Somewhere in the last few weeks, Meta’s AI shopping assistant Muse became the most talked-about robot customer in corporate America. The company has painted a vivid picture of a future that is already here: a new kind of software, called an agent, that can log onto the internet and take care of real tasks for ordinary people—scanning store inventories, comparing prices, applying coupon codes, checking out, and even handling returns if something goes wrong. Meta says this is not a distant experiment. According to the company, millions of people are already using Muse to shop, and the average order is more than $200. Mark Zuckerberg added on Meta’s earnings call that more than 10 million people in the US and Canada have activated Muse, and more than a quarter of them have used it for shopping, making shopping the most common way people are testing the tool. That pitched investors into a state of unease. If every consumer suddenly has a tireless, ruthless digital assistant that can find the lowest price in seconds, the elaborate machinery of brand loyalty will begin to rust. No longer will shoppers be swayed by pretty packaging, emotional ads, or the comfortable inertia of buying the same product again. They will simply ask a bot for the best deal, and the best bot will win. That could ignite price wars, crush profit margins, and make the old rules of retail irrelevant. It is no wonder that Muse became one of the hottest topics on recent earnings calls. The questions are so new that even the people being asked them don’t have a script, and the fear is so vivid that it has started to move markets before any real damage has been done. Investor anxiety is not entirely irrational. The last few years have shown how quickly AI can move from curiosity to utility, and agents are being positioned as the next big interface for everything digital. If a chatbot can book a vacation, order groceries, and switch insurance plans, the companies that traditionally profited from those interactions could find themselves turned into invisible suppliers. The only name the consumer sees is the agent, and the agent belongs to Meta. That threat is bigger than a discount code; it is a threat to the relationship between brands and the people who buy from them. It explains why so many analysts are paying attention to a tool that, by every available metric, is still barely moving the needle.
But when the numbers from companies actually doing business arrived, the tone changed. In conversations with WIRED and in earnings calls across roughly 50 companies—travel, internet providers, food, self-storage, banking, and more—executives delivered a collective shrug. DoorDash said the number of orders coming from Muse and similar agents is in the hundreds. That is not hundreds of thousands. It is hundreds. T-Mobile told analysts that it hasn’t seen anything of significance from agents. Hilton, a company that could be a natural fit for people using bots to plan trips, was even more direct, calling its agent-generated bookings insignificant. Ally Financial, an online bank, described activity as quite low. eBay, the marketplace that would seem tailor-made for robot bargain hunters, said the numbers are very, very small. If you put these comments side by side with Meta’s claims, the picture gets confusing. Ten million people activated Muse, but the businesses that would presumably receive that spending are not seeing it. Some of that can be explained by timing, since the tool is still new. Some of it might be explained by the way people experiment with agents: they ask for a recommendation, get a handful of options, then open a browser and buy the thing themselves. The agent is a concierge, not a cashier. That is a meaningful distinction for the economy. It means agents are influencing behavior, perhaps even shaping preferences, but they are not yet replacing the final step in a purchase. The old world, at least for now, is still standing. Most executives, in fact, said they expected to benefit if agentic commerce ever becomes mainstream, because agents could deliver new customers who otherwise would never have found them. That optimism is easy to mock, but it is also the most rational response to a technology that has not yet earned a permanent place in the checkout line.
The fear that agents would become robotic unsubscribe machines has also not materialized. One of the first things financial experts expected users to do was ask Muse to cancel unused subscriptions and memberships. Why call a gym, wait on hold, or fiddle with a shoddy account page when a bot can do it in seconds? If that behavior took off, it would threaten the recurring-revenue businesses that have become the backbone of modern corporate profits. But the companies in the crosshairs are reporting an unusual number: zero. Life Time Group, the fitness chain, said it has seen zero cancellations from agents. Public Storage, which rents out millions of units to people who have often forgotten they have them, gave a similar answer. Hartford Insurance Group did as well. Eternal, a delivery app giant in India, estimated agentic commerce at close to zero. This is a remarkable result, given how easy it is to imagine people handing over their least favorite chores to a bot. Maybe consumers do not yet trust an agent with something as final as cancellation. Maybe they prefer to deal with a human representative when changing a contract. Maybe the agents are simply not being asked. Whatever the reason, the nightmare that had investors dreaming of empty gyms and unsubscribing households remains a thought experiment. The only thing being automated, as one might say, is the anxiety. Zero is a strange number in corporate reporting. It is easy to dismiss as anecdotal, but when several different companies in several different industries all report the same zero, it starts to look like a pattern. It suggests that the hypothetical, all-powerful agent that can cancel anything in seconds has not yet won the trust of consumers. That may change as the technology matures, but it is not the kind of change that happens overnight. For now, the percentage of people using agents to prune their monthly bills is so small that it rounds to nothing.
So where are the 10 million people actually spending their money? Mark Zuckerberg says more than a quarter of Muse users have tried shopping, and Meta says the average purchase is above $200. That implies there is a real flow of dollars somewhere. Meta declined to comment beyond the earnings call, but it pointed to partnerships with 10 retailers, including Walmart, Best Buy, and Gap, which are supposedly integrated with Muse to simplify shopping. These are exactly the kinds of companies that could see a noticeable bump in sales from agent-driven orders. But here is the awkward part: most of them reported earnings before Muse launched, or before agentic shopping had enough volume to matter. They have not shared data about it yet. In a market where every small change is scrutinized, silence is meaningful. If Walmart and Best Buy had seen an explosion of high-value orders from AI agents, they would have had every incentive to mention it. The fact that they haven’t says something. It could be that the numbers are too small to bother, or that executives consider them too volatile. It could also be that users of Muse, Instinct, and other AI agents are spending on small-scale purchases that don’t show up in the earnings of public companies—things like food, parking, or impulse items from local shops. But the available evidence from DoorDash, Hilton, and others suggests that even those categories have not seen a clear surge. There is a gap between the confidence of the AI companies and the receipts of the rest of the economy, and for now nobody seems to know exactly how to close it. Perhaps the most sensible reading is that the ten million activations are not the same as ten million weekly users. People download apps, click through a demo, and then forget. If Muse is experiencing that kind of drop-off, its shopping numbers might be impressive in the abstract but too small to move the needle at any single company. That is a problem for Meta’s narrative, but not necessarily for its long-term plans.
Executives are the first to say that agents are at an early stage. Yet the immense public anxiety about AI has left companies searching for lessons from the moment the technology goes live. They cannot afford to wait for a problem to arrive; they have to build for it in advance. That is why some are already trying to optimize their websites for automated shoppers. Chipotle’s chief digital officer, Arlie Sisson, recently told reporters that the restaurant chain watched agents try to complete purchases and fail in ways that human customers rarely do. Some were tripped up by loyalty discounts, unable to connect a rewards account to the total. Others tried to order after the kitchen had closed and got stuck in a loop. When Muse debuted, it increased agentic orders by a single-digit percentage—not a revolution, but enough to attract attention. Sisson’s response was not to wait for AI companies to fix the problem. Chipotle began cleaning up its website code, making it easier for agents to navigate menus, understand prices, and actually finish an order. This is a small but telling shift. Retailers used to build websites for people with eyes and patience. Now they are starting to build them for machines with no visual instincts but a disturbing ability to follow every link. Think of it as search-engine optimization for a different kind of customer. Instead of keywords and backlinks, the currency is clean code, fast load times, and clear layouts. The work is not glamorous. It is the digital equivalent of widening the driveway for a self-driving car that hasn’t quite mastered the entrance yet. But it matters, because the company that is easiest for an agent to shop at is the company an agent will choose next time. And behind every agent, there is always a hungry human waiting.
For investors, the immediate message is one of relief. The agentic shopping apocalypse, at least as described in breathless forecasts, has been postponed. A technology that produces hundreds of orders at DoorDash and zero cancellations at Life Time is not yet a threat to the broader economy. But the longer-term story is not finished. The tools will get better, consumers will get more comfortable, and companies will keep optimizing for the day when agent traffic matters. Meta, for its part, has a clear interest in making Muse a permanent habit rather than a novelty. The platform power that would come from standing between consumers and every retailer is enormous. That alone should keep executives alert. They are watching, tweaking their websites, and waiting for the next quarter’s numbers to tell them whether agents are finally arriving in force. If Walmart, Best Buy, and Gap eventually report that agent-driven sales have jumped, the old fears will resurface. If they echo the same “very, very small” language we are hearing now, the conversation will pivot to the slow, boring timeline of human adoption. For now, the most honest way to describe the agentic shopping era is that it exists, but barely. Robots are browsing. They are even buying a little—a few hundred burritos here, a handful of hotel rooms there. But they are not yet replacing the thing that commerce has always depended on: the human decision to trust a brand, pull out a wallet, and click buy. The revolution may still come, and it may come faster than anyone expects, but it hasn’t arrived yet. And as every marketer knows, if you can’t see the customer’s face, you can’t tell whether they’re a human or a bot—but you can watch your receipt scanner. Right now, the scanner is mostly quiet.
