Dots, Muse and Grok Bot: what always-on AI agents are

A normal chatbot waits. You type, it answers, and when you close the window it stops. Three companies now sell something different: an AI agent that keeps working after you leave. OpenAI calls its version Dots. Meta calls its AI agent Muse. xAI calls its version Grok Bot.

Here’s what each one is, who can use it today, and what to check before you give one a real job. All three are new. We pulled the details from press coverage and announcements dated 11 August to 4 October 2026, and plans are changing fast, so check each company’s own page before you pay.

The basic idea

An AI agent is an AI that takes steps instead of only giving answers. An always-on AI agent adds one more thing: it runs in the cloud, so it can work while your laptop is shut. You hand it a job such as “watch my inbox for unpaid invoices,” it gets on with it, and it comes back when it finishes or needs a decision.

Your laptop

You close it and walk away.

AI agent, in the cloud

• Watches your inbox
• Finds an unpaid invoice
• Drafts the reminder

Needs your decision: send it?

It’s the difference between asking a friend a question and giving an assistant a standing job. The assistant can only do the job if it can get into your tools, and that access is the risky part.

OpenAI Dots

OpenAI announced Dots at its developer conference on 29 September 2026. Each Dot is an AI agent that runs in OpenAI’s cloud with its own computer and browser. It can use more than 4,000 connected apps, work across several projects, and send you progress updates along the way. You can watch what it does on its computer while it works.

Who can use it: Pro and Business Premium subscribers get one Dot at launch, and Enterprise workspaces can get access with admin approval. You create it in ChatGPT on the desktop or in a browser. After that you can reach it from the mobile apps, Slack or Microsoft Teams.

You can set rules for when a Dot acts alone and when it has to ask first. Start with “ask first” for everything.

For more on the company behind it, read what OpenAI is.

Meta’s Muse

Muse is Meta’s personal AI agent. It works across the apps you connect, such as messages, calendar and files, and it can run tasks in the background. Meta has also launched Muse for Small Business. That version can connect to business accounts like Instagram and Facebook Pages, plus tools such as QuickBooks, Stripe, Canva and Notion.

Who can use it: Muse is free for most uses. Paid tiers are reported at about $20 and $100 a month, with more daily actions at the higher tier. Those prices come from news reports, not from Meta, so confirm them before you pay.

Muse is the only one of the three with a free way in, so it’s the easiest to test.

xAI’s Grok Bot

Grok Bot launched in beta on 11 August 2026. Each Bot runs on a persistent cloud computer with a browser, file access and its own logins. It works through multi-step jobs across your apps and inboxes and comes back when it needs approval. You can also show it a task once and ask it to repeat the steps.

It is a separate product from the Grok chatbot. Using one doesn’t require the other.

Who can use it: at launch it was in beta for SuperGrok Heavy subscribers and some Cursor plans, with enterprise access on a waitlist. The coverage we read says it’s included in those plans at no extra cost.

Two limits are worth knowing. All your Bots share one cloud computer, so there’s no separation between them. And the same coverage says audit trails of what a Bot did are “coming, not shipped.”

How the three compare

	Dots	Muse	Grok Bot
Made by	OpenAI	Meta	xAI
Where it runs	OpenAI’s cloud	Meta’s cloud	xAI’s cloud
Who can use it	Pro and Business Premium plans, Enterprise with approval	Free tier, plus paid tiers	Beta for SuperGrok Heavy and some Cursor plans
Best known for	Thousands of connected apps	Free start, small-business version	Learning a routine you show it
Check first	When it acts alone	What it can see	No audit trail yet

What to check before you give one a job

1. Start with a job where a mistake costs you nothing, such as sorting newsletters.
2. Connect one app at a time. Email, bank and calendar shouldn’t all go on day one.
3. Turn on approvals. Anything that sends a message, spends money or deletes something should wait for your yes.
4. Read what it did, the way you’d check a new hire’s first week.
5. If you handle client records, find out what your industry allows before you connect anything.

AI agents suit some jobs well. Late-payment reminders and review replies are two. Give the AI agent the job, and keep the passwords and the final say yourself.

If you’re a personal trainer, you don’t need an AI agent to start. The free Trainer Admin Audit shows which admin job eats the most of your week, and the guides for personal trainers show how to hand it to a plain chat assistant.

What we don’t know yet

These AI agents will get things wrong. They can misread an email or act on old information, and none of them has a long track record. If an AI agent writes to a customer for you, the message still carries your name.

Treat the first month as a trial. Pick one job, watch what happens, and widen it only if the results hold up. If you’d rather start with a plain chat assistant, which AI should I use? walks through the options.