Booking thinks AI travel agents need an operating stack

Glenn Fogel's sharpest point is not that Booking Holdings already has a permanent defense against AI travel agents. It is almost the opposite: he thinks the company has to assume there is no permanent defense at all. In his No Priors conversation with Elad Gil, the Booking Holdings CEO treats AI as a better interface for travel, but he keeps returning to the messy operating layer underneath it: inventory, customer service, partner relationships, regulation, payments, failures, and the economics of every model call. 1

Fogel's case for AI starts with a mundane observation: planning a serious trip is work. A multi-city family trip can involve flights, rooms, transfers, restaurants, loyalty miles, different budgets, and the problem of what happens when one leg changes. Wealthy travelers have long paid human concierges to absorb that complexity. Fogel's view is that AI lets a mass-market travel company offer some version of that help at software scale. 1

That framing matters because it separates travel agents from generic search. Fogel is not describing a chatbot that finds a hotel and hands the user a link. He describes a system that remembers preferences, tests permutations, backs out of bad plans, asks clarifying questions, and eventually becomes the single contact point when travel breaks. In travel, the breakage is part of the product. Weather, maintenance, missed connections, and cancellations can turn one bad flight into a chain reaction. 1

His concrete example is Priceline's Penny, which he says he used for a complicated Europe trip involving two cabins, adult children, different return cities, hotels, shuttles, restaurants, and frequent-flyer miles. The important part is not that Penny answered a prompt. It moved back and forth through constraints that a normal booking form was never designed to hold. 1

Gil notes in the episode that Booking's team had said Penny adoption doubled every month over the past few months, with signs of faster search, fewer steps to booking, lower cancellation, and better customer success. Fogel immediately narrows the claim. He says the absolute numbers are still small relative to a company that handled $186 billion of travel and more than a billion room nights last year. 1

That is a useful correction. AI agents can look impressive in demos long before they matter inside a large P&L. Fogel wants to know the unit economics: how many tokens the interaction consumes, which model should handle which task, what the cost of that booking was, whether the user returns, and whether the lifetime value justifies the spend. 1

This is where the episode becomes more practical than the usual agent discussion. In consumer travel, a model is not just a reasoning engine. It is a cost center attached to a conversion funnel, a support operation, and a repeat-use relationship. If the agent makes the trip easier but burns too much inference cost, the business has a problem. If it reduces support time but leaves customers desperate to talk to a human, the business has another problem.

Fogel says Booking's cost per customer-service contact is down while customer satisfaction is up, but he also says some users still want a person. That is the product boundary: automate enough to remove waiting and handoffs, but not so much that the user feels trapped inside a bot when the stakes are high. 1

Booking's scale is the obvious incumbent advantage. The episode says the company did $186 billion of travel last year, and Fogel points to Booking's global lodging footprint, alternative accommodations, hotels, and partner operations as assets that are hard to reproduce quickly. 1

But he rejects the clean moat story. His line is blunt: "There is no such thing as a moat." He says competitive advantages can disappear, so the only long-term answer is to keep developing new services and new ways of doing the work. 1

The distinction is subtle but important. Fogel is not saying scale is useless. He is saying scale has to be converted into operational advantage every day. Thousands of people work with hotels and property managers. Merchant-of-record travel is regulated differently across countries. A travel marketplace has to serve both travelers and partners. Those are not just database rows waiting for a better conversational interface. They are ongoing business systems. 1

That is the strongest argument against the simple "ChatGPT replaces travel sites" thesis. A general-purpose agent may own the user's intent. But fulfilling that intent still requires supply, pricing, availability, payments, policy handling, local compliance, and recovery when the trip goes sideways. The interface can move quickly; the operating system underneath is slower to copy.

Fogel joined Priceline in February 2000, right as the Nasdaq peaked. In the episode, he says the company's market cap fell from tens of billions to a few hundred million, with the stock near a dollar before a reverse split. Booking's official leadership page says he joined the company in February 2000 and has served as Booking Holdings CEO and President since January 2017. 2

That history shapes how he talks about AI. He sees the same speculative energy that surrounded the late-1990s internet boom: real technology, real new companies, and a lot of money that will still be lost. Asked whether today's AI company failure rate will resemble the dot-com era, he avoids giving a ratio. His point is simpler: speculative booms can fund genuine innovation and still destroy capital along the way. 1

For founders, his advice is not a spreadsheet rule about when to sell. He says the decision depends on the specific company, the confidence of management and investors, and what the founder is trying to accomplish. Wanting to make money is legitimate. Wanting to build something that matters is also legitimate. The hard part is being honest about which game you are playing. 1

The last third of the conversation turns to jobs, and Fogel's view is more worried than his travel-agent optimism might suggest. He uses Booking's own history as an example. In the mid-2000s, Booking.com needed humans to translate hotel content and customer-service scripts across more than 40 languages. Machine translation eventually erased much of that work. 1

He accepts the standard long-run argument that technology creates new work. His concern is timing. Jobs may disappear faster than new jobs appear, and not every displaced worker can move easily into the next category. The image he chooses is a 50-something truck driver whose identity and income vanish when the job is automated. 1

Inside Booking, his answer is upskilling. He says he talks with his people organization about training employees to become AI-literate, even when the training may help them find work elsewhere. That is partly self-interest: a company needs workers who can use new tools. But he also frames it as a duty to reduce the chance that people reject the technology because they fear being discarded by it. 1

The episode's useful tension is that Fogel is both an AI adopter and an operator who keeps asking where the bill lands. AI travel agents may become a better front door for customers. They may also make old support and translation jobs disappear, shift cost into tokens, and force incumbents to prove that their supply networks still matter. Booking's bet is that the agent is not the whole product. It is the new interface to a much larger machine.