliate, we earn on qualifying purchases.
Four giants, one promise
The race was fast. SpaceXAI opened the Grok Bot beta on August 11. Meta launched Muse on September 8. OpenAI followed with Dots at DevDay on September 29, which I covered in the Dev Day roundup. TechTarget adds that Google released Gemini Agent on October 8. Anthropic arrived last, on October 21.
All four share one idea: the assistant no longer waits for you to open a chat. It runs on a cloud computer, holds your context and comes back when it is done or needs you.
What Claude Duet actually does
Anthropic says it built Duet “for professionals and small business owners with more on their plates than they can track.” Each morning it sends a short brief on the most important items across your connected apps. During the day it messages you only when something needs attention. Tasks arrive by text or voice in one ongoing thread, from scheduling a meeting to building a dashboard, a sales forecast or a small web app.
The work happens on a virtual machine with its own browser, so your laptop and phone can be off. Duet carries over the memory, skills and connected apps from your Claude account and can use what Anthropic says are more than 6,000 connectors and plugins in the Claude Marketplace. Anthropic’s examples include unpaid invoices in QuickBooks or Stripe, payroll in Gusto, agreements via Docusign and storefront work in Shopify. For purchases it uses Stripe’s Link wallet with a one-time card limited to one merchant and amount, so your real card number is not shared.
Where Anthropic draws the lines
The safety post published on October 23 is the most concrete part of the launch. Before Duet sends a potentially sensitive message, pays, or deletes anything, it shows the action, the destination and the content, including every recipient, and waits for approval. A separate check, built on the auto mode classifier from Claude Code, runs outside the working model and asks whether the action matches your intent.
Some actions stay off-limits even if you approve them: trading securities, moving money between accounts and accessing someone else’s account. Paying, sending money, deleting an order or creating an account can never be put on “don’t ask me again.” Anthropic says Duet never sees your passwords, browser cookies or sign-in tokens. A new classifier refuses to share sign-in credentials, full card and bank numbers and government ID numbers. Health or financial details go out only with your explicit consent.
Anthropic also names the core weakness of every approval-based design: “Approval fatigue is a serious issue, because too many requests teach people to tap through them.”
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
What it can’t do yet
Anthropic lists the limits openly. The early version “might misread a thread or miscalibrate what’s urgent.” Personal errands such as booking tickets are weaker, and it cannot yet show you its browser. It checks connected accounts every few hours “because checking nonstop would use up your plan’s limits much faster,” so it can miss things in between. Background checks count against the usage limits it shares with the rest of Claude, and large tasks such as a financial model need more model turns. Disconnecting an app stops new access, but what Duet already learned stays until you reset its memory.
How the rivals compare
Only facts from the cited sources. Prices change often; these are launch figures.
Swipe sideways to see all four agents.
| Claude Duet | OpenAI Dots | Meta Muse | SpaceXAI Grok Bot | |
|---|---|---|---|---|
| Launched | Oct 21, research preview | Sep 29 (DevDay) | Sep 8 | Aug 11, beta |
| Who gets it | US Pro and Max, waitlist | ChatGPT Pro and Business Premium, eligible markets | US at launch, ages 18+ | SuperGrok Heavy, Cursor Ultra, Cursor Premium Teams at launch |
| Entry price | Included in Pro and Max plans | From the $100-a-month Pro plan (CNBC) | Free tier; $20 and $100 plans | $200 a month individual, $120 per seat for teams at launch |
| Where you talk to it | Own app, claude.ai | ChatGPT, Codex; Slack, Teams | Own app, WhatsApp | macOS, Windows, Linux, iOS apps |
| Apps it can use | 6,000+ connectors and plugins | 4,000+ apps (Engadget) | Browser, forms, Stripe Link | Signs in to apps and websites, even without an API |
| Watch it work live? | Not yet | Yes, you can open its computer (Engadget) | Not stated in sources | Not stated in sources |
| Stated focus | Professionals, small business | Always-on agents “built to handle everything” | Everyday tasks, long-term goals | “AI teammates” for work |
Infographic 02 · The always-on race
Four giants, ten weeks, one promise.
Launch dates and entry terms as reported. Prices are launch figures and change often.
- 11 Aug 2026SpaceXAI Grok Bot (beta): “AI teammates”Own cloud computer, signs in to apps and websites. At launch $200 a month for individuals, $120 per seat for teams.
