Meta's Muse launched in September as a consumer assistant—filling in forms, booking travel, theoretically making purchases on your behalf. It passed five million downloads in North America within three weeks and briefly sat above ChatGPT at the top of the US app store. A nice debut. And then, before anyone had finished processing the implications of that, Meta extended it to businesses.

The same agent can now connect to the software a seller actually runs on: Shopify, Stripe, Intuit QuickBooks and Klaviyo sit alongside Asana, Box, Canva, Dropbox, Figma, Notion, Slack and Zoom, together with a company's Facebook and Instagram business accounts and its Meta ad accounts. Custom connectors are supported for anything not yet on the list. Meta's pitch is that an agent given a goal—phrased as broadly as "run my business" or "find new customers"—can act across all of it at once.

Which is a sentence that should make you feel at least two things simultaneously.

For anyone who's watched Meta's advertising playbook evolve over the past decade, this is a pattern so familiar it's almost comforting. Almost. We've gone from "let us target your ads" to "let us make your ads" to "let us choose your audiences" to "actually, just connect your bank account, we've got this." Every year, one more piece of the funnel gets absorbed. The trajectory has all the subtlety of a python eating a rabbit—you can see exactly what's happening, and yet somehow you're still watching.

Muse for Small Business extends that logic past the ad account and into the back office. The company that wanted to run your customer acquisition now proposes to run your storefront admin, your email flows, your books and your scheduling. The framing is time saved—owners are short on hours, not ideas—and that framing is genuinely persuasive because it's genuinely true. Running a small ecommerce operation involves a staggering amount of administrative repetition that nobody got into business to do.

The harder part of the trade is less visible, and it lives on the supply side. An agent wired into a seller's storefront, payment processor, email platform and accounting system can read what a product costs to sell, where the margin sits, which customers repeat and which quietly refund. That's operational intelligence most sellers struggle to assemble for themselves, now sitting inside Meta's infrastructure.

Meta has pre-empted the obvious objection by saying it will not feed the data from the agent's working environment into its advertising systems. That assurance is worth taking seriously, and worth reading precisely. It's a statement about the current configuration of the business product. Not a constitutional amendment. Not a blood oath.

It's tempting to treat the consumer version of Muse as the minor half. The maths seem to invite that reading. One widely cited estimate puts the average US consumer at around 47 payments a month. Strip out automated bills, fuel and physical-shop purchases, and analysts at one equity research firm reckon perhaps eleven are the kind an agent could plausibly handle. That doesn't sound like a revolution. It sounds like it might cover your Uber Eats habit and a replacement phone charger.

But reading that as "the consumer side is small" mistakes the fee for the point. A transaction cut was never the most valuable thing Meta could draw from a consumer agent. The valuable thing is the behaviour. Every goal a person hands the agent—the products they weigh, the comparisons they ask it to run, the prices they balk at, the categories they keep returning to—is intent captured at a resolution that a feed of likes and clicks never reached. That's the raw material of ad targeting, which is the business Meta actually runs.

It's why the same research firm that doubts the fee model expects Muse to become a surface where Meta sells ads. The eleven-payments figure doesn't shrink the opportunity. It explains why the money is more likely to arrive as advertising than as commissions.

Here's where you should sit up slightly straighter in your chair.

The consumer agent gathers demand-side behaviour: what people want, how they decide, what moves them off a purchase. The operator agent gathers the supply side: what products cost to sell, where margins sit, which customers are profitable. Hold the two together and Meta is positioned to see both halves of the same market—what buyers want and what sellers can afford to offer—which is a more durable advantage than a slice of any single transaction.

Call it the last-mile monopoly: the tendency for one platform to end up owning every step from discovery to checkout, so that each step stops being a service a seller buys and becomes rent paid to a single landlord. We've been watching this building for years in Meta's ad stack. Now the landlord is politely asking whether you'd also like help with the plumbing, the electrics and maybe the accounts.

