Marketplaces
When AI agents bypass the middleman
An AI agent might find the same rental for less on a host’s site. The hard part is proving that the offer is real and the booking is safe. That is where Airbnb’s fee faces its test.
- Booking the same home directly could cost less. In one example, the guest saves $75 and the host earns $52.25 more.
- The agent must find the real host, compare the full offer and finish a safe booking. Many homes may be available only on Airbnb.
- Airbnb can still charge for exclusive homes and support. Whoever controls the booking agent could become the next middleman.
Suppose a home costs $1,000 on Airbnb and $925 on the host’s website for the same nights. An AI agent could spot the cheaper offer and book it. But first it needs to know that the website belongs to the real host, the dates are available and someone will help if the stay goes wrong.
That is the challenge behind Mark Jenney’s recent “bear case” for Airbnb. Jenney says his rentals bring in almost $1 million a month. After trying AI agents, he argued that they could find the same home outside Airbnb and avoid its fee. He predicts Airbnb’s growth will slow within two to three years and turn negative within five. His exception: homes available only on Airbnb.
The forecast is uncertain. The useful question is simpler: which parts of Airbnb’s fee can an agent replace, and which parts still protect the guest?
What Airbnb’s fee buys
Airbnb helps guests find homes and hosts find guests. More homes attract more guests, which in turn attracts more hosts.
Under Airbnb’s single-fee plan, most hosts pay 15.5 percent of the booking price. Airbnb deducts it from the host’s payout. On a $1,000 stay, that is $155. Hosts can account for the fee when they set their prices, so guests may bear some of the cost too.
That fee pays for more than search. It also pays for work before and after the guest clicks “book.”
| What Airbnb does | What an agent could do |
|---|---|
| Help guests find a home | Compare listings and prices it can access |
| Help guests trust the listing | Check evidence, if reliable evidence is available |
| Take payment and resolve problems | Use authorized payment tools, but someone must handle refunds, fraud and support |
Airbnb chief executive Brian Chesky made the same distinction on the company’s February 2026 earnings call. A chatbot, he said, “can’t provide global payment processing, customer support or insurance.” Finding a cheaper listing is easier than making sure the guest gets the home.
Who keeps the saving?
Jenney says a direct booking could save “everyone 15%.” But the host decides how much of the avoided fee to pass on. Here is one possible split, before taxes and cleaning fees.
On Airbnb, the host gets $845 after Airbnb’s $155 fee. Direct at $925, the host gets $897.25 after a $27.75 card fee. The guest saves $75 and the host gets $52.25 more. The rest goes to card processing.
The host could instead charge $1,000 direct. Then the guest saves nothing, while the host gets $970 after the card fee. Direct bookings also have costs this example leaves out, such as finding guests and handling payment disputes. An agent can compare prices, but it cannot force the host to offer a discount.
Can the agent make a safe direct booking?
Some professional hosts list the same home on Airbnb and their own websites. Software such as Hostaway can keep prices and calendars in sync across those sites. That creates a possible route around Airbnb.
The agent still faces three checks. It needs a current calendar; using Hostaway’s API for that requires the operator’s permission. It needs to prove that the cheaper listing is the same home and belongs to the real host; copied photos are weak evidence. And it needs a way to pay and get help if something goes wrong. Airbnb’s terms also forbid bots from collecting its data, so scraping Airbnb is not a dependable way to compare listings.
Many homes may have no direct route at all. Chesky said in February that “the vast majority of our homes and our unique inventory is only on Airbnb.” That claim has not been independently measured. If it holds, an agent could suggest another home, but could not book the same one for less elsewhere.
Guests also give up Airbnb’s AirCover protection when they pay outside the platform. A direct site may offer its own guarantee, but the agent must check it. Saving $75 is a bad trade if a fake listing could cost the guest the whole stay.
Four things must happen before a booking leaves
Jenney’s forecast depends on four conditions. A direct booking happens only when every one of them holds:
- The same home is available elsewhere. Some homes are listed only on Airbnb.
- The guest uses an agent that checks other booking sites. Many guests may stay in Airbnb’s app.
- The direct offer is better overall. A lower price may not make up for weaker protection.
- The agent can finish a safe booking. It needs a real host, a live calendar and a working checkout.
To see why the conditions matter, imagine that half of homes pass the first check. Suppose 40 percent of those guests use a comparing agent, half of those find a better direct offer, and 80 percent of those bookings succeed. That leaves 8 percent of all bookings: 50% × 40% × 50% × 80%. These are made-up rates, not a forecast. The real rates are what we need to measure.
When any check fails, the agent cannot book that same home direct. Even a small share of direct bookings could matter if it includes valuable, professionally managed homes that appear on several sites.
How we could tell if Jenney is right
We do not need to wait five years to test the mechanism. Take a sample of Airbnb homes across destinations, prices and host types. For each one, ask an agent to find a legitimate direct offer for the same nights. Then check the final price, the host’s identity, the calendar, the checkout and the protection the guest would lose or gain.
The test should include fake host sites and stale calendars. It should give the agent credit for refusing an unsafe booking. A low price means little if the agent cannot show that the offer is real.
Airbnb’s revenue rose 17 percent in the second quarter of 2026. That tells us how the company was doing then; it does not settle what agents will do to bookings over the next five years. Better signals would be completed direct bookings through agents, the share of homes with a verified direct option, and any change in Airbnb’s effective fee.
Three ways this could play out
- Guests book direct. Agents find safe, cheaper offers with hosts. Airbnb loses those bookings.
- Guests still book on Airbnb. Outside agents send them there for the home or the protection. Airbnb keeps a role, though it may have to share revenue with the agent.
- Guests use Airbnb’s own agent. Airbnb handles both the search and the booking. It has already added natural-language search and AI summaries to its app.
The pressure reaches other markets too. An agent can compare a hotel room across booking sites or an identical product across stores. The marketplace can still earn its fee if it offers something the agent cannot easily supply, such as a room available nowhere else, reliable returns or help when a purchase fails.
The next middleman
Jenney expects agents to remove a middleman. But an agent used by millions of guests would become the first place they ask for a stay. Hosts and booking sites might offer discounts or pay to appear in its answer.
My last essay showed how travel booking screens and Google shopping results have favoured their owners. An agent could hide a similar bias inside a single recommendation. The guest would need to know whose interests it serves and why it chose that booking.
AI agents may make it cheaper to find a home. Whether they make the booking cheaper and safer depends on the homes they can reach, the evidence they can check and who takes responsibility when something goes wrong. Those are the parts of a marketplace’s fee that will face the clearest test.
Further reading
- Mark Jenney, the bear case for Airbnb (11 October 2026), the post this essay responds to.
- Airbnb’s help article on service fees, on the single-fee and split-fee structures.
- Gu and Zhu, Trust and disintermediation: Evidence from an online freelance marketplace (Management Science, 2021), the randomized trial on going around a platform.
- Rochet and Tirole, Platform competition in two-sided markets (Journal of the European Economic Association, 2003), on how platforms price two groups of users.
- Airbnb’s fourth-quarter 2025 earnings call (12 February 2026), where Brian Chesky discusses exclusive inventory and chatbots.
- Airbnb, Q2 2026 financial results (6 August 2026).
- Chen-Yang Su, When the buyer has an agent (10 October 2026), on how buyer agents change the reward for product quality.