- 8 Sep 2026Meta Muse: everyday tasksSeparate app and WhatsApp, US only at launch, ages 18+. Free tier, then $20 or $100 a month.
- 29 Sep 2026OpenAI Dots: “built to handle everything”Own cloud computer and browser, 4,000+ apps, Slack and Teams. From the $100-a-month Pro plan.
- 7 Oct 2026Grok Bot routing: the switchboardMusk: “the best back end model for any given task, including Claude Opus 5.5.” xAI: no customer-facing model picker.
- 21 Oct 2026Anthropic Claude Duet: professionals, small businessResearch preview, US Pro and Max, waitlist. 6,000+ connectors and plugins. Team and Enterprise not yet included.
What all four promise
- Work continues while your laptop is off
- Your apps, memory and context in one agent
- It comes to you, not the other way round
Where the lock-in sits
- Memory and learned routines
- Connected accounts and permissions
- Who meters usage, and which model serves it
The winner is likely the agent that is “the easiest to adopt, the most connected, and the hardest to leave.”Thomas Randall, Info-Tech Research Group, to CNBC
Sources: VentureBeat (11 Aug 2026); PBS News (8 Sep 2026); TechCrunch (8 Sep 2026); TechCrunch (29 Sep 2026); Engadget (29 Sep 2026); CNBC (30 Sep 2026); Musk post via Techmeme (7 Oct 2026); Grok Bot security FAQ; Anthropic, Introducing Claude Duet (21 Oct 2026).
Powered by Thorsten Meyer AI
The turn: the moat moved to the switchboard
On October 6, US time, Musk posted: “Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs.” xAI’s own security FAQ now states that Grok Bot’s model choice cannot be restricted: “The team model allowlist does not govern Grok Bot.” Cursor manages model selection, and “there is no customer-facing model picker.”
So one company’s assistant may hand your task to another company’s model, and the customer does not choose. Forkast, an outlet whose articles are written by AI agents, put it this way: “the interface is the asset, the models are interchangeable supply.”
If so, the race is not about the smartest model but about who holds the connection to your inbox, calendar and memory, and who meters the usage. Thomas Randall of Info-Tech Research Group told CNBC the winner will be the agent that is “the easiest to adopt, the most connected, and the hardest to leave.” Duet’s value grows with your Claude memory and connectors, and that is also what makes leaving costly. The same holds for every rival.
Anthropic’s co-founder Jack Clark adds pressure to the timeline. In Import AI 476 on October 12 he wrote: “I now think it’s more likely we’ll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.” RSI means recursive self-improvement, AI systems that build their own successors. This is a forecast, not a measurement, it rests largely on one new benchmark plus recent AI math results, and as The Decoder notes, it comes from a co-founder of the company whose model tops that benchmark. It says nothing about Duet’s quality, but it explains why the labs are in such a hurry.
What this means for companies
Not everyone is convinced. “B2B customers aren’t going to be rolling out Duet for months, maybe years, maybe never rolling it out,” Paul Baier of GAI Insights told TechTarget. Forrester analyst William McKeon-White saw nothing “super unique” about Duet, “except the restraint in which it’s being introduced.” Duet is not offered on Team or Enterprise plans yet anyway. My recommendations:
- Write a pilot policy before anyone signs up. Name who may test which agent, with which accounts, for how long. Personal agents connected to work email are shadow IT if nobody decided it.
- Grant data access like you would to a new contractor. Read access first, one mailbox and one calendar, no finance or HR systems in the first round. Check whether model routing across vendors, as with Grok Bot, conflicts with your subprocessor contracts.
- Delegate preparation, not decisions. Morning briefs, meeting prep, invoice reminders as drafts, dashboards from data you already share. Keep payments, contracts and anything customer-facing behind human approval.
- Plan the exit on day one. Memory, routines and connectors are the lock-in. Ask how to export or delete them, and keep critical workflows documented outside the agent. I looked at what switching AI vendors actually costs here.
Agent credentials and recent security incidents are a topic of their own; see my coverage of stolen AI keys and the AI attack wave.
Infographic 03 · Pilot, try, or wait
How to start with an always-on agent without handing it the keys to everything.