And no equivalent data wall has been described around the consumer agent's behavioural data, which is the half that most naturally belongs in a targeting system. A platform doesn't need to misuse a vantage point for the vantage point to matter.

For Amazon sellers, the most revealing thing about Muse for Small Business is what's absent from it. The connector list is the plumbing of a direct-to-consumer brand. Amazon Seller Central is not there.

Nor is this an oversight that time will quietly fix. Amazon has already asked Meta to remove its store from the consumer side of Muse, saying it had not agreed to be available through the tool. Amazon took Perplexity to court last year over agent-driven purchases on its marketplace. An appeals court lifted a temporary restriction in August, and the dispute continues. Amazon is defending the same perimeter it always defends—and it's building its own agents, Rufus/ Alexa among them, to sit inside that perimeter rather than let anyone else's in.

The practical effect is that an Amazon-first seller sits outside Muse twice over. The operator agent doesn't plug into Seller Central, so the time-saving pitch largely doesn't apply. The consumer agent can't reliably reach the marketplace, so the demand-side story doesn't apply either. Depending on perspective, that's either exclusion or a reprieve. (I suspect most Amazon sellers will file this under "things I didn't need another platform trying to get me to worry about," which is an increasingly thick filing cabinet.)

A brand that operates across both Amazon and its own Shopify store will increasingly be running in two different agentic environments with incompatible access politics: a Meta-wired operating layer on the DTC side and a walled, Amazon-controlled one on the marketplace side. Managing a catalogue, a brand voice and a set of customer relationships consistently across those two worlds is a problem that didn't exist eighteen months ago.

There are guardrails, at least on paper. Meta says nothing publishes, sends or spends without a person's approval. It uses single-use card numbers with new merchants. The product is free up to a usage cap, with paid tiers running from roughly $20 to $100 a month.

The approval gate is the main safeguard on offer, and it's also the main brake on the autonomy the product is selling. The more you insist on approving each action, the closer the agent gets to being a slightly faster assistant rather than an operator. Which may be the right setting for now, but is not the thing the marketing describes.

And the risk isn't purely theoretical. Press reporting in the launch week described the consumer version of Muse handing a marketplace buyer the home address of a seller and arranging a meeting the seller hadn't agreed to. Meta said it was investigating. The detail matters because it marks the line that agents cross: an assistant that generates a wrong answer produces text a person can ignore. An agent that acts on information across live systems produces consequences in the actual world. For a small business weighing whether to let software send messages to its customer list or move money through its payment processor, that distinction is the entire decision.

Meta's Muse is not doing anything conceptually new—it's extending a decade-old playbook into new territory. The company that progressively absorbed your ad creative, your audience targeting and your measurement is now offering to absorb your operations. The pitch is compelling because the administrative load is real. The question is what you're trading for the convenience.

Industry analysts have questioned why a seller would wire its core systems to Meta when established cross-system providers—Microsoft and Google among them—already sit inside many of those workflows. Others point out that smaller operators weigh cost and hours more heavily than data caution, which is precisely the group Muse is aimed at. Both observations can be true simultaneously.

The question each seller has to answer is narrower and more concrete than the headline about agents replacing shopping. It's which parts of your operation are worth handing to a platform in exchange for speed, which customer relationships and data need to stay in your own hands, and what it would cost to unwind the arrangement later if the terms—or the fees—change.

Whether or not the agent that shops for your customers ever buys at scale, it's already learning how they shop. The one asking to run your business is asking for the other half of that picture. Worth reading what it's actually asking for before you hand over the keys.

📅 Howie.ai- Say goodbye to scheduling hassles with advanced AI coordination

📊 Zocket - A generative AI model trained with local and global purchase data for the highest level of marketing performance.

🖥️ Numerous AI - Stop spending time on spreadsheet busywork.

🤖 AgentX - Make specialized AI Agents work together, AgentX is the best place to build AI teams. Cross vendor LLM collaboration.

📹 Top View Ai - Create any video, just tell your agent.

😍It was great, all I needed

🙂Good, not great

🙄It sucked