Meta's Muse: an AI that makes phone calls for you. 40 days, 10M installs in the US

Muse is Meta's free 'personal AI agent', released in Sep 8: you describe a task in the chat box, and it opens a browser, fills in forms, and makes calls to get it done itself. It took off so fast for two reasons. Half of it is that the product gave ordinary people their first AI that handles tasks on its own. The other half is that Meta put the whole company's traffic behind it. Below: where it gained traction, where the money comes from, and what a small team can take from it and what it can't.

A call from a stranger

In mid-September 2026, New York Times reporter Eli Tan got a call from an unknown number in San Francisco. On the other end was an AI voice: 'I've reached Anthem Dental and will connect you right away.' Ten minutes earlier, the AI had called his dental insurer for him, given his member ID, navigated the phone menu, and waited in the queue. That AI was Muse. His report, published 9-22, was titled 'I Handed My Life to Meta's AI Agent, and It Stunned Me'.

That same week, X was full of similar posts. Product blogger Peter Yang said that in 9-18, Muse called his cable and phone carriers to negotiate, saving him over 800 dollars a year. Another user uploaded their car insurance policy and asked for the same coverage cheaper. Five minutes later, they said, Muse had found a policy $3,500 a year cheaper, bought it, and canceled the old one. These savings figures are self-reported and unverified, but posts like these drove most of Muse's organic growth in its first two weeks.

What it is

An AI agent you use like messaging. You type in the Muse app, or message it in WhatsApp like a contact. It reads your email, manages your calendar, books restaurants, compares prices and orders, watches for price drops, and follows up on tasks. For anything that takes time, it keeps working after you close the app, and it comes back to you when there's progress or when it needs your approval (before sending an email or paying).

The examples in the official announcement are all everyday tasks: turning a recipe video you saved on Instagram into a shopping list, remembering which friend is vegetarian when you plan a dinner, getting a better price for something you're selling, and cutting down a bill.

To address 'will people really hand their accounts to Meta?', it made safety a selling point. Each user gets a separate cloud virtual machine. A separate approval agent sits alongside, and nothing Muse does reaches the internet without its sign-off. Muse can't see your passwords or card number. Payments go through Stripe's Link, which creates a one-time card for each purchase. It's also the first AI agent covered by Link's purchase protection. Conversation data stays out of Meta's ad system.

It's not all-powerful. Someone compiled hundreds of user cases: some carriers treat it as a bot and block it at the door, verification codes, e-signatures and bank accounts still require the person themselves, and Amazon reportedly banned it outright. Gizmodo 10-12 used it to book a vacation flight, and the headline was "Didn't Get Very Far."

How fast it grew

• Google Play: 11-06 installs, 9M; about 7M new in the last 1.61 days. Rating 145k reviews, steadily adding about 3,000 a day. Ranked #1 on the overall free chart in the US, Canada, and Singapore, and #1 on the productivity chart in Japan.
• App Store: 460k reviews in the US, about 69k added in 7 days. Ranked #2 on the US free chart; #1 (11-06) on the free charts in Japan and Canada.
• Sensor Tower(9to5Mac 10-20 引用):美国上线 40 天下载 1,000 万次,是 2015 年有统计以来第 12 个做到的 App,上一个是 Meta 自家的 Threads。同期 ChatGPT 是 370 万,Grok 50 万,Claude 20 万。
• Retention: 7-day retention at 30%, 14-day at 23%. ChatGPT, Claude, and Grok averaged 19% and 16% over the same window (same data source).

It's currently only live in four countries: the US (9-08), Canada, Japan (10-23), and Singapore (10-28). With just those four, it added 1.61M on Play in a week, three-quarters of Claude's global weekly adds (2.12M) and more than half of Meta AI's (3.05M). ChatGPT added 13.8M globally in a week, over eight times Muse's number (both figures from the near 7 days of 11-06).

Why now

Early 2026, the open-source OpenClaw showed the whole tech world what an AI with its own computer that does work for you looks like, but you had to self-host it. Big companies followed, all at premium tiers: Google's Gemini Spark (May) is only for AI Ultra subscribers, xAI's Grok Bot (August) is about $120 a month and aimed at work, and OpenAI's Dots came three weeks after Muse and was initially limited to Pro and enterprise users.

Muse is the first one that's free, open to all adults, and needs no learning curve. The concept has already been explained to everyone, yet ordinary people still couldn't use it. Muse is catching that group.

Where users come from

To be clear up front: most of its reach comes from advantages only Meta has.

1. Meta's own ad slots across its apps. According to TechCrunch 9-25, citing Sensor Tower, Muse started running its own promotion on Facebook and Instagram the day after launch, and within 10 days it had taken the largest share of Meta's in-app promotion slots, ahead of Meta's own promotions for FB, IG and WhatsApp. The creative approach is worth noting: it doesn't pitch how powerful the model is, but shows one concrete thing it can do for you, based on what Meta already knows about you.
2. Paid acquisition off-platform. Mobile ad networks, Reddit, TikTok, and YouTube are all running it. 9-22 Muse entered the top ten US advertisers by spend. October, Muse accounted for 39% of Meta's US ad spend (23% in September), and its ad impression rank among US brands rose from 73 to 21.
3. Usable right inside WhatsApp. It sits directly in the chat app billions open every day.
4. Launch events and Zuckerberg himself. 9-17 Zuckerberg announced a Mac version; 9-23 at Connect he unveiled video calls, Mac computer control, a dedicated email address, and glasses integration. Daily active users rose 27% the day after the event. Meta also hired creators to make collab videos, like a travel blogger who used it to plan a full family day in Chicago.
5. Word of mouth. Posts along the lines of "it saved me $X" were widely reshared on X and Threads. From 9-08 to 9-19, ads made up only 6% of Muse's impressions; in the first two weeks, organic traffic did most of the work.
6. Going country by country. Each new country gets another push to reach its local free chart, plus local service integrations. Singapore, for example, got Shopee and Grab.

Where the money comes from

Most features are free, and heavy users buy subscriptions by usage: the Power tier is $20 a month, the Maximum tier is $100 a month, and the difference is how much compute you get per week. In-app purchases on Play range from $0.99 to $1,049.

Meta hasn't disclosed revenue. The indirect signal is the grossing charts: 11-06 it ranks #12 in Canada, #17 in the US, and #18 in Japan on Google Play's Productivity top-grossing chart. People are paying, but the scale is unknown.

Who it's competing with

Its real rivals are the other three big companies. CNBC's headline for 9-30 read "OpenAI jumps into the hot personal-agent market after Meta." On Play there are also a bunch of wrapper apps with Agent or Muse in their names riding the keywords. They aren't real competitors.

What a small team can take

You can't just build a Muse and take its place. It rests on in-house models, a cloud VM per user (a serious compute bill), Meta's own ad slots, the WhatsApp entry point, and data from Instagram. Doing it in another country won't work either: Meta is rolling out country by country on its own, with OpenAI and Google behind it.

Two things you can take.

First, the playbook, which applies to any "AI that does things for you" product:

• Live inside a chat app users already have, so nobody has to learn a new interface;
• In ads, show only one concrete thing it did for you, ideally with a dollar figure ("cut $180 off my electric bill this year"). That sells far better than talking about model capability;
• Build a habit for free first, then charge by usage tiers;
• Make safety and payment protection a selling point (one-time cards, asking your approval before paying, every action auditable), directly addressing "can I trust it with my account?"

Second, build a business around it. When a new platform takes off, users' first questions are "what can it do, and how do I get it to do it?" People are already doing this: musecase.dev has collected 1,216 real use cases, its homepage sells sponsor slots at $5 a week; botdidit, musetaskhub, and agentusecases turn use cases into copy-paste prompts; someone posts a daily how-to on "making Muse do X"; and others sell the part it can't do, like writing the email it gets stuck on. These businesses are easy to start but small, there are already five or six of them, and each time Muse ships a new feature they have to rewrite.

What we don't know yet

• Conversion rates and revenue haven't been disclosed. Whether people stick around once the novelty fades needs a few more months to judge.
• In October, security researchers found that shared conversations leak personal and company-sensitive information, and Meta paused the sharing feature (Ars Technica). Its impact on word of mouth isn't clear yet.
• Recent Play reviews are mostly vague five-star ones ("most useful app ever," "I love my juno"). How many were solicited, I can't tell.
• Download estimates vary: TechCrunch's 9-25 piece cites Sensor Tower at about 10M in the first 2.8 days, Apptopia at 4.3M cumulative at the time, and Appfigures at about 2.3M. None specify global versus US. Another report puts US downloads at 730K in the first 10 days.

If it were you

In one line: C. It's a product you can learn a lot from but can't copy. Meta is pushing it with its whole company's distribution, and every major outlet has already covered it.

Who it's for: teams building "AI that does things for people" who want to borrow the playbook: chat entry point, single-concrete-outcome ads, usage-based pricing, safety as a selling point. Also individual writers and SEO people who can build a use-case library, a tutorial site, or prompt templates around Muse.

Who it's not for: small teams trying to build a general consumer personal agent head-on. Compute cost is a VM per user, and acquisition means competing with Meta's own ad slots as well as OpenAI and Google. You lose on both fronts.

How to do it: if you want a vertical product, pick one scenario and go deeper on one task than Muse can. Best bets are where users complain in posts that "it got stuck": the part that needs a signature, a verification code, or gets blocked by merchants. Or regions Muse hasn't reached, or scenarios that need local services, like local carriers or local insurance. If you want to build around it, make the most practical "get Muse to do X" prompt library, with SEO landing pages by scenario (cancel bills, switch insurance, book reservations).

How to get users: copy Muse's own approach. Post real screen recordings of "it got this done and saved me $X" on X, Threads, or TikTok. For use-case sites, go after long-tail searches like "Muse how to + [task]."

When to move: for businesses built around it, the window is probably a few months, before it rolls out globally and before the official side builds its own use-case hub.

Signs you've read it wrong: (1) Muse's 7-day retention falls to ChatGPT's level (below 20%), or it drops out of the US free top ten. That would mean ads are what's pushing it up, and its playbook isn't worth copying. (2) Meta opens third-party plugins or a developer platform. Then "building around it" goes from writing tutorials to building plugins, a bigger opportunity, and the rating should go up.