Grok Bot, ChatGPT dots and Claude Duet: what to settle before staff use one

OpenAI introduced dots on 29 September 2026. A dot is an always-on agent powered by GPT-6 Astra, with its own cloud computer and browser. It learns from feedback and works toward goals around the clock. Plugins connect it to more than 4,000 apps.

Staff reach their dot in ChatGPT on desktop, web and mobile, in Slack and in Teams, and by voice call. Texting is coming soon, and context carries across every channel. You can open the dot's computer at any time to inspect its work. Your own computer stays separate unless you give the dot permission to use it.

In the background, a dot runs what OpenAI calls proactive research. It uses the apps you've connected, with tools limited to reading. It can't send messages, change what is in an app or control your browser or computer.

Permissions run through the ChatGPT app controls a workspace already has. Built-in rules decide when a dot acts and when it asks first. Custom Rules let you allow a specific action, require approval for it or block it. An Activity View shows the dot's progress, background work included.

Auto-review checks any action that could affect an account or share information against your instructions, your Custom Rules and OpenAI's safety requirements. Some sensitive tasks, such as changing a password, always stay with you. Monitoring can pause or stop a dot over a safety concern, and saved passwords for supported sites are used without the model seeing them. OpenAI still tells users to review consequential work, because dots can make mistakes.

OpenAI doesn't use content from ChatGPT Business, Enterprise or Edu workspaces to improve its models by default. On personal plans, users choose whether a dot's conversations and work are used. OpenAI doesn't train directly on proactive research or on a dot's notes to itself. Information from them may still inform an eligible conversation or task, depending on your settings.

Dots are rolling out to Pro and Business Premium in eligible markets. Enterprise, including Edu and Healthcare, can try the beta once a workspace admin enables it.

The first dot comes with Pro or Business Premium at no extra cost. The plan includes an allowance for deeper work, with extended limits for the first month. Talking to a dot doesn't count toward ChatGPT usage limits. Tasks it starts in Codex or ChatGPT Work count as usual.

Specialist dots are a preview, run as focused enterprise pilots with OpenAI engineers. Each one has its own identity, credentials and IT-provisioned hardware, with deep integrations into systems of record. OpenAI is working with Microsoft to bring them under the governance and security controls in Microsoft Agent 365.

Dots arrive inside a ChatGPT workspace that a business may already run. An admin decides when the Enterprise beta starts, and a dot's rules live in controls the IT team already manages.

Keeping background research to reading lowers the risk of a dot acting while nobody is watching. Anything that sends, shares or changes goes through the rules and auto-review. Those rules only cover what someone writes into them, so writing the Custom Rules is the first piece of work.

The first dot comes with Pro and Business Premium. A company on another Business plan would likely need to change plan to get one, and OpenAI hasn't published a price for extra dots.

Anthropic introduced Claude Duet on 21 October 2026 as a research preview. Access is by waitlist, starting with Pro and Max subscribers in the US. Anthropic says Duet is built for professionals and small business owners.

Duet is its own app on iOS and Android and on claude.ai, separate from the main Claude app for now. It is built around Claude Opus 5.5 and runs on a virtual machine with its own browser, so your laptop and phone can be off. It can't yet show you that browser.

Duet brings over the memory, skills and connected apps from your Claude account. You talk to it in one ongoing thread, by typing or by voice. Each morning it sends a brief of the important items across your connected apps.

It checks those accounts every few hours, because Anthropic says checking nonstop would use up plan limits much faster. It messages you when something needs attention, and you choose what you hear about and how often.

Before Duet sends a message, makes a payment, signs a document or deletes anything, it shows exactly what it will do. Then it waits for your approval. Some actions stay off-limits even with approval, such as trading securities or moving money between accounts. You can disconnect any app, and Anthropic says its auto mode classifier approves the actions Duet takes.

Duet's sign-ins follow the design of Claude Managed Agents, so it never directly sees passwords or sign-in tokens. A new classifier and safety instructions check each message for leaks of sensitive information. Duet refuses to share the strictest kind, such as credit card numbers and government ID. It shares other sensitive details, such as health or personal finances, only when you have told it to.

