The agent needs to know what a place actually is
So, what's an AI agent then?
Mark M
21 January 2026
Agent 2 Agent Booking
This small moment, the gap between the recommendation and the booking, is where a lot of the interesting questions about AI agents actually live.
That whole interaction, from question to confirmed booking, is what people mean when they talk about AI agents. And it's already happening.
A friend of mine booked a hotel last month. Nothing unusual about that, except for how it happened.
She asked ChatGPT for somewhere in the Cotswolds for her parents' anniversary. Quiet, good food, proper garden. It came back with three places. Not a list of 200 to compare and filter. Three recommendations, with an explanation of why each one fit.
She looked at the first one, thought "yes, that's right," clicked the link and made the reservation, done.
The mechanism
Most of what people call AI right now is really just a conversation. You ask a question and it gives you an answer, you ask for options and it suggests some. Helpful, certainly, but you're still the one in charge of what happens next. The AI responds, you decide what to do about it, and then you go and do it yourself.
An agent is a different thing. When we talk about an agent, we mean a system that actually acts on your behalf. The conversation that used to end with "here's what I found" becomes "I've booked it, confirmation's in your email." That's a meaningful shift, because to do that the agent has to be confident. Not just fairly sure, but confident enough to commit. Confident it understood what you actually wanted, that the place genuinely fits, that the booking will go through without a problem. It can't hedge the way a search engine can. It has to choose, and it has to be right.
That kind of confidence requires some specific things to be true. The agent needs to know what a place actually is, not marketing copy it has to interpret, but structured identity it can reason about directly. It needs to trust what it's seeing, which means verification from sources that matter rather than just taking claims at face value. And it needs a way to actually complete the transaction: availability, terms, a clean path from "yes, this one" to "done." Identity, trust, booking. Without all three, the agent gets stuck and hands you back to the old way of doing things.
These layers exist. Agent-to-agent protocols handle how systems discover each other, exchange structured information, and complete tasks. For hospitality, there's a layer that sits on top: schemas that describe what a venue is actually like, trust signals that flow from authoritative sources like tourism boards and verified operators, availability and booking that connect through. We've been working on that hospitality layer, and the specifications are published openly at AgenticBooking.
What has to be true for venues
For a venue to be one of those three recommendations, it needs to be understood by the system making the choice. Not just listed somewhere, but genuinely legible. The agent needs structured information about what the place is actually like, verification that the claims are accurate, and a way to complete the booking directly.
That's a translation problem, not a marketing problem. The information often exists already, it's just not in a form agents can work with.
The anniversary
My friend's parents had a lovely weekend, by the way. The place was exactly right. Quiet, good food, proper garden.
She found it, booked it, and confirmed it in a single conversation. No tabs, no comparison sites, no re-entering her card details somewhere else. The agent understood what she wanted, found somewhere that fit, and finished the job.
There's still plenty to figure out as we work through old systems like reviews and legacy booking platforms. But that's what an agent is. And that's how she booked a hotel.
Mark M