Brian Chesky says AI agents need an operating system of their own

The interface is still unsettled

Chesky sees chatbots as a poor fit for e-commerce: they show only a few options at a time, and users may need several prompts to find what they want. But he does not consider Airbnb’s current search interface the final answer either. He expects an intermediate format between a chatbot and Airbnb’s first version of search.

That distinction matters for travel. Some people want an agent to book a ticket without involving them in the details. Airbnb users, Chesky argues, often want to browse. Research on travel suggests that planning and anticipation can be more enjoyable than the trip itself, so a useful interface must preserve discovery, inspiration and curated options.

Chatbots also tend to serve one person at a time, while Airbnb trips are often planned with family or friends. Over the next three to six months, the company plans to study interfaces for multi-user AI that let a group work together on a trip.

Other companies are exploring interfaces that change as a user works. Startups Monogram and later Wabi have experimented with generative interfaces, where AI creates screens on the fly rather than relying only on a designer’s preset layout. Chesky expects future products to combine generated elements with predictable, designed ones.

Agents need more than access

Airbnb is preparing to make its service easier for consumer agents such as Instinct and Muse to use. Chesky sees agents as a possible source of new customers, but the company also needs reliable software infrastructure and a way to hand control back to users.

That handoff matters because booking a stay can involve more than selecting a room. Users may want to compare options, message a host, verify identity, use a map or add services. Airbnb could build a richer set of developer tools so an agent can show users what the app can do, or let the agent pass control to someone who wants to stay inside Airbnb.

Chesky also expects agents to interact with one another, both within Airbnb and across apps. Apps have not generally connected without separate business agreements between their companies. Agents, he argues, could make that coordination easier. He imagines an Airbnb agent appearing in search, customer support and other services, with a broader Airbnb agent eventually communicating with agents on other platforms through MCP, a standard for connecting AI agents to software.

But the infrastructure is not there. Companies are building AI apps for iOS, macOS and Windows; Chesky says those platforms are not operating systems for AI. In his ideal, AI would work at the operating-system kernel level, allowing agents and other components to interact.

For now, companies are competing to become the user’s central AI agent — the “quarterback.” Chesky’s objection is that no one is building the full developer toolkit and operating system that applications need. He said ChatGPT tried to launch an app store early last year, and that he told Sam Altman an App Store-like marketplace needs both an operating system and developer tools. Without them, app features cannot work. Chesky said the attempt did not work well.

The missing layer

Chesky has tried Airbnb in Muse and Instinct. He says the service works poorly there, especially for homes compared with hotels, and that consumer AI does not yet handle tasks well. His proposed fix is structural: agents need to become apps, apps need to become agents, and both need better interfaces and ways to work together.

I think this is the more consequential part of Chesky’s case. He is not arguing that Airbnb needs a better prompt box; he is arguing that the platform layer beneath the interface is incomplete. That makes his vision dependent on a new platform from Apple, Google or another company. Without one, AI agents may remain features embedded in existing apps rather than apps that can work across services.

Chesky rejects two common Silicon Valley assumptions: that interfaces will disappear in favor of speaking or typing, and that software will become entirely generative. He believes designers can prompt for better software than ordinary users can. He also expects apps to remain central while the iPhone remains dominant and Apple and Google control the main platforms.

What I’d want to know is how much of Airbnb’s agent strategy can work without that new platform. The company can make its own app more useful to agents, but the broader promise — agents coordinating across services — depends on standards and infrastructure that Chesky says do not yet exist.

Voice moves closer to the product

Chesky expects people to speak to computers more often than they type. He pointed to newer voice-note apps such as Wispr Flow, Willow and Superwhisper, as well as voice input built into AI tools. Airbnb plans to launch voice agents in the fall to help users with search and customer service.

He also says AI is helping Airbnb ship features faster and helping him get information with fewer meetings. That is a more immediate use of AI than the agent ecosystem he describes elsewhere: a reduction in communication steps inside the company, even as the larger software stack remains unfinished.