Brian Chesky thinks the AI industry is building on the wrong foundation. In an interview with TechCrunch AI, timed to Airbnb’s new AI-powered search in its fall update, the Airbnb CEO said the race to become everyone’s main AI agent is missing a key piece: an operating system designed for AI from the start.
That’s a bold claim from a company that’s been slow and careful about AI. It’s also worth hearing out, because Chesky has called a few of these shifts correctly before.
🧭 Chatbots still aren’t the answer for shopping
Chesky hasn’t changed his mind about chatbots. He says they’re bad at browsing and shopping because they show you a few options at a time and take several back-and-forth turns to get anywhere.
He also isn’t calling Airbnb’s new search the finished product. “I think that there’s going to be some interface that is, you know, I don’t want to call it a midpoint, but something between a chatbot and what you see in the first version we shipped,” he told TechCrunch AI.
His reasoning is specific to travel. Some tasks should be fast and invisible, like booking a flight. But according to Chesky, travel studies show people often enjoy planning a trip more than taking it. “You don’t want to take the inspiration and the browsing and the library experience of dreaming away.”
He sees a second problem too. Chatbots are built for one person, and trips usually aren’t. Over the next three to six months, Airbnb plans to explore “multiplayer” AI, meaning interfaces a group can use together while planning a trip.
🤖 Agents as lead generators, not threats
For a while, consumer platforms worried that AI agents would cut them out of the customer relationship. Chesky now says chatbots have turned out to be good at lead generation. So Airbnb wants to be more agent-friendly, not less.
That doesn’t mean he expects apps to shrink into data feeds behind one universal interface. His view:
- Interfaces should fit the job. Airbnb’s app looks different from DoorDash or Amazon for a reason.
- Agents can’t do everything yet. Messaging hosts, checking identity, using maps and comparing listings all need real UI.
- Something has to give. Either the agent hands the user off to the app, or agent platforms build SDKs (software developer kits) strong enough to “bring the app to you.”
He also expects Airbnb itself to become an agent, or a set of them: one in the explore tab, one in customer service, and eventually a “macro Airbnb agent” that talks to outside agents through MCP, the standard that lets AI agents connect to other software. He calls agents “apps with faces,” and argues they could make every app interoperable without the business deals that’s usually taken.
🏗️ The missing operating system
This is where Chesky’s argument gets sharpest. Today’s AI apps run on iOS, macOS and Windows, none of which were built as AI operating systems. Ideally, he says, AI would run at the kernel level (the core of the OS), with agents working together natively.
Instead, everyone’s fighting to be the “quarterback,” the main agent users go through for everything. The problem, he says, is that nobody’s building the full SDK that would make that work.
He has a receipt. Chesky says he told Sam Altman that ChatGPT’s early app store push needed a real developer kit and operating system behind it. “Otherwise, none of our functionality is going to exist. And of course, it didn’t work very well.” He’s equally blunt about today’s consumer agents. He’s tried Airbnb on Muse and Instinct, and says hotels don’t work that well, “but Airbnbs work even worse.”
🔮 What the next 1–3 years could look like
What stands out here is that a major consumer platform is saying, on the record, that the agent stack isn’t ready. If Chesky’s right, a few things follow:
- Hybrid interfaces win. Expect fixed, designed screens mixed with generative ones that AI builds on the fly, an approach startups like Monogram and Wabi are already testing.
- The OS layer becomes the real prize. Whoever ships a true agent SDK, whether that’s Apple, Google, Microsoft or an AI lab, sets the rules for everyone else.
- MCP-style interoperability becomes table stakes. Platforms without agent-ready endpoints will get skipped.
If you’re building a consumer product, don’t wait for a winner. Make your core actions (search, booking, messaging, verification) available to agents now. Keep investing in the parts of your UI people actually enjoy using. And if your product is used by groups, pay attention to multiplayer AI, because single-user chat leaves a lot on the table.
The full conversation, including Chesky’s thoughts on voice, is on TechCrunch AI.