Airbnb CEO Brian Chesky has waded into a public debate on X over whether AI agents will eventually replace the rich, visual interfaces that apps like Airbnb are built around, arguing that the opposite is more likely to happen.
The back-and-forth started with a post from an account called Mostly Borrowed Ideas, which argued that Chesky was wrong to not see agents as a threat to Airbnb’s interface-heavy, visually driven booking experience. The post pointed to Meta’s Muse, an AI agent that can click through websites on a user’s behalf, as an example of the kind of straw man Chesky was arguing against. “Muse runs in a browser in Meta’s cloud, not on your phone, so a VM clicking through websites as you is a workaround, not a platform, and it exists because no agent gets to sit in front of the apps on the device,” Chesky wrote in response. “This creates a poor user experience.”
That reply drew a pushback of its own from X user Thomas Bekkers, who argued Chesky had it backwards. “You’re mistaken if you don’t think that’s the future. People will want their own agents doing the browsing. They’ll use agents to create personalized dashboards with house comparisons, etc. Everyone hates using agents that are unique to their website,” Bekkers wrote.
Chesky’s longer response is where the more interesting prediction shows up. “Here’s my prediction: Rich, diverse interfaces aren’t going away. And in the future, you’ll interact with brands more, not less,” he wrote. “This isn’t really about Airbnb. It’s about something much bigger. You’re describing an experience built on computing platforms that have fundamentally remained unchanged for 19 years on the phone and 42 years on the computer. Now, I do believe we’ll have a central agent, but it will operate at a much lower substrate than the app level. It will coordinate with many other agents, most of which you’ll also interact with directly. We’re living in a temporary state of computing purgatory. But not forever.”
In other words, Chesky’s bet is that a central, coordinating agent will sit beneath the operating system layer, orchestrating tasks across other agents, but that this won’t flatten the visually rich, brand-specific interfaces people use today into a single chat window. Instead, he expects people to keep interacting directly with those interfaces, and even more so than they do now.
The numbers so far back up Chesky’s skepticism
The exchange on X is playing out just as Airbnb has started attaching real numbers to how much of its business agents are actually driving. On Airbnb’s Q3 2026 earnings call, Chesky said traffic and bookings from agents like Muse and the startup Instinct are negligible against the company’s roughly half a million bookings a day, with the small trickle of agent traffic Airbnb does see having plateaued since early October.
Airbnb isn’t alone in seeing this. Uber’s CFO disclosed that AI agents accounted for just around 500 Uber Eats orders and four Uber rides out of roughly 320 million trips and orders during one of the company’s busiest weeks of the year. Not every AI integration has fallen flat, though. OpenTable’s integration with OpenAI’s ChatGPT has fared considerably better, generating 190,000 bookings in September alone, suggesting that simpler, more transactional bookings like restaurant reservations may be more AI-friendly than something as logistically involved as planning a trip.
That distinction lines up with the argument Chesky has made elsewhere about why travel resists being handed off wholesale to an agent. He’s previously described Airbnb as a “project-based” retailer rather than an “item-based” one, comparable to planning a trip with Ikea or Home Depot rather than picking up a single item at Walgreens. Booking a stay, in his telling, involves decisions about where to stay, what to do, and how to get there, all bundled into one highly logistical and often social process, with most guests browsing a dozen or more listings and messaging hosts directly before booking.
A bet on agents as infrastructure, not interface
What makes Chesky’s X exchange notable is that it reframes the entire debate. Rather than arguing narrowly about whether an agent should be allowed to book an Airbnb on a user’s behalf, he’s making a broader claim about how computing itself is structured. His view is that the phone’s app-grid model and the computer’s windowed, browser-based model have gone essentially unchanged for decades, and that the next shift won’t be agents replacing interfaces, but agents becoming an orchestration layer underneath them, one that coordinates with the rich interfaces people already use rather than erasing them.
That’s a meaningfully different vision from the one implied by tools like Muse, which effectively simulate a human clicking through a website inside a cloud-hosted browser. Chesky’s objection to that approach is structural: an agent working around a website through a VM, rather than being invited to sit inside a platform by design, produces what he calls a poor user experience, since no agent actually gets a privileged place in front of the apps running on a device.
Whether Chesky’s prediction holds will likely become clearer over the next year. Airbnb itself isn’t sitting out the agent race; the company has said it plans to launch its own AI travel agent in the first half of 2027, built around a richer, more visual interface rather than a basic chatbot, a design choice that tracks closely with the argument Chesky has now made publicly on X.