In the 1990s, the average traveler searched about ten times for every booking. By late 2025, Sabre estimated that look-to-book ratioshad reached 1,000:1 and could soon hit 20,000:1 as AI-driven search expanded.
That number is worth sitting with. It describes a buyer that never gets tired, never satisfies its curiosity with "good enough," and evaluates every option it can reach before committing. A buyer with no interest in your logo, your landing page, or the feeling your brand evokes.
Companies are already building for this buyer. MCP servers, A2A Agent Cards, structured tool schemas, machine-readable endpoints are being stood up alongside human-facing products across travel, e-commerce, and SaaS. The default assumption about what follows is commoditization: if an agent optimizes on price, latency, and reliability, every service becomes interchangeable and the cheapest one wins. AI-referred traffic to U.S. retail sites grew 805% year-over-year on Black Friday 2025 while traditional search traffic declined 10%. BCG's analysis of European e-commerce is blunt: brands without real differentiation will be squeezed. The commodity thesis has teeth.
The commodity thesis also has a history, and the history suggests something more complicated. When comparison shopping engines first intermediated airline bookings, airlines didn't simply race to the bottom on price. They introduced à la carte pricing, unbundling fare components into custom configurations that a peer-reviewed study identified as deliberate decommoditization: creating configuration complexity that the aggregator's structured schema couldn't fully flatten. The intermediary won discovery. The supplier won back margin by making the product harder to reduce to a single sortable field.
Something analogous may be emerging in agent-facing services, though the vocabulary for it barely exists. A stress test of 100 MCP servers found a stark bimodal distribution: the top decile all shipped typed schemas, idempotency support, cancellation handling, and quota tracking. The bottom decile shipped none of these. The difference predicted reliability better than any popularity signal. Nobody chooses a restaurant because its API handles cancellation gracefully. But an agent selecting a service at machine speed, across hundreds of options, would weight that property heavily. A whole new axis of product quality, one with no consumer-facing equivalent, inverting decades of competitive logic. The properties an agent evaluator cares about most are precisely the ones a human buyer never sees and a marketing team never measures.
Where this leaves defensibility is genuinely unclear. A company's moat might live entirely in the quality of its tool schema definitions, in how gracefully its service degrades under load, in whether its structured output includes the provenance metadata an agent needs to trust the result. One MCP-integrated service already reports structured quality scores designed specifically for agent consumption, achieving measurably higher quality ratings than human-only alternatives. That's an outcome guarantee expressed in a format no human customer would ever read. If this is where competitive advantage migrates, then the entire apparatus of brand-building, the visual identity systems, the emotional resonance, the carefully crafted onboarding flows, becomes a sideshow to a machine-readable contract that most product teams don't yet know how to write. Differentiation becomes invisible to everyone except the entity making the purchasing decision.