Booking Holdings CEO on AI, Scale, and Survival

Glenn Fogel joined Priceline in early 2000 with the worst possible timing. He collected his Wall Street bonus in February, walked in the door, and watched the Nasdaq peak the following week. Within nine months, the company's market cap had fallen from roughly $15 billion to a few hundred million. The stock touched a dollar a share. They did a reverse split to avoid delisting.

He stayed anyway, and twenty-seven years later he runs Booking Holdings, a company worth north of $100 billion. That arc — the improbable survival, the grinding rebuild, the eventual dominance — is what makes Fogel's views on the current AI moment worth examining rather than dismissing. He has seen a speculative boom swallow the company he worked for and kept working. That tends to concentrate the mind.

Fogel sat down recently with investor Elad Gil on the No Priors podcast, and the conversation covered enough ground to reward close reading: agentic travel planning, the economics of deploying large language models at scale, what Booking's size actually means as a defensive asset, and the job displacement problem that most tech executives prefer to discuss abstractly. Fogel doesn't entirely avoid the abstractions, but he gets more specific than most.

The moat question

The most striking thing Fogel says — and he says it clearly, not as a hedge — is this: "There is no such thing as a moat. There is no such thing as somewhere you're going to be protected against innovation."

He means it as a motivational principle for his 25,000 employees. Stay hungry, keep building, assume nothing is permanent. I believe him on the philosophy. What I'd push back on slightly is the implication that Booking's advantages are therefore equivalent in fragility to a startup's advantages. They are not. The regulatory complexity Fogel himself describes — operating as a merchant of record across dozens of jurisdictions, navigating travel rules that vary by country and are tightening — is a genuine friction cost for any new entrant. So is the supplier relationship infrastructure: thousands of people working directly with hotels and property managers, not just indexing them. None of that is a moat in the sense of permanent protection, but it is friction measured in years and capital, not months. Fogel knows this, which is why he says it in the same breath as "understand what the business is before you decide to commit your capital." The no-moat framing is real as a mindset. As a competitive analysis, it undersells what Booking has actually built.

What Penny does, and what it costs

Priceline's AI agent, called Penny, has been doubling in adoption month over month, according to figures Gil cited from a prior conversation with Booking's team. Fogel walked through a genuinely complex family trip he ran through the system — split cabins, miles versus cash tradeoffs, the logistics of arriving in one city and actually starting your trip in another — and described the output as impressive. I have no reason to doubt the demo worked.

The more interesting part of the conversation is what Fogel said immediately after. Penny's adoption is growing fast, but in the context of the volume Fogel described on the podcast — he cited the company's annual travel processing and room nights as the frame of reference — it remains small in absolute terms. Booking isn't pushing the product fully yet, partly because the economics aren't fully understood. "How many tokens are we consuming? What is the cost of us getting that trip for that person, and what is our ROI?" Those questions, he said, don't have settled answers yet.

That's an honest thing for a CEO to say on a podcast, and it matters. The capability exists. The business model for deploying it at scale, with full confidence in the unit economics, is still being worked out. Fogel is right to hold both of those things simultaneously, and right to resist the pressure to roll out faster than the numbers justify.

On customer service, Fogel said on the podcast that cost per contact is down and customer satisfaction is up since deploying AI — and that the company is investing approximately $700 million this year across technology and related areas. He was also explicit that the math on full AI replacement in customer service is more complicated than it looks, because some customers want a human and forcing AI on them produces worse outcomes. "In the end it's always what's best for the customer." That's not a sentiment, it's a constraint on the deployment strategy.

The speculative boom question

Gil raised the dot-com comparison directly. His framing: roughly 1,500 companies went public around 1999-2000, and by his count perhaps two dozen survive. Does the same math apply to AI companies?

Fogel wouldn't guess at the ratio, which is the right answer. But his broader point is sound. Speculative booms have happened repeatedly across American economic history — the California gold rush, the Detroit auto boom, the internet — and they follow a consistent pattern: genuine innovation, massive capital inflows, massive failures, and a handful of survivors that reshape the landscape. The innovation is real; the valuations assigned during the boom are not. This cycle is bigger in absolute dollar terms, which means the eventual disappointment will also be bigger in absolute dollar terms.

The interesting wrinkle now is that some AI companies have built substantial revenue quickly, which wasn't universally true of internet companies in 1999. That changes the failure distribution somewhat, though not the overall pattern.

The truck driver problem

Fogel's most honest passage in the conversation is about job displacement, and the specific image he reaches for is a 50-something truck driver whose livelihood evaporates when long-haul automation arrives. He describes that person's situation with genuine discomfort — "that person's going to feel really bad" — and acknowledges that the rate of job disappearance and the rate of new job creation are probably not running at the same speed.

His prescription, to the extent he offers one, is corporate upskilling. He described talking regularly with his head of HR about AI literacy programs — building that into how the company operates, treating it as an obligation to employees rather than a nice-to-have.

I want to take that seriously, because Fogel seems to mean it. But I also can't pretend the history here is encouraging. Readers old enough to remember the steel mill closures, the auto plant consolidations, the offshoring waves of the 1990s and 2000s, know what corporate retraining programs have delivered in practice. They have not, as a general matter, made displaced 50-year-olds into technology workers. Government retraining programs have done no better. Fogel himself says as much about the government programs. The honest answer is that nobody has cracked this, and the people who will bear the cost of getting it wrong are not the people running $100 billion companies. Fogel's instinct to name this problem rather than paper over it is right. His solution — upskilling at the firm level — is better than nothing and insufficient to the scale of what's coming. Both things are true.

What the survival story actually tells us

The dot-com crash is a genuine data point about Booking's resilience, and Fogel wears it honestly rather than as mythology. The company didn't survive because it had a moat. It survived because it had a real product, stayed solvent when others didn't, and rebuilt over years when the market stopped rewarding optimism and started rewarding execution.

The question worth sitting with is whether that story is transferable to a faster cycle. The dot-com correction played out over years. AI capabilities are compressing development timelines in ways that could make the next disruption sharper and shorter. The survival skills Booking developed — patience, capital discipline, deep supplier relationships — were shaped by a world where you had time to recover and rebuild.

My read: those skills still matter, probably more than the AI boosters acknowledge. The companies that are making considered bets on deployment economics, as Fogel describes Booking doing, are better positioned than those racing to announce capability. Whether that advantage holds if the cycle moves faster than anyone expects is the open question. Fogel has earned the right to be optimistic about his company's adaptability. He hasn't earned the right to assume the next disruption will be as slow as the last one.

---

Bob Reynolds is Senior Technology Correspondent at BuzzRAG.