Muse Hype vs. Checkout Reality: The Real State of Agentic Commerce
On Meta's third-quarter 2026 earnings call, Mark Zuckerberg presented numbers that looked like the prelude to a revolution. More than 10 million people in the United States and Canada had activated Muse, Meta's AI assistant, and over a quarter of them used the agent for shopping. The average purchase size hovered around $200. Wall Street applauded.
Six weeks later, the story told by executives at roughly 50 companies—from hospitality to telecom, food to self-storage—sounds quite different. The actual impact of Muse on these companies' revenues is, at best, negligible. This isn't a matter of a small adjustment or a tiny fraction: it's the difference between a product still being built and one that is already selling.
What Meta Says
Zuckerberg didn't hide the numbers on the call. He cited internal Plaid data—the payments company that connects wallets and bank accounts—to support the claim that Muse's connector had recorded "unprecedented" levels of activity across any service before. According to Zuckerberg, the top three categories of Muse use were, in order, shopping, personal productivity, and creativity. Commerce was, by far, the most frequent activity.
Meta announced partnerships with Walmart, Best Buy, Sephora, Wayfair, and Gap, along with integrations with Shopify (Shop Pay), Stripe (Link), and PayPal so that Muse could complete purchases inside its own environment. The ambition is striking: the agent doesn't just suggest—it browses, compares prices, and closes the transaction without human intervention.
Zuckerberg also revealed that Meta plans to charge companies a fee for transactions facilitated by Muse, and that the intention is for this revenue to eventually surpass that of the Family of Apps—the collection of Facebook, Instagram, WhatsApp, Threads, and Messenger.
What Companies Are Actually Seeing
The contrast emerges when you ask who is actually receiving those orders. Wired spoke with executives at roughly 50 companies across different sectors. The near-unanimous answer was: nothing unusual.
DoorDash provided the most concrete data point: CEO Tony Xu revealed on the same earnings call that the platform had received "several hundred" total orders from third-party agents—Muse, Instinct, and Fo—since the capability launched. Hundreds against one billion orders in the quarter. The volume is an order-of-magnitude mismatch. Xu described agent-based ordering as "a very low use case" at this point, but noted that "you never want to fight gravity."
T-Mobile said it had seen nothing significant in travel bookings. Hilton called agent-originated bookings "insignificant." eBay described the activity as "very, very small." None of these companies announced plans to redesign their website experience to accommodate agents.
Chipotle is an interesting exception—not because it is seeing high volume, but because it is encountering problems. Muse agents stumbled over loyalty discounts and after-hours orders. The company is now cleaning up its website code so agents can navigate more smoothly. Even so, Chipotle's CEO admitted that Muse's debut raised agent orders by only a single-digit percentage.
The Data Tells a Different Story
Numbers from Forter, the fraud-prevention firm, suggest the growth is real—but concentrated in a handful of retailers. From July to October, agent-placed orders more than quadrupled on Forter's network, which covers more than 2 billion consumers and roughly half a trillion dollars in annual retail transactions. The week Muse appeared (September 21) recorded a 229% month-over-month increase in agentic orders.
But the most revealing data point is that 70% of agents still "default to scraping"—when there is no direct integration, the agent scrapes the retailer's web interface, a process 10 times slower and 5 times more token-intensive than a protocol-based integration. Most retailers—fewer than 11% of small merchants, per Forter data—are simply unprepared for agentic traffic. It's as if retail is opting in by default, without knowing.
OpenAI's ChatGPT Agent and xAI's Grok Bot are also competing for the same space. ChatGPT Agent now accounts for 15% of OpenAI's agentic orders—up from 1%—and nearly a quarter of its orders come from outside the United States. Grok Bot began recording orders in September and already accounts for a significant share.
The Structural Problem
What separates Zuckerberg's narrative from operational reality is an infrastructure problem. Muse, Instinct, and Grok Bot can operate in two modes: using dedicated checkout protocols (which require explicit retailer integration) or scraping the web interface (which works on any site, but with significantly lower efficiency). Protocols offer faster experience, higher conversion rates, and lower fraud risk—but they only work where the retailer has activated them. Muse has connectors with Shopify, Stripe, PayPal, Walmart, and other major names. But the vast majority of retailers, especially small ones, are not participating.
Amazon blocks personal agents entirely. Walmart, meanwhile, has launched the Agent Gateway program and adopted the Model Context Protocol to facilitate agent access. This divide between "blocking" and "welcoming" will determine which companies capture value and which lose market share in the next retail cycle.
Where Agentic Commerce Is Heading
The current landscape resembles the early days of mobile commerce: adoption metrics are impressive, but the universe of active users remains a small fraction of the total, and retail infrastructure is largely outdated. The difference is that, for mobile, companies knew what to look for—the screen was smaller, behavior was predictable. With agents, the interaction pattern is radically different: no scroll, no banner, no shopping cart. The agent decides, compares, and purchases in milliseconds.
Meta is building the infrastructure. But until companies adapt their checkout experience for agents—or until agents become good enough to scrape pages without friction—agentic commerce will continue to be what it looks like today: a product Wall Street believes will dominate the world, while retailers who haven't prepared for it keep operating as if nothing is changing.
The lesson the executives Wired gathered isn't that agents won't transform retail. It's that transformation happens in cycles—and the current cycle is about adaptation, not scale adoption. The question remains: when Chipotle has cleaned up its code and T-Mobile still hasn't decided whether to accept agents, who will pay the price of inaction?
---
Article based on sources from Wired, TechCrunch, CNBC, and Forter data on agentic commerce.
Sources: WIRED, TechCrunch, CNBC
✓ Independent sources cross-checked and verified before publishing
