- Airbnb CEO Brian Chesky told TechCrunch that a chatbot is the wrong interface for travel. He says apps must become interoperable agents and that no AI-native operating system exists yet.
- He said Airbnb works poorly inside agents like Muse and Instinct. This is his account, not an independent test.
- Leaders should decide which tasks an agent may finish alone, which return to the app, and who approves each agent connection.
The interface is not the hard part
Airbnb CEO Brian Chesky told TechCrunch that the bigger gap sits beneath the screen. In his view, no one is building a full operating system for AI agents.
What follows is WebPulse's interpretation, not Chesky's claim. His three paths for agents each decide who handles the customer at a key moment. Read that way, the question for a company is less which chat window wins. It is where the handoff between an agent and your own app takes place.
What Chesky said
Airbnb has just added AI to its search, as part of a seasonal product release. TechCrunch notes the company has been slow and deliberate about AI. Chesky has not changed his mind that a chatbot alone is the wrong tool for travel discovery.
His reasons are practical. Chatbots show a few options at a time, so users need several turns to reach a result. They are also built for one person, while trips are often planned with family or friends. Airbnb plans to explore "multiplayer" AI, which several people can use at once.
He also said Airbnb's current search is not the endgame. He expects something between a chatbot and the first version Airbnb shipped. He expects a mix of pre-designed screens and screens an AI builds on the fly.
He credits chatbots with one strength: generating leads. Some people had feared that AI tools would push companies like Airbnb out of the customer relationship. Chesky raised that fear and then pointed to this strength.
How agents would reach Airbnb
Chesky described three ways a consumer AI agent could work with Airbnb. Each places the work in a different spot.
First, the agent hands the user off to the Airbnb app. That covers tasks such as messaging a host, comparing listings, verifying identity or using the map. Second, the agent application builds a more robust software developer kit (SDK), a set of tools that lets outside apps run inside it. That would bring the Airbnb app to the user. Third, Airbnb's own agents talk directly to agents from other apps.
For the third path, he named MCP as the likely link. It is a standard that lets AI agents connect with other software. He imagines separate Airbnb agents for exploring and customer service, and eventually a "macro Airbnb agent."
His argument is that apps used to be interoperable only when companies struck a business deal. Agents, which he called "apps with faces," could make every app interoperable once they talk to each other.
Why he says it does not work yet
Chesky was blunt about today's results. He said he has used Airbnb on the agents Muse and Instinct, and it "doesn't work very well." He said hotels work poorly there too, and Airbnbs work worse. His conclusion was that consumer AI has not been cracked.
This is his own account, not a test. The article offers no independent measurements. He also leads a company with a stake in how this plays out.
His diagnosis is that today's AI apps are built for iOS, macOS and Windows, which are not AI operating systems. He would like AI to work at the kernel level, the core of an operating system, so agents can interact from there. He said it is up to Apple, Google or someone else to build that platform. Until then, he expects apps that have agents inside them, not agents that replace apps.
He also named two misconceptions in Silicon Valley. One is that the interface will vanish and people will just talk and type. The other is that software will become completely generative.
What this means for your organisation
The source is an interview, so it describes intentions, not shipped results. Still, it shows a design question that any company with a customer-facing app may face. The questions below are WebPulse's own, not Chesky's.
Which tasks can an outside agent finish alone, and which must return to your own app? Identity checks are one example Chesky gave of work that may need the app.
If you connect to agents through MCP, who approves each connection? If an agent platform offers a developer kit to run your app inside it, who reviews what that agent can see or change on a customer's behalf?
Do you track leads that arrive through agents? Chesky credits chatbots with generating leads. Your team should confirm whether that holds for your own business.
The takeaway
Chesky's own position is narrow. Until Apple, Google or someone else builds a new platform, he expects apps with agents inside them. Whether he is right is open. That is his view, not a measured trend.
WebPulse's interpretation goes one step further. If each of his three paths puts the handoff in a different place, then the handoff is a design decision. It is worth making on purpose.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: TechCrunch.
