AI agents now book travel, but where does the real money and control end up in that booking process?

This explores where the money and leverage go in travel booking once AI agents start doing the searching. The corpus gives a fairly clear answer. Value is moving away from discovery (finding and comparing options) and toward the parts of a booking that agents can't easily replace: settling the payment, holding the inventory, taking responsibility for the transaction, and controlling which prices agents actually get to see.

The first surprise is that the online travel agencies are not being cut out. They're being built further into the plumbing. Independent hotels' reliance on OTAs rose to 63.4% even as AI agents arrived. OTAs stay the merchant of record, the party legally on the hook for the sale, because agents reach them through APIs rather than going around them. Expedia's business-to-business segment, which sells this booking machinery to other companies, grew 24% and now makes up 38% of its revenue Are AI agents actually removing online travel agencies from bookings?. Stock-market behavior points the same way. Expedia, Booking and Airbnb gained in 2026 despite years of 'disintermediation' warnings, because AI makes search cheap but leaves payment, supply relationships, trust and completing the transaction in OTA hands Are AI agents actually disintermediating online travel platforms?. The layer that loses is the one that made money from human attention. Tripadvisor's hotel revenue fell 23% in a single quarter. When an agent ranks options by what the traveler said they wanted, paid placement and ad clicks stop mattering Are AI agents making attention-based ads irrelevant for aggregators?.

The less obvious place value collects is the pricing layer underneath everything. Agents search far more than people do, approaching 200,000 searches per ticket sold. So Amadeus stopped pricing every query live and now serves precomputed, cached fares for common routes. That sounds like a technical shortcut, but it means whoever decides what gets cached, and how often it's refreshed, shapes which offers agents can see at all. A cost-saving measure becomes a quiet source of market power Does caching fares give infrastructure providers pricing power?. On the supply side, hotels face a new gatekeeping question: whether they offer MCP endpoints, a standard way for AI assistants to check availability and book directly. Bookings are already going through Perplexity, Gemini and OpenAI, and Sabre and RateGain are building this infrastructure. Early adopters may capture that channel before new middlemen settle in Will hotels need MCP endpoints to reach AI travel assistants?.

The pull toward fulfillment is reinforced by trust and by payments. Travelers trust AI for brainstorming far more than for high-stakes tasks like visa requirements Do travelers trust AI differently for brainstorming versus visa tasks?. That leaves the accountable, error-sensitive steps with established players. Payment systems built around a human clicking 'buy' also strain when an agent starts the transaction. McKinsey proposes cryptographically signed mandates, a verifiable record of what the traveler authorized, to replace the human signals fraud systems rely on Does agent-initiated commerce break existing payment infrastructure?. Whoever issues and checks those authorizations gains a new hold on the transaction. One brake on all of this is that travel's data is split across airlines, hotels, distributors and agencies, so most agentic projects in the industry are still stuck as pilots Why is travel's agentic AI investment stuck in pilot mode?. Regulators like the UK's Competition and Markets Authority (CMA) warn that without data portability, the ability to move your data between services, this concentration could harden into lock-in Should data portability be infrastructure for agent-driven markets?.

The takeaway: in agent-driven travel, the strongest position is not having the smartest assistant. It's owning a step every agent must go through, whether that's the merchant-of-record contract, the cached fare, the booking endpoint or the payment authorization. One limit to keep in mind is that the corpus here is mostly analyst and industry commentary from 2026, not controlled studies.