Airbnb CEO: Building AI Agents Requires a New Operating System, Not Chatbot-Only Interfaces

Airbnb's rollout of AI-powered search marks a deliberate shift for a company historically cautious about AI adoption. CEO Brian Chesky has consistently resisted the idea that conversational interfaces alone can drive discovery and booking—and his latest thinking goes deeper, suggesting the entire tech industry is building AI solutions on the wrong foundation.

The Chatbot Limitation#

Chesky remains skeptical of chat-only experiences for travel and shopping. Traditional chatbots force users into sequential back-and-forths, offering just a handful of options per turn. This approach misses what makes travel planning compelling: the exploration, the inspiration, the browsing through possibilities. Studies show people derive as much pleasure from anticipating a trip as from taking it—stripping away that discovery process could undermine the core appeal.

There's also a structural problem: chatbots assume a single user. Travel planning is inherently social. Groups of friends or families working together need interfaces that support collaboration, not just individual queries. Over the next few months, Airbnb is exploring "multiplayer" AI—tools that enable multiple people to interact with and guide the same AI simultaneously.

Beyond Single-User Interfaces#

Airbnb's upcoming search is not, in Chesky's view, the final answer. The company expects the interface to evolve between pure chat and traditional pre-designed screens, mixing what he calls deterministic (fixed, designer-controlled) and generative (AI-built-on-demand) components. Other startups like Monogram and Wabi are already experimenting with this hybrid territory.

The Missing Operating System Layer#

Chesky's core argument is more fundamental: the industry lacks a purpose-built operating system for AI. Currently, companies are creating AI apps for iOS and Android—but not "AI operating systems." Agents, he believes, need kernel-level access and standardized interoperability. Right now, the competition is fierce to become the primary "quarterback" agent users route everything through. That model, he contends, is flawed.

He points to ChatGPT's app store attempt as a cautionary tale. Without a full SDK and operating system architecture—the way Apple's App Store functions—such ecosystems fail. Airbnb has tested itself on consumer agents like Instinct and Muse; the results were underwhelming. "I don't think we've cracked consumer AI," Chesky said. "For consumer AI to be cracked, we have to crack at a lower substrate."