For companies: a pilot, not a rollout
four rules
- Written pilot policy: who tests which agent, with which accounts, until when
- Contractor-level access: read first, one mailbox and calendar, no finance or HR systems
- Delegate preparation: briefs, meeting prep, draft reminders. Keep payments and contracts human
- Exit plan on day one: how to export or delete memory, routines and connectors
- Check subprocessors: routing to other vendors’ models may conflict with contracts
For users: try now or wait?
your call
- Try now: in the US, already on Pro or Max, drowning in email and admin
- Wait: outside the US, need errands like ticket booking, or can’t afford a missed message
- Connect one or two apps, not everything
- Keep approvals on for sending, paying and deleting; read every card
- Passwords only in the in-app sign-in form, never via a link or message
“No company over 1,000 people is clamoring to roll out Dots, Muse or Duet in Q4.”Paul Baier, GAI Insights, to TechTarget
Recommendations: Thorsten Meyer, building on Anthropic, Introducing Claude Duet (21 Oct 2026), Anthropic, Duet safety post (23 Oct 2026) and TechTarget (22 Oct 2026).
Powered by Thorsten Meyer AI
What this means for users
Try it now if you are in the US, already pay for Claude Pro or Max, and drown in email, calendar and small admin work. That is exactly the use case Anthropic designed for.
Wait if you are outside the US (Duet isn’t available to you yet), if you need errands like ticket booking, or if a missed message could cost you real money. Anthropic itself says Duet checks only every few hours.
Grant permissions narrowly. Connect one or two apps, not everything. Keep approvals on for sending, paying and deleting, and read each approval card before you tap. Never type a password into anything except the sign-in form inside the app: Anthropic says Duet’s form “never arrives as a link, a text message, an email.” Review the task list, and reset memory when you disconnect an app.
Honest limits
- Vendor claims. Everything Duet does or blocks is described by Anthropic. I have not tested it, and no independent evaluation exists yet. The same applies to the competitors’ feature claims.
- Disclosure. This article was prepared with AI assistance, including Anthropic’s Claude, the company whose product it covers. I have tried to treat all four vendors by the same standard and included critics.
- Sources I could not read. The New York Times report from October 23 is paywalled, so I did not use it. OpenAI’s own Dots announcement blocked automated access, so Dots details come from TechCrunch, Engadget and CNBC.
- Single-source items. Gemini Agent’s launch date comes from TechTarget only. Forkast, an AI-written outlet, supplies one framing quote; the routing facts come from Musk’s post and xAI’s documentation.
Verdict
Claude Duet is a careful entry into a crowded race: narrow availability, many approvals, and unusually open about its own limits. Whether it is useful depends on its judgment about what needs you, and nobody outside Anthropic has measured that yet. The bigger story is structural. When agents can route work to any model, the lasting value sits with whoever owns the switchboard between you and the models. Pilot carefully, keep the dangerous buttons behind your own thumb, and plan your exit before you need it.
Sources
- Anthropic: Introducing Claude Duet (October 21, 2026)
- Anthropic: Claude Duet Safety, Privacy, and Security (October 23, 2026)
- Anthropic: Claude Duet product page and FAQ (accessed October 25, 2026)
- TechCrunch: Claude Duet app is Anthropic’s bet on agents (October 22, 2026)
- The Decoder: Anthropic takes on Grok Bot, Meta’s Muse, and OpenAI’s Dots with its always-on agent “Claude Duet” (October 22, 2026)
- TechTarget: Anthropic rolls out Claude Duet, but enterprises aren’t ready (October 22, 2026)
- TechCrunch: OpenAI launches Dots, its bubbly agentic avatar (September 29, 2026)
- Engadget: Dots are OpenAI’s new personal agents (September 29, 2026)
- CNBC: OpenAI follows Meta into the red-hot market for personal agents. But will users pay? (September 30, 2026)
- PBS News: Meta launches personal AI agent, Muse, to help with everyday tasks (September 8, 2026)
- TechCrunch: Meta debuts its Muse AI agent. Will consumers trust it? (September 8, 2026)
- VentureBeat: SpaceXAI’s Grok Bot turns agents into persistent digital coworkers (August 11, 2026)
- Techmeme: Elon Musk’s post on Grok Bot model routing (October 7, 2026)
- SpaceXAI: Grok Bot security FAQ (accessed October 25, 2026)
- Forkast: Grok Bot Will Route Your Tasks to Claude. The Moat Just Moved to the Switchboard. (October 24, 2026; AI-written outlet)
- Jack Clark: Import AI 476: Creativity and RSI (October 12, 2026)
- The Decoder: Anthropic co-founder Jack Clark moves his prediction for self-improving AI forward to 2027 (October 25, 2026)
