Key Takeaways
- Booking Holdings, led by CEO Glenn Fogel, isn't just bolting general AI onto its platforms. They're developing highly specialized, agentic AI like Priceline's “Penny” assistant, designed for complex, multi-variable travel planning.
- This focused, domain-aware agent can handle intricate requests (e.g., “my wife and I want to go up in the front of the bus” for a family trip) far more effectively than generic AI models, which often fail at travel commerce.
- The practical business impact is clear: Fogel states that their “cost per customer service per contact are down” while “customer satisfaction is up” directly due to this agentic approach.
- Booking's ambition extends beyond reactive support. Fogel aims for systems that “predict well enough what the problem may be before it happens. And suggest changing, fixing”—a proactive, predictive agent.
The Method: Agentic AI's Unfair Advantage
Most founders are experimenting with general-purpose AI. They're integrating LLM APIs into chatbots or automating simple tasks. Glenn Fogel, the CEO of Booking Holdings, offers a sharper path. His team isn't thinking broadly about AI; they're building pinpoint-accurate agents for specific, high-friction problems. Fogel sees AI as “an incredible beneficial tool and a way for us to be able to do our mission easier, cheaper, and better for our customers.”
The core of Booking's strategy is developing what Fogel calls “personalized agents that... know everything about you, everything you want.” The prime example is Priceline's "Penny" assistant. This isn't a chatbot that just summarizes information; it's an intelligent agent designed to execute complex, multi-constraint travel planning. Fogel shared an anecdote where he tested Penny himself:
What makes Penny so powerful? It's the deep integration of specific travel knowledge, pricing models, inventory, and user preferences, all within a narrow domain. This allows it to not only understand complex human intent but also to act on it by navigating the intricate web of booking systems. This stands in stark contrast to how general AI platforms often stumble when attempting real-world commerce or multi-step tasks. Booking's agent doesn't try to know everything; it knows everything about travel.
The impact isn't just better customer experience. It’s a direct hit to the bottom line. Fogel confirmed, “Our cost per customer service per contact are down… Customer satisfaction is up.” The vision extends to proactive problem-solving. Fogel's goal is to “have a system that actually we are able to predict well enough what the problem may be before it happens. And suggest changing, fixing.” Imagine an agent that re-routes your trip before a flight is even canceled. That’s the power of agentic AI applied with precision.
Where This Breaks Down
Booking Holdings has immense resources: decades of travel data, a massive engineering team, and established infrastructure. Replicating their agentic AI approach requires significant investment in domain-specific data and engineering talent. Simply hooking into an LLM API won't cut it. Your agent needs deep, proprietary knowledge of your product, your pricing, and your customer behavior to be truly effective. This method also struggles if your problem space is too broad or too dynamic. If the rules of your domain constantly shift, maintaining an effective agent becomes a continuous, high-cost battle.
What to Do With This
Forget general chatbots. This week, identify the single most complex, multi-variable problem your users or customers face that currently requires significant human intervention or frequently causes frustration. Instead of building a generic assistant, design an "agent" to solve just that one problem, end-to-end. Equip it with all your proprietary data, business rules, and integration points to act autonomously. Measure the specific impact on resolution time, customer satisfaction for that problem, and the cost of human support. Aim for a specialized tool that executes, not just informs.