Airbnb chief executive Brian Chesky has a blunt diagnosis for why consumer AI agents still feel clumsy: they are being built on top of operating systems that were never designed for them. In an interview with TechCrunch published on October 1, a day after Airbnb switched on AI-powered search for guests in the United States, Chesky argued that the industry is missing a genuine AI operating system, and that until one exists, agents will remain trapped inside individual apps rather than cooperating with one another.
The comments matter because they come from the head of one of the world's largest consumer marketplaces, a company that has been notably deliberate about how it brings generative and agentic AI to its products.
The Missing Layer in the Agent Stack
Chesky's core argument is architectural. Today's agents run on iPhones, Macs and Windows PCs, platforms whose app model, permissions and developer kits were built for humans tapping through screens. In his view, nobody is yet building the equivalent of a full operating system and software developer kit for agents, and that gap shapes everything above it.
"No one is developing a full operating system software developer kit," Chesky told TechCrunch.
Without that layer, he suggested, the industry ends up with apps that carry agents inside them, rather than independent agents that can discover and work with each other. He pointed to earlier attempts by chatbot platforms to create their own app stores as an example of trying to bolt a new paradigm onto old foundations.
Crucially, Chesky does not think a startup or a single AI lab can solve this alone. He said meaningful change in how apps become agents largely depends on Apple or Google choosing to build a new platform, because they control the devices where consumers actually live.
Agents as "Apps With Faces"
Despite his skepticism about today's plumbing, Chesky sketched an optimistic long-term picture. He described future agents as something like apps with faces, able to talk to one another across platforms through open standards such as the Model Context Protocol (MCP). That kind of interoperability, he noted, could replace the bespoke business partnerships that apps once needed to integrate.
He expects Airbnb will eventually run a broader Airbnb agent that is interoperable with other agents via MCP, and said the company wants to make its own app more agentic while adding more user-interface controls.
Consumer AI Is Not Cracked Yet
Chesky was candid about the current state of consumer agents. He said he had tried booking Airbnb stays through personal agent products including Muse and Instinct, and that the experience did not work very well, adding that even hotels perform poorly in such agents today.
He also repeated a long-held view that a chatbot is not the right interface for e-commerce. Chat windows show only a handful of options at a time, take multiple turns to converge on an answer and strip out much of the pleasure people get from browsing and planning a trip.
That thinking shaped Airbnb's new search experience, which blends AI understanding with a more visual layout rather than replacing the app with a chat box. Chesky indicated the current version is not the final form, predicting an interface that sits somewhere between a chatbot and a traditional search grid.
What Airbnb plans next
- Multiplayer AI: Over the next three to six months, Airbnb will explore interfaces that several people can use at once, such as a group of friends planning a trip together.
- Voice agents: The company plans to roll out voice-enabled agents this fall for search and customer support, and Chesky expects voice to eventually overtake typing as a primary input.
- A more agentic app: Airbnb wants more of its own product to be operable by AI, with richer controls for users.
Why It Matters
Chesky's comments land at a moment when the agentic AI industry is racing to define its own infrastructure. Protocols like MCP have given agents a common way to call tools and services, and a wave of personal agent products is competing to become the primary interface for consumers. But protocols alone do not solve identity, permissions, payments or discovery across devices, the functions an operating system traditionally handles.
For businesses, the takeaway is practical. If Chesky is right, the near-term winners will be companies that make their services easy for any agent to use through open standards, while hedging against a future in which Apple or Google defines the rules. For developers, his remarks are a reminder that agent experiences built on today's app model may need rethinking once an AI-native platform emerges.
There is also a cautionary note for the hype cycle. When the CEO of a major travel platform says his own service performs poorly inside leading personal agents, it suggests that agentic commerce still has significant reliability and design problems to overcome before mainstream consumers trust agents to book, buy and plan on their behalf.
The Bigger Picture
Airbnb is not alone in rethinking interfaces. Retailers, travel companies and software vendors are all experimenting with agent-friendly surfaces, from conversational store builders to MCP endpoints. What makes Chesky's intervention notable is his insistence that the decisive move will happen at the platform level. Whether Apple and Google answer that call, or whether AI labs build their own operating layers, may determine how the next generation of consumer software is structured.
