Agentic AI starts to crack Booking and Expedia's business (Airbnb is holding up better)

By Investing.com · October 4, 2026.

Bernstein analysts have put numbers on something the industry has sensed for months: artificial intelligence agents that search, compare prices and complete travel bookings —what is already known as "agentic commerce", autonomous software that acts end to end on the user's behalf— are beginning to strain the business model of the big online travel agencies. Booking Holdings (owner of Booking.com) and Expedia face, according to their analysis, greater long-term risk than Airbnb.

The immediate blow is small. Much of the demand these agents generate today still ends up channeled through the agencies themselves, which for them means almost "free" bookings: without the marketing cost they normally pay to win that same customer. But the report identifies three risks that build up over the medium term: disintermediation (the agent searches hotels directly and dispenses with the middleman), pressure on commissions, and far more aggressive price competition.

The most uncomfortable figure for Booking and Expedia: their brands offered the cheapest hotel rate in only around 15% of searches made in the United States in 2026, according to Bernstein. Smaller travel agencies hit the lowest price 77% of the time. An AI agent genuinely working for the user's wallet —and not for whatever commercial deal is in play— has ample reason to bypass the market leader.

That is the underlying problem: part of Booking's business depends on steering customers toward the inventory that pays it the best commission, not necessarily toward the cheapest or the best for the traveler. Bernstein estimates that practice accounts for around 10% of the company's revenue. It is exactly the kind of friction that an autonomous agent, programmed to find the best deal rather than to fill the house's basket, is designed to eliminate.

Airbnb, by contrast, comes off better for structural reasons rather than technological merit: around 70% of its nine million active listings are exclusive to the platform, so an agent cannot simply "find the same thing" somewhere else. It receives close to 45% of its traffic directly, which reduces its dependence on paid search. And 90% of guests message their host after booking, an ongoing bond no price comparison engine replicates.

There is still a technological cushion for the big OTAs (online travel agencies): searching for a hotel remains a complex task —maps, amenities, multiple filters— that today's conversational agents handle worse than a website with good filters, and aggregating a broad inventory in a single source retains value. Bernstein also points to the variable that will decide much of the outcome: who pays the agent. If the user pays directly for an agent that seeks the best deal with no interest other than their own, the risk to the agencies is at its highest. If the agent is funded by advertising —as much of internet search is today— it is far easier for it to reproduce the same incentives, and the same biases, as the current system.

We have been pointing out for months that the next big fight over AI will not be waged only in the labs that train models, but in the layer of agents that acts on the user's behalf in front of every commercial platform. This report is a miniature case study of that phenomenon: any business that lives off information asymmetry —showing a price without showing the best one, charging a commission to place one product ahead of another— is structurally vulnerable as soon as an agent exists that works solely for whoever is using it. And who funds that agent, the user or an advertiser, determines whether the change benefits the consumer or simply shifts the toll to another pocket.

This fits something we have already seen sector by sector: AI does not eliminate tourism or travel, it eliminates the value of purely administrative intermediation —comparing, listing, charging a commission— while what cannot be replicated holds firm: Airbnb's exclusive inventory, the ongoing relationship with the host, human judgment on a complicated booking. In the short term, it is concrete bad news for the jobs and margins tied to the commission model of traditional OTAs —marketing, sales, supplier deal management— and real pressure on their business if the pattern holds over time. In the long term, it is the kind of friction whose disappearance brings more efficient and transparent markets closer: the traveler pays less for the same thing, and the capital and talent that today manage opaque commissions can be redirected toward building the next layer of value. It is a small but representative sample of where the abundance AI promises is pushing: not overnight, and not without losers along the way, but with the direction fairly clear.