Six weeks after Meta launched its Muse personal AI agent on September 8, 2026, we finally have real usage data instead of speculation. On Meta's third-quarter earnings call on Wednesday, October 21, Mark Zuckerberg shared the first hard numbers on how people actually use the agent — and the headline is blunt: shopping is Muse's number one task. About 25 percent of users task Muse with shopping, including travel planning, average purchases run around $200, and the payment volumes in week one were so far above expectations that Meta had to ask Stripe to raise its credit limits.
Payments Dive's October 23 report has the details. If agentic commerce has felt hypothetical until now, this is the moment it acquired a receipt. This piece walks through the numbers, why the bank pushback got louder the same week, what it means that Canada is one of only two live markets, and what a small business should actually do about it.
The numbers Zuckerberg actually shared
| Metric | Figure (Oct 21, 2026 earnings call) |
|---|---|
| Muse signups since Sept 8 | About 10 million people |
| Live markets | United States and Canada only |
| Top user task | Shopping, including travel planning |
| Share of users tasking Muse with shopping | About 25% |
| Average purchase size | Around $200 |
| Payment method | Stripe's Link digital wallet |
| First-week surprise | Stripe credit limits raised after volumes exceeded expectations |
Two data points deserve emphasis. First, the $200 average basket: these are not token test transactions. Users are letting the agent buy things that matter, at price points where a mistake is felt. Second, the Stripe credit-limit fix: Meta engineers expected meaningful volume, and real demand still outran their ceiling in the first week. Early adopter enthusiasm is a bias, obviously, but demand outrunning infrastructure is a genuinely different signal from "people tried it once."
Beyond shopping, the next most common uses are what you would expect from an assistant with access to your inbox and calendar: increasing personal productivity, managing email, keeping schedules straight. The mundane tasks are the retention engine; the shopping is where the money is.
Stripe saw the same curve
The payments data corroborates Meta's story from the other side of the rail. Stripe president John Collison, speaking on his podcast and in conversation with Zuckerberg, said Stripe's agentic commerce activity "started going vertical" in September — the same month Muse launched. As Payments Dive notes, Stripe's CEO Patrick Collison sits on Meta's board, a relationship worth knowing about when weighing how independently these two data points stack up. Still, Stripe's transaction curves bending upward at the same time as an agent's signups is the kind of convergent evidence that makes a trend hard to dismiss.
The counterweight: "almost none of the customer protections are in place"
The same week Muse's numbers landed, the pushback found its clearest voice. Cameron Fowler, CEO of Early Warning Services — the consortium owned by seven of the largest U.S. banks, several of which have blocked agents like Muse from their customers' accounts — told Payments Dive in an interview published October 23, speaking amid Money20/20 in Las Vegas:
"Turning that engine onto spending our money for us is just such a wildly different proposition, and almost none of the customer protections that you would expect to see are in place."
Fowler is a former banking executive at Montreal-based BMO, a detail Canadian readers should not miss: the loudest caution about agentic commerce in North America is coming from someone who spent his career inside Canadian banking leadership. JPMorgan Chase was among the first banks to block agent access, with CEO Jamie Dimon discussing the issue on the bank's own earnings call the prior week.
So the picture in late October is a real tension, not a marketing story: millions of consumers happily delegating $200 purchases to an agent, while the institutions holding the money say the liability, dispute and fraud frameworks for delegated spending do not exist yet. Both things are true at once, and both matter to anyone selling online. We covered the bank lockout side in our piece on Citi, Bank of America and JPMorgan blocking agents; the earnings call is the first systematic look at the demand side.
Why only the U.S. and Canada — and why Canada should pay attention
Zuckerberg was candid that Muse is limited to two markets because the computing infrastructure remains too cost-prohibitive for a wider release: "If we actually did roll it out more broadly, then you would expect that number to be significantly larger. But we have to go do the work to bring the efficiency in line to make that make sense."
For Canadian businesses, this accident of geography is an opportunity. Canadian merchants, retailers and service businesses are among the first in the world to face agent-mediated shoppers — real consumers letting an AI read their catalog, compare options and complete a purchase with a payment wallet. That also makes Canadian businesses early test cases for the trust questions Fowler raised: what happens when an agent buy goes wrong, who fields the dispute, and what your terms of sale need to say about delegated purchases.
Meta's monetization timeline reinforces the point. Zuckerberg said Meta aims to make money from Muse sooner than its "normal playbook" for new products, driven by compute costs and the size of the personal-agent opportunity. The Muse network he described — consumers' agents and businesses' own agents interacting — is a commerce channel with every incentive to scale fast.
The climb in context
The shape of the adoption curve matters more than any single number. Ten million signups in six weeks, a quarter of users shopping through it, $200 average baskets and a payments rail that "went vertical" the same month are the classic markers of a channel moving from experiment to infrastructure. Put the rest of the month beside it: banks publicly tightened agent access all through October, Walmart opened its Agent Gateway to approved agents, Shopify opened its catalog to personal agents with a human sign-off, and Anthropic's Claude Duet is previewing similar purchase capabilities for small-business users, while open source agents like OpenClaw widen the funnel further.
When a new commerce channel forms, the businesses that get found early win compounding visibility — the familiar source effect applies to agents as much as to answer engines. An agent can only buy from what it can read.
What small businesses should do this quarter
You do not need to build anything exotic. You need to be legible to agents and defensible with humans:
- Fix your product data. Agents read structure: clean product feeds, explicit prices, availability, shipping windows and return policies — plain, unambiguous, current.
- Make key pages reachable by AI search. If your catalog or policies are locked inside interfaces an agent cannot parse, you are effectively sold out.
- Decide your delegation line. Where does an agent-initiated purchase require human confirmation? Say it in your terms before your first dispute, not after.
- Instrument for agent traffic. Watch server logs and AI crawler activity, and tag what you can — October's ChatGPT attribution bug showed how easily this traffic hides.
- Prepare the trust paper trail. Itemized confirmation emails, clear receipts and a one-click human escalation path protect both your customer and you.
Conclusion
Muse's first quarter of data settles one argument and sharpens another. Settled: consumers will hand real, meaningful purchases to a personal AI agent — ten million signed up in six weeks, a quarter of them shop through it, average baskets run $200, and Stripe's limits had to be raised in week one. Sharpened: the infrastructure of trust has not kept pace — the consortium behind the rails says customer protections for delegated spending are almost entirely absent, and the money moved this month while the protections did not. For businesses, the window is open but uneven: agents can already read and buy, while disputes, refunds and liability remain unwritten. Canada sits in the first wave by accident of geography, which turns preparation into competitive advantage rather than a theoretical exercise. If you want help making your store and your content legible to this new class of buyer — product data, AI crawler access, policy language — Vivid can help. Contact us and we will take a look at where you stand.
