Your customer just hired an assistant that never clocks out.

Picture a member of your studio on a Sunday night. She has not opened your app in two weeks. She has not read your newsletter. But on Monday morning her phone shows her a short note: "Your class pack expires Friday. You usually go Tuesdays at 6am. Want me to book it?"

You did not send that note. Her assistant did. It read the receipt in her inbox, remembered when she likes to work out, and noticed the deadline before she did.

Nothing in that scene is far-fetched anymore. Between September 8 and October 21, four of the largest AI companies each shipped what they call a personal agent. This series is about what that means for a gym, a studio, a med spa or a salon: who reads your website, your email and your ads now, what persuades a piece of software that already knows what its person wants, and what to do about it.

We covered the plumbing side in an earlier series, starting with Your next customer won't visit your website. This one is about the person side: memory, the inbox, and an assistant that acts before it is asked.

Four agents in seven weeks

Here is what shipped, from the companies' own pages where we could open them and from reporting where we could not.

Meta Muse (September 8). Meta calls it "a personal AI agent." People message it in the Muse app or directly in WhatsApp, and it works on its own dedicated computer in the cloud, so it "keeps working after people close the app." Meta says Muse "remembers what matters to a person, so it can make suggestions unprompted," and that people can tell it to forget things. People choose whether it can only read their email or also send on their behalf, and it checks with them before sending an email or making a purchase. It launched in the US on iOS, Android and the web, free for most use with paid plans for heavier use. On October 23 Meta brought it to Japan, which Meta calls its first market outside North America.

OpenAI dots (September 29). Announced at OpenAI's DevDay, dots are always-on agents inside ChatGPT, each with its own cloud computer the owner can open to watch it work. You give a dot a goal, and it keeps going while you do other things. When nobody has given it a task, it does what OpenAI calls "proactive research," looking through connected apps for ways to help, with read-only access so it cannot send or change anything in that mode. As reported by SiliconANGLE, it is rolling out to ChatGPT Pro and Business Premium subscribers, except Pro users in the European Economic Area, Switzerland and the UK, with an admin-enabled beta for Enterprise plans. Engadget notes Pro plans start at $100 a month.

Google's Gemini agent (October 8). Google Cloud introduced it at its Gemini at Work event as "a universal agent for work." It runs inside Gmail, Calendar, Docs and the rest of Workspace, keeps "a single set of memories" across devices, and keeps working "after you close your laptop." It can even become a coworker with its own email address and calendar. TechCrunch reports Google is launching it for businesses first and bringing it to consumers later, with no consumer date given. Google's own post does not state a release stage, so we treat it as early.

Claude Duet (October 21). Anthropic's entry is a research preview, by waitlist, for US subscribers on the Pro and Max plans; it is not on Team or Enterprise plans yet. It has its own iOS and Android app and runs on claude.ai. Each morning it sends "a short brief on the most important items across your connected apps," checks those accounts every few hours, and carries over the memory from the person's Claude account. Anthropic's help page lists Gmail and Google Calendar among its connectors and says Microsoft 365 is not available yet. Before it sends a message, makes a payment, signs a document or deletes anything, it "shows you exactly what it's about to do and waits for your approval."

"Muse is a personal AI agent. It doesn't just answer questions, it actually does the work."

What "personal" means

Search engines, chatbots and the shopping agents we wrote about earlier all wait to be asked, and these four do not. Read the launch pages side by side and the same three ideas keep coming back.

1. 1RemembersHer schedule, budget, favorite class time and what she said once in passing
2. 2WatchesChecks her inbox, calendar and apps every few hours, in the background
3. 3BriefsSends a short morning note on what needs her attention
4. 4AsksShows the action it wants to take and waits for a yes
5. 5ActsBooks, replies, buys or reschedules on its own computer

Memory. Every one of them keeps a running picture of its person. Meta says Muse learns from conversations and gets "sharper along the way." Google lists four kinds of memory, including a record of everything the agent has done before. Duet starts with whatever the person's Claude account already knows.

Background. Each runs on a computer in the cloud, not on the phone. The person can close the app and the work continues.

Action. They book, draft, fill in forms and pay, with an approval step before anything consequential.

Put together, that is a customer represented all day by software that knows their preferences better than your front desk does.

How many people have one

It is early, and the numbers are uneven. Muse is the only one that is free and open to the public in the US, and it has a head start.

Source: Sensor Tower via VentureBeat (Muse); SiliconANGLE (dots); VentureBeat (Duet)

Two caveats. First, dots and Duet sit behind paid plans and waitlists, and Google's agent is starting with businesses, so most of your customers do not have one yet. Second, the companies themselves warn about mistakes. OpenAI tells users agents "can still make mistakes." Anthropic says the Duet preview may misread a conversation or misjudge what is urgent. Meta says plainly that "prompt injection remains unsolved," meaning a page or message can try to trick the agent.

There is also a privacy fight starting. TIME reported on October 6 that Muse keeps detailed, regularly updated profiles of its users and the people they mention. Meta told TIME that Muse data is not shared with its advertising system.

The series map

Over ten parts we will take the consequences one at a time.

1. Your customer just hired an assistant: what shipped, and what personal means. You are here.
2. The agent already knows what they want: memory, and why marketing moves from persuading a stranger to matching a known brief.
3. Your email gets read by an agent first: what happens to promos, newsletters and win-back emails when a summary comes first.
4. Your website has a second reader: the website as a spec sheet.
5. Agents shop before your customer asks: retention when the renewal reminder comes with two cheaper options.
6. Ads for an audience that delegates: what paid media is for when an assistant builds the shortlist.
7. What an agent counts as proof: reviews and listings, read all at once.
8. When an agent calls the front desk: your front desk now answers software too.
9. Measuring marketing when agents shop: attribution when the shopping happened inside an assistant.
10. How we are getting clients ready for personal agents: our own work, and what is live.

Next, in Part 2, we look at memory: what it means when a customer's agent already carries a brief like "6am, near the office, no long contract," and what your business has to say plainly to match it. If you want to see how an agent's browser handles your site today, The browser is becoming the assistant covers that side.

---

Sources

• Meta, "Introducing Muse: The World's First Personal AI Agent Built for Everyone," September 8, 2026, updated September 30, 2026. about.fb.com
• Meta, Japan launch announcement for Muse (Japanese), October 23, 2026. about.fb.com
• The Rundown AI, "Meta introduces Muse as personal AI agents face growing competition," September 9, 2026. therundown.ai
• SiliconANGLE, Duncan Riley, "OpenAI launches Dots, always-on AI agents in ChatGPT with their own cloud computers," September 29, 2026. siliconangle.com
• Google Cloud, Thomas Kurian, "Welcome to Gemini at Work 2026," October 8, 2026. cloud.google.com
• Google, "Google Cloud introduces the Gemini agent," October 8, 2026. blog.google
• TechCrunch, Sarah Perez, "Google brings agentic AI to Gemini, starting with businesses," October 8, 2026. techcrunch.com
• Anthropic, "Claude Duet," October 21, 2026. claude.com
• Anthropic Help Center, "Get started with Claude Duet." support.claude.com
• VentureBeat, Carl Franzen, "Anthropic's answer to Muse is here: Claude Duet, a mobile and desktop AI agent that takes actions proactively," October 21, 2026. venturebeat.com
• Engadget, Igor Bonifacic, "Claude Duet is Anthropic's answer to Meta Muse and OpenAI Dots," October 21, 2026. engadget.com
• TIME, Harry Booth, "Meta's Muse AI Agent Is Building a Dossier On You," October 6, 2026. time.com
• Drag, Nick Timms, "Claude Duet vs OpenAI dots vs Meta Muse vs Gemini agent: AI Agents and Email (2026)," October 22, 2026. dragapp.com