Will AI Replace Travel Agents? The Barrier Is Not Itinerary Generation

Introduction

Glenn Fogel is not cautious about what AI may do. He wants Booking Holdings agents to understand traveler preferences, assemble complicated trips, and eventually propose changes before weather or mechanical problems break an itinerary. Yet the CEO who lived through the internet crash keeps asking a second set of questions: how many tokens a booking consumes, whether conversion and loyalty improve, who resolves failures, and why a platform is equipped to do the work.

One crash separates technological value from investment outcomes

Fogel joined Priceline in 2000, almost exactly as the Nasdaq peaked. He recalls its market capitalization falling from billions to a few hundred million dollars within nine months and the stock approaching delisting. He stayed through the recovery that followed. Those historical figures are his account and have not been independently checked here.

The experience leads him to reject the inference that because AI matters, every AI company is a good investment. Gold, automobiles, and the internet attracted capital and talent while producing extensive failure. A technology wave can change the world and still destroy the capital committed to many participants. He is equally unwilling to prescribe whether a founder should sell or continue. The answer depends on the business, financing, personal purpose, and the years someone is willing to devote.

A travel agent must solve a chain of dependent problems

Travel planning is not a list of search results. A family trip can include different cabins, mileage accounts, arrival and departure cities, hotels, transfers, restaurants, and individual preferences. A flight disruption can knock over every later reservation. Fogel therefore describes AI as a tool for serving Booking's two customer groups, travelers and partners, more cheaply and effectively, not as an automatic replacement for travel platforms.

He uses Priceline's Penny to describe the target experience. The system can ask about mileage balances, cabin preferences, and different family itineraries, then revise combinations. The higher-value next step is not prettier planning copy but coordinated recovery when a trip breaks. Fogel wants a system that can anticipate a likely problem and recommend a change before it happens. That remains a product ambition, not a demonstrated capability in this episode.

After the demonstration come tokens, conversion, and service costs

Penny adoption and outcomes are described positively in the conversation, but Fogel introduces an important denominator: absolute usage remains small beside Booking's annual travel volume.

Production requires a different ledger. How many model turns does a trip need? Which model should handle each task? Do inference costs produce better conversion, repeat use, loyalty, or lifetime value? When AI reduces customer-service cost per contact, does satisfaction rise too? When must a person take over? Fogel's standard is not whether the system can perform a task, but whether it creates net value at real scale.

That standard matches his capital-allocation rule. Invest internally when expected returns justify it, then consider acquisitions; if neither is attractive, return money to shareholders. AI does not receive an automatic exemption from that discipline.

Scale is an advantage, not a permanent moat

Booking has hotel and alternative-accommodation inventory, global partner relationships, payment capabilities, and operating teams. Fogel also emphasizes that acting as merchant of record in travel means meeting different regulatory requirements around the world. A new AI interface can appear quickly. Connecting supply, helping partners grow, serving customers, and meeting regulation is a different undertaking.

Even so, Fogel repeatedly says there is no moat. Current scale is a competitive advantage rather than permanent protection from innovation. The platform must keep improving the traveler experience and help properties and other partners solve demand and operating problems. Scale matters because it provides more real situations in which a service must prove itself, not because it lets a company stop changing.

Beyond efficiency lies a transition-speed problem

Fogel uses translation work at Booking.com to show that technological substitution has already occurred. The company's early multilingual content and service operation required many human translators; machine translation removed that category of work. This is his company history and does not establish the labor-market total.

His concern is the mismatch in speed. Jobs may disappear faster than new roles emerge, while a displaced truck driver or junior analyst may not move into a new occupation quickly enough. Eventual productivity gains do not automatically solve income, dignity, and skill problems during the transition.

Booking's response is to upskill employees and build AI literacy, improving their prospects even if a particular role cannot be preserved. Fogel offers no complete policy and acknowledges that government retraining has not always worked. He identifies a management responsibility: a company calculating AI return cannot pretend that human transition costs do not exist.

The conversation moves competition in travel AI beyond who can generate an itinerary. The harder question is who can connect preferences, inventory, partners, payments, regulation, and failure recovery at a sustainable cost, while accepting the real-world consequences of greater efficiency. Planning is only the entrance.