Just like the CIO of Hilton, I may have underestimated how quickly people would trust AI with their credit cards. My assumption was that agentic AI in travel would arrive gradually. It took the internet years to build enough trust for people to comfortably put in a credit card and book a holiday online. But with AI everything goes a lot faster.
Meta has Muse. Instinct is claiming $1B in transaction. Priceline has made Penny agentic and is adding Lola. Agoda is testing an AI Assistant that moves from inspiration towards bookable trips. Dots from OpenAI are coming. It is moving so quickly that anything I write here will probably be outdated before some of you read it.
I argued that the big opportunity might be AI plus human, not AI replacing the human. A gig economy of travel assistants, where almost anyone could manage trips professionally because AI does most of the heavy lifting, which several are building. The human becomes the person you message when you need to book, change, cancel or fix something. AI provides the productivity. The human provides judgment and accountability.
I still think that model has legs. But perhaps I have been too conservative.
The problem is that almost all the proof of success we see today tests the easy part of travel: booking it. The interesting test begins when things go wrong.
Imagine an AI finds you and the family a clever combination of two cheap flights on different airlines with a 120-minute connection in New York. Great price. Perfect itinerary. Then the first flight is delayed. You miss the second one. Because they are separate tickets, nobody has responsibility for getting you home. The clever saving suddenly costs thousands.
Or the agent finds a hotel through an obscure intermediary because it is $80 cheaper. You arrive at 2am with two children and discover the hotel never received the reservation. The hotel is full. Your money has disappeared somewhere into the distribution chain. Now what?
These are not edge cases in travel. Storms happen. Flights get cancelled. Hotels overbook. Connections are missed. Travel is a collection of interdependent services, and when one breaks, three others can break behind it.
This is where the question becomes much more interesting. Who is responsible when the AI made the decision? Who pays when the optimization was technically logical but practically stupid?
For many households travel is one of the largest expenses of the year, planned for months and supercharged with emotions.
Agentic AI is clearly moving faster than I expected. People seem surprisingly willing to let it transact. Maybe full autonomous travel really does arrive faster than the human-plus-AI model I imagined.
But booking will not prove that entirely. The moment to watch is when the family is stranded at 2am, the flight has gone, the hotel has no room, and several thousand dollars are at stake.
That is when we find out whether we built a travel agent or just a very clever booking engine. So maybe the human plus agent model still has legs.
