“Agentic booking” has become one of those phrases everyone uses and few define. It sounds futuristic, powerful, inevitable. But when you listen closely, people mean very different things.
If we don’t clarify the term, we risk building the wrong infrastructure for the wrong model.
From what I’m seeing in the market today, there are at least three fundamentally different interpretations of “agentic hotel booking.” And they are not minor variations. They require different technical architecture, different payment rails, different identity models, and very different data flows.
1) AI-Assisted Booking (Inside a Consumer AI Platform)
This is what most people currently imagine.
A traveler opens ChatGPT, Gemini, or another consumer AI platform. They type: “Find me a 4-star hotel in Chicago next Tuesday under $300 near the convention center.” The AI returns recommendations. The user selects one, clicks to book, and either completes the transaction inside the AI interface or gets redirected to a supplier or intermediary to finalize payment. Confirmation follows.
The AI assists discovery and decision-making, but the human still drives the transaction.
In this model:
-
The AI platform becomes the discovery interface.
-
The booking may be completed via UCP or redirected to OTA or brand.com.
-
Payment rails often remain traditional (card-on-file, redirect checkout).
-
Identity is still user-controlled and explicit.
This is AI-assisted booking – enhanced search, conversational UI, but human-led execution.
2) AI-Mediated Booking (On the Hotel’s Own Website)
Here the interaction happens inside the hotel ecosystem.
The user visits brand.com and interacts with an embedded AI agent or chatbot. They enter dates, preferences, loyalty number. The chatbot surfaces room types, answers questions, applies promotions, and completes the booking – all within the hotel’s own digital environment.
The AI mediates the transaction but remains inside the brand’s infrastructure.
In this model:
-
The hotel controls the agent.
-
The booking engine, CRS, and PMS integrations remain internal.
-
Payment processing is directly tied to the hotel’s merchant account.
-
Identity is verified within the brand’s own system.
-
Data stays (mostly) within the brand ecosystem.
This is not the same as AI-assisted booking on ChatGPT. The distribution dynamics are entirely different. The hotel retains control over customer data, payment flow, and merchandising logic.
AI-mediated booking might be the more accurate term here – the AI is the interface layer, but the transaction remains within the hotel’s domain.
3) AI-Executed Booking (Agent-to-Agent / A2A)
This is where things truly become “agentic.”
In this model, the user may not actively participate in the booking process at all.
A meeting is added to the user’s calendar in another city. Their personal AI agent detects the travel need. It communicates directly with a hotel’s AI agent (or a distribution intermediary’s agent). The agents negotiate availability, rate, loyalty recognition, cancellation terms, payment credentials, and identity verification – autonomously.
The user receives a confirmation email.
No clicking. No browsing. No chat.
This is agent-to-agent (A2A) booking – AI executed.
In this scenario:
-
Identity must be machine-verifiable and portable.
-
Payment credentials must be securely delegated to agents.
-
Policies and preferences must be encoded in advance.
-
Contracting logic may be automated.
-
APIs and interoperability standards become mission-critical.
-
Trust frameworks and authentication layers are non-negotiable.
This is a radically different infrastructure challenge compared to a chatbot on a website.
Why the Distinction Matters
Lumping these three scenarios under one label (“agentic booking”) creates confusion.
Each model shifts power differently:
-
AI-assisted booking shifts discovery control to consumer AI platforms.
-
AI-mediated booking keeps control within the hotel ecosystem.
-
AI-executed booking shifts control to autonomous digital agents.
Each requires:
-
Different technical stack.
-
Different payment infrastructure.
-
Different identity and authentication layers.
-
Different commercial agreements.
-
Different risk management frameworks.
A chatbot implementation does not prepare a hotel for A2A.
An API partnership with ChatGPT does not equal autonomous commerce readiness.
And traditional booking engines are not built for delegated AI payments.
We are not talking about one future. We are talking about three parallel paths that may coexist… and possibly compete.
If we want meaningful industry progress, we need to stop using “agentic booking” as a buzzword and start preparing for each of these 3 different models.