The hotel discovery layer: Agentic discovery and the next intermediary
The first three shifts moved discovery to new surfaces. The next one moves it to a new actor
A traveler asks an AI agent to book three nights in Lisbon for late October — near Alfama, under €250 a night, with a workspace and a kitchen. The agent compares hotels, weighs reviews, applies preferences the traveler has expressed in past conversations, and returns one or two options. The traveler approves. The booking is made.
Some of these capabilities are real today. Most of them are partial. But the structural question they raise is sharper than the agentic adoption curve, and it is worth examining now.
The previous articles in this series tracked functions moving to new surfaces — TikTok, AI summaries, AI Overviews. Agents move them to a new actor. The traveler is no longer on a surface at all. The agent is.
What's actually happening
The announcements are real. Major OTAs have launched agent-facing infrastructure. Booking.com and Expedia have invested in agent integrations. The broader adoption of the Model Context Protocol lets agents query hotel inventory and rates programmatically. Hilton's AI Planner sits on the brand side. Google, OpenAI, and the major hyperscalers have all positioned agentic AI as a 2026 priority. Phocuswright reports that more than 60% of travel businesses are experimenting with or scaling agentic AI.
The operation is more limited. Agentic bookings remain a small share of total volume. Most agent flows still require human approval at the booking step. Full-funnel autonomy — agent searches, agent compares, agent books — exists in pilot form but not at scale. Phocuswright's Travel Forward 2026 found that around a quarter of U.S. travelers were open to agent-led booking as of spring 2025 — readiness, not adoption.
The gap between announcements and operational reality is wide.
What changes structurally
The structural shift agents represent holds up regardless of how fast adoption moves.
The first three forces in the series moved decisions to new surfaces — but the traveler still operated on those surfaces. The traveler scrolled TikTok. The traveler read the AI summary. The traveler clicked through the AI Overview. The discovery layer's audience was still a person.
With an agent, the traveler delegates. Hotels are then discovered, compared, and chosen by a system the traveler trusts to act for them. The discovery layer's audience changes from a person reading content to a system parsing structured signals.
This is a different intermediation than the previous three. With AI summaries, a model interpreted reviews for the traveler — but the traveler still made the choice. With an agent, the agent makes the choice. What hotels write, how they tag their inventory, which signals their content carries — these are increasingly being authored for an audience that does not browse, does not scroll, and does not respond to atmosphere or design.
The hotel's audience is no longer the traveler. It is the agent acting on the traveler's preferences.
The asymmetry
The asymmetry agents create is the most extreme of any force in the series. The agent's logic is opaque. The signals it weighs are not published. The hotel cannot see how its property is being represented to the traveler before the booking happens — or whether it is being represented at all.
The previous articles described intermediaries hotels could not fully see. Agents extend that. Hotels cannot easily verify which agents are in the loop, what queries they ran on the hotel's behalf, what alternatives they considered, or what tipped the choice. The first signal a hotel may receive that an agent considered them is the booking that arrives — or, more often, the booking that doesn't.
What the series has shown
The series began with the observation that hotel booking channels have stopped doing the work. It tracked four forces moving that work elsewhere — social platforms, AI search, the absorption of metasearch, and now agentic discovery.
The pattern is the same in every article. The function migrates upstream. The commercial controls do not migrate with it. The data hotels need to manage the new surface arrives later than the surface itself. The skills hotels have built remain valuable on the surfaces that already shrank.
Each shift has left hotels with less commercial relationship than the one before. Booking.com is a contractually negotiable partner. TikTok is not. AI Mode offers no negotiation at all. Agents may offer less than that.
The harder question — what kind of commercial function is fit for a layer hotels do not own, cannot fully see, and increasingly depend on — has not been answered by the series. It is the question the next phase of distribution thinking will need to answer.
by Markus Busch, Editor/Publisher Hospitality.today