Anthropic also cites defences against prompt injection, drawn from its research on browser use. Prompt injection is an instruction hidden in a page or message the agent reads. Red teams and an outside firm tested Duet for security weaknesses, and it is covered by Anthropic's public bug bounty.

The Claude Marketplace lists over 6,000 connectors and plugins. Anthropic's examples include Gmail, Slack, QuickBooks, Stripe, Gusto payroll, Docusign, HubSpot, Canva, Shopify and LinkedIn job posts after approval. Duet buys with a virtual card through Link, Stripe's wallet, and checks out with Shop Pay. Signing in with logins saved in 1Password is coming soon.

Duet is included in Pro and Max during the preview and uses the same usage limits as the rest of Claude. Background checks count toward those limits, and larger jobs such as a financial model or a deck use more. Anthropic says Duet may misread a thread or misjudge how urgent something is, and that it is weaker at personal errands than at work tasks.

Duet is aimed at one person, and the announcement doesn't mention Team or Enterprise plans or admin controls. A member of staff on a personal Pro or Max plan could connect work email or a payroll tool. That connection is likely to sit outside anything your IT team can see or switch off.

The approval step covers the actions that cost most to undo: sending, paying, signing and deleting. The block on trading securities and moving money between accounts holds even when someone approves. A business still needs its own rule on which company accounts may be connected at all. The preview gives an admin nowhere to set one.

The announcement doesn't state Duet's terms on training with your data. Confirm them before any company data goes in.

Grok Bot's pages sit on x.ai, whose footer reads SpaceXAI LLC. They describe AI teammates that sign in to your tools, use them as you do and come back with finished work. Grok Bot runs on the desktop and on iOS. The page names macOS, and the FAQ adds Windows.

You can run many Bots at once, around the clock, even with your laptop closed. Each Bot has a name, an avatar and a title, remembers its conversations and keeps its own memory and Routines. Tools and Skills sit at account level, so every Bot on the account can use them.

Several Bots can share one thread and pass work between themselves. The page says you watch them take action rather than approve every step. The design article puts the limits at about 50 Bots per account and six in a group chat.

With "teach a task", a Bot follows along while you complete a workflow once, saves it as a Routine and runs it itself next time. Routines run on a schedule or on events. The examples include a pull request being opened, an incident starting, a message from another system (a webhook) and new messages in a channel. The computer view shows a status icon, a preview panel or a full takeover, where you take control and hand it back.

The FAQ says every Grok Bot shares one persistent cloud computer, with its files, browser and logins. Isolation is per user rather than per Bot.

That computer is encrypted in transit and at rest, and the FAQ mentions a training opt-out. Sensitive actions can go through Auto Review before they run. Enterprise admins can set data loss prevention (DLP), certificates, proxies and network controls when the computer starts.

Individual plans are billed monthly. Premium is $30 a month and includes a cloud computer for your Bots, sign-in to your tools and scheduled Routines.

Super is $100 with 3.5 times Premium's usage and priority model access at busy times. Ultra is $200 with 12 times Premium's usage and earliest access. Each plan includes weekly usage, and extra usage is billed at token cost.

Grok Bot is included with eligible Premium, Super, Ultra, Teams and Enterprise plans, and is generally available on Teams and Enterprise. Teams costs $30 per user a month, with central billing and admin, team-wide privacy settings and single sign-on.

Enterprise is priced through the sales team. It adds audit logs, service accounts, spend controls, and access and network controls for Grok Bot.

Grok Bot is generally available on business plans today. Its FAQ lists network controls an Enterprise admin sets when the computer starts, and the Enterprise plan lists audit logs and service accounts.

The shared computer is the design choice to understand first. A finance Bot and a sales Bot on one account share files, a browser and logins. Separating those duties means giving each one its own user, because isolation is per user.

In a group thread, Bots act without approval at each step. A business that wants a person in the loop would rely on Auto Review for sensitive actions. It needs to know which actions Auto Review covers.

By our arithmetic, Super costs about $28.57 a month for each Premium-sized share of usage, and Ultra about $16.67, against $30 on Premium. Usage past the weekly allowance is billed at token cost, so the monthly bill can move with how hard the Bots work.