On 8 September Meta introduced a personal AI agent called Muse. It is available in a dedicated app, on the web, and directly inside WhatsApp.
Most coverage treated it as another assistant. But Meta's own announcement contains two details that make it a business announcement, and they are worth noticing precisely because they are written quietly.
Errors or inaccuracies are possible. Spotted something that looks wrong? Write to me and I will correct it.
The First Detail: the Agent Pays
This is not an agent that recommends. It is an agent that closes the transaction.
Per the announcement, Muse "can open a browser, fill out forms, and negotiate" on the user's behalf. It books travel, sends emails, and reaches the payment step.
And at the payment step Meta did not build its own infrastructure - it wired itself to Stripe.
| The payment mechanism | |
|---|---|
| The instrument | Stripe's Link |
| The protection | Meta notes this is the first agent covered by Link's purchase protections |
| What that includes | cover for damaged or lost items, price drops, no-fee returns |
| The security | a wallet generating a one-time-use card, so real card details stay hidden |
| Coming | Shop Pay as an additional method, and 1Password for existing logins |
That is a list of commerce partnerships, not of features. Anyone building an assistant does not need a wallet with a one-time card and purchase protection. Anyone building buying infrastructure does.
The Second Detail: What Meta Takes Out of the Picture
And here comes the line in the announcement that surprised me most.
Meta writes explicitly: "Muse doesn't share a person's conversations or the data in their VM with Meta's ad systems."
To see why that is unusual, remember what Meta is. Its revenue comes almost entirely from advertising. An agent that knows what a user is searching for, what they are deliberating over, and what price they eventually accepted - is perhaps the most precise advertising data source ever built.
And Meta writes that it is not touching it.
There are two possible explanations, and they do not conflict:
One is regulatory. An agent that reads your email and holds your passwords, at a company with Meta's regulatory history, could not have launched with an open advertising pipe. The isolation is an entry condition, not a concession.
And the second, which is the more interesting one: perhaps that is simply not the product. If the agent earns revenue from transactions - a fee, a subscription, or a share of commerce - then the ad system is not its destination but a different business inside the same company.
Meta writes that the service is "free for most of what people need, with subscription plans for people who want to do more". That is a subscription model, not an advertising model.
And What About the Shares
Almost nothing.
On launch day, Tuesday 8 September, Meta shares closed at $613.48 - down 0.53% from the prior close of $616.77.
(Those are closing prices from the last completed session. As of writing, today's session is still open, so I am not quoting a price from it.)
And that is the thing worth noticing: a company with billions of WhatsApp users launched an agent that makes purchases, with Stripe's payment infrastructure behind it, and the market marked it as a non-event.
What Would Have to Happen for It to Matter
So it is worth stating exactly what is missing. Meta's announcement contains:
- No number. Not how many users, not a target, not a revenue forecast, and not what a subscription costs.
- No detailed revenue model. Subscription plans are mentioned; what they are and what they include is not.
- No figure on the cost of running it. A dedicated virtual machine per user is not a cheap product, and at Meta's scale that is a material question.
- No reference to a share of transactions. It does not say whether Meta receives anything from the purchases the agent makes.
Anyone wanting to price this needs at least one of them, and none exists yet. That is not a criticism of the announcement - this is what a product announcement looks like at this stage - but it is why it cannot be priced.
הזווית שלי
דעה אישית של אילן אברמוב - לא ייעוץ ולא המלצה
What draws me here is not the agent but where it runs.
Meta did not launch a new app that people must be persuaded to download. It put the agent inside WhatsApp - somewhere people already are, already typing, and already messaging businesses. That is an enormous difference in friction, and it is the real asset Meta brings to the table. No competitor in agents has a distribution channel like it.
And what I think gets missed in reading the announcement is the Stripe side. A one-time card, purchase protection, and Shop Pay to follow - that is not a list of user conveniences, it is a list of what you need before anyone will let software buy on their behalf. Whoever built that is thinking about transaction volume.
And even so I am in no hurry to conclude anything, and not because the product is unimpressive. Rather because Meta's history with large product launches is very mixed, and because the announcement contains not one number anyone can track. Infrastructure that buys on your behalf is a big idea; a big idea with no stated revenue model is still only an idea.
And what I will look for is one specific thing: when Meta starts talking about Muse on the earnings call. Not as a mention in the vision paragraph, but as a line with a number attached - users, subscribers, or transaction volume. The moment that happens, it can be priced. Until then, anyone pricing it is pricing a story.
(It is important to stress: this is my personal opinion only, and nothing herein constitutes a recommendation to take any action. Ilan Abramov, the writer, or parties on his behalf may hold positions in the securities mentioned, and those holdings change from time to time.)
