Booking.com Builds AgentHub, an Internal Platform for AI Agents

Booking.com wants every team to be able to build an AI agent without starting from scratch. The company has described AgentHub, an internal platform for building modular agents, in a blog post covered by PhocusWire.

AgentHub uses configuration-driven setups and shared infrastructure. That lets people outside engineering, such as UX writers and product managers, build and test agents alongside developers. According to Booking.com, teams can deploy an agent "in a matter of hours."

Guy Nadav, director of machine learning and AI at Booking.com, wrote on LinkedIn that as the company shipped more generative AI use cases, it asked whether each team should keep solving the same problems. AgentHub is the answer. An agent can reason about a task and call tools that expose existing capabilities: retrieval, property data, recommendations, content intelligence, internal APIs and external services. Nadav said the company "simply added a new layer" on top of its existing ranking, recommendation and content systems that decides when and how to use them.

The first agent

The first agent live on the platform is AI Trip Planner Q&A. It answers travelers' questions about a property. It matches a property mentioned in a question to the right property ID, pulls information from internal systems, and works out distances and travel times to landmarks and transport hubs. Booking.com rebuilt an older, rule-based flow so the agent decides which systems to call. The company reports better answer relevance, better context relevance and higher factual accuracy, along with a lower rate of rejected LLM responses.

Planned uses cover the whole booking funnel: discovery and planning, booking, cross-selling and trip management. The roadmap includes agents that work together, support for images and other non-text input, long-term memory and personalisation, and better tool calling.

The wider picture

Booking.com is not alone. Expedia Group's chief AI and data officer, Xavier Amatriain, recently described a "composed, multi-agent system" for travel, with specialised agents for discovery, property comparison, itinerary building, questions and trip changes.

For property operators, the practical point is that traveler questions about your property will increasingly be answered by an agent using the data in your listing. Content that is complete and accurate, including location details and amenities, is what these agents will draw on. Gaps in your listing become gaps in the answers.