OTA moats against AI disruption may be shallower than they appear

I wrote an article around 18 months ago on why intermediaries are in trouble in the land of AI. Mario Gavira wrote a rebuttal arguing that the OTAs would be supercharged in this new era. 

And now, just a few weeks ago, he's written a follow-up explaining how intermediaries will be damaged, but only some and only slightly. The argument got much smaller. It’s almost a 180.

Mario opened the piece with a great anecdote, which I've seen a few times now. On stage at the Phocuswright AI Marketing Summit in front of 250 travel marketing executives, he asked how many had tracked at least one end-to-end agentic booking. One hand went up.

He used this to argue that agentic disruption is overhyped. The bookings aren't there yet. The LinkedIn posts are louder than the reality.

There are a few problems with this argument.

• Nobody in any audience actually knows the definition of an “end-to-end agentic booking.” Just the word "agentic" has lost its meaning. Almost everything in AI right now is some version of agentic.
• One definition could be: I ask for a hotel in Barcelona June 4–5, send, and the next thing I see is a reservation in my email box. If that's the definition, I might suggest that will never happen in travel, nor does it need to, for all of this to have a massive disruption on the current travel players. The part that's getting disrupted, and the thing that's a threat to all OTAs, is that the discovery part is moving away from websites and into the chatbot. Right now that's owned by OpenAI, a bit of Anthropic and a lot of Google. But Google today is not the same as the Google of two years ago.
• You could have asked the exact same question about mobile bookings in 2008. Here we are in 2026, and making all kinds of travel bookings on a mobile phone is second nature to most people.

Loyalty

The first big moat Mario claims for OTAs and chains is loyalty. Bonvoy, Genius, One Key, etc. The argument is that agents will respect these preferences, so the chains and OTAs become the default before the search even begins. This is almost exactly backwards.

Loyalty programs exist because comparison shopping is expensive for humans and our tiny data-compressing brains. Reading reviews, checking prices on 57 sites, trying to keep your tab previews big enough to see the icons—it’s tedious. So you accept slightly worse value in exchange for not having to think about it. Status, points, recognition, sunk cost. They're all taxes on human shortcomings.

Agents don't have inertia. The whole point of an agent is that it strips friction from comparison shopping. The rational user instruction in an agent-first world isn't "protect my Bonvoy balance." It's "find me the best stay for these dates." Loyalty just gets stripped down to dollars and cents. That's all loyalty points are anyway: Discounts wrapped up in special tricks to fool the humans.

Unique inventory

Next is the unique inventory argument. The Airbnb treehouse in Costa Rica that an agent supposedly can't source anywhere else.

That may be the case for some listings in 2026, but it only really works looking backwards. If agents are bypassing intermediaries because they can, then every piece of inventory will end up with a direct booking path, because now it comes with distribution attached through agents. 

Booking.com already has inventory pretty close to Airbnb. Vrbo isn't far behind. It's a chicken and egg argument. When the demand comes, the supply will meet it. And the agents create the direct demand for anybody who wants to multi-home their listing.

Bundling

This is an easy one. It's not really the hardest thing for an AI agent to come up with bundling technology. Google last week announced "Universal Cart," which does just that. 

People will cry out that, "No, no, travel is different. It’s very special. Travel is more complicated." Yes, in some cases, travel does come with extra options and choices as part of the booking transaction, but this also isn't the most difficult thing to create these days. 

When you're booking travel, having everything in a single shopping cart accessible to an agent should be quite a valuable addition. The agent could explain layovers, distances between planned destinations, transfers, time required at locations and anything else that currently makes booking travel quite complicated.

Bundling as a defense was always a workaround for the fact that human attention is scarce. Software attention isn’t scarce. The moment agents (or third-party services) are good enough to do dynamic bundling, OTA bundling becomes more expensive, less flexible and slower than what an agent can construct on demand. It switches from moat to liability.

What does the ideal booking actually look like?

The arrival of agentic bookings isn't really the thing to look out for. ChatGPT, Gemini and others have already disrupted the booking discovery channel massively in the last three years. Most data not coming out of an OTA shows the huge increase in travel discovery on these models, and it's not going to stop anytime soon. Even without pure agentic bookings, when discovery moves, there's disruption.

A thought experiment that's interesting to try right now: What would the absolute ideal booking experience be if we weren't constrained by technology? I think it's fair to say that's