In 1858, a dry goods merchant named Rowland Macy moved his store to Sixth Avenue and 14th Street and started putting plate glass in the windows. The idea was simple and, in retrospect, foundational: if you could make someone feel something about a product before they walked through the door, you'd already done most of the selling. The entire subsequent history of commerce, from the Sears catalog to the Instagram ad, has been an elaboration on that insight. Arrange things so the buyer desires, doubts, or panics at the right moment. The web perfected it. Every countdown timer, every "only 2 left" nudge, every hero image exists because the person on the other side of the screen has eyes, impulses, and a susceptibility to narrative.
That person is, in a growing share of transactions, being replaced by something that has none of those things.
The signals are already structural. The Universal Commerce Protocol, backed by Google, Shopify, and more than twenty major retailers, defines how an AI agent discovers a product, evaluates fulfillment terms, and completes a purchase. What the merchant publishes amounts to a capability profile: structured attributes, pricing rules, inventory accuracy. Meanwhile, AGENTS.md files are proliferating across websites, declaring machine-readable permissions for autonomous visitors. That these files exist at all is an economic concession: sites are already deciding what to show an entity that can't be charmed, and the answer looks nothing like what they show humans.
A remarkable amount of commerce rests on the assumption of a human buyer. Somewhere between 20% and 40% of e-commerce spending is impulse-driven, purchases triggered by emotion, boredom, a well-timed discount. Applied to a $6 trillion global market, that's one to two trillion dollars annually in transactions that depend on a buyer feeling something at the moment of decision. Global advertising crossed $1 trillion in 2025, three-quarters of it digital, nearly all of it engineered for an audience with a nucleus accumbens. An agent has no reward center to activate.
There's a comforting precedent here. When comparison shopping engines arrived in the early 2000s, researchers found that brand premiums survived full price transparency. But the reason was telling: enough friction persisted that information asymmetry never fully collapsed. Consumers got distracted, didn't find the comparison site, couldn't evaluate quality differences across tabs. Brand premium survived because transparency was incomplete.
An agent, in principle, completes that transparency. It comparison-shops by default, across every source, without fatigue or distraction. The friction that protected brand premium is largely what agents eliminate. Early data suggests the brands surfacing in agent recommendations tend to be the most data-complete, the most machine-legible, the most transparent about what they actually deliver, regardless of recognition. Google search traffic to publishers fell by a third last year. The funnel that turned attention into revenue is narrowing from the top.
One possibility worth sitting with: if the competitive axis shifts from persuasion to structured data completeness, brand migrates rather than disappears. Brand becomes epistemic trust, whether an agent's training data associates you with verified, consistent attributes. The merchants already adapting to UCP are learning this. Agent-discoverable commerce runs on clean, structured data. The competitive moat becomes machine-readable reliability. That's a genuinely different kind of advantage than a beautiful hero image.
Humans still browse, still impulse-buy, still fall for a well-timed ad. None of this vanishes tomorrow. But each transaction an agent handles is one where six centuries of commercial logic, the whole apparatus of display and desire, simply has no purchase. We may be watching the first commerce infrastructure built for a buyer that Rowland Macy's plate glass could never have reached. The glass is as beautiful as ever. Nobody's looking through it.
- UCP's expanding scope: Google's March 2026 update added Cart, Catalog, and Identity Linking capabilities, turning UCP from a single-item checkout spec into something closer to a full commerce protocol for AI agents.
- The trust gap in agent purchasing: A Forrester survey found only about a third of consumers would complete payment through an answer engine, suggesting the shift to agent-mediated commerce still hinges on consumer willingness to delegate, not just technical capability.
- Discovery as unsolved problem: Across every major agent protocol, the hardest structural challenge isn't transaction or negotiation but how agents find and evaluate merchants in the first place, which will shape who benefits from the new commerce layer.
- What hotels learned first: OTAs now handle 51% of travel bookings, and the hotel industry's response offers a preview of how merchants defend margin when an intermediary controls discovery and the buyer never visits your property.