An AI agent booked around Airbnb. The commission is now a price tag

You know the moment. You find the perfect rental, hit “Reserve”, and only at the bottom do you see what the platform adds on top. Until now you paid. Now an agent asks first: is the same place cheaper somewhere else?

That is exactly what Nicolas Bustamante did, an AI executive at Microsoft according to Business Insider. Before paying for an Airbnb stay for his family, he had an AI agent search the web for the same property on its own website. The agent found it, checked availability for the same dates and concluded that booking direct was cheaper. Savings, per the report: 14%. In the same piece, Airbnb CEO Brian Chesky is quoted as saying agents are barely making bookings on his platform. Airbnb is betting on an agent of its own.

The commission paid for two jobs. The agent now does one of them

Sounds like a travel hack. It isn't. Savvy travelers have done this by hand for years. What's new is that it now costs nothing: no searching, no comparing, no ten open tabs.

As I see it, a marketplace fee always paid for two jobs. One, discovery: the marketplace gets you to the listing. Two, trust: payment, cancellation, reviews, someone to call when things go wrong. The agent now handles discovery in seconds. That leaves trust as the only reason to pay the fee.

Marketers have seen this movie. In 2009, Cornell researchers listed hotels on Expedia and pulled them off again in alternating weeks. In listing weeks, bookings through the hotels' own channels rose by an estimated 7.5 to 26%. The “billboard effect”: the marketplace was the billboard, the booking often went direct. An agent turns that effect into automation. The billboard stays. The toll booth gets driven around.

The catch: without web search, the AI stays loyal to the platform

Today, Nov 14, 20:57 CET, I asked ChatGPT and Claude (in German) how to pay the least for a rental I found on Airbnb. Both told me to book through Airbnb and warned against paying off-platform because you lose buyer protection. Neither suggested looking for the property's own website. Limits: an API sample without web search, one answer each, results vary.

That's the real point. The chat defends the platform's trust layer. Only the agent with web search routes around it. And that's where the winner is decided: does the agent find your direct offer, does it recognize it as the same place, and does it trust your checkout? Marketplaces get this. Amazon is trying to become the agents' data source instead of being bypassed. Let agents in and you get compared. Lock them out and you get routed around.

What I'd do now

• Run the test yourself: hand an agent with web search your own marketplace listing (rental, product, course) and ask for the cheapest way to buy. Note whether it finds your website.
• Make the direct offer unmistakable: same name, same photos, price and availability as structured data, and don't block AI bots. Whether big sites do, see the AI Agent Radar.
• Make trust readable: spell out cancellation, payment protection and reviews on your own page. And read your marketplace's terms before you actively steer customers away from it. More: Marketing to AI agents.

Bottom lineOnce the agent does the searching, nobody pays for discovery anymore. Only for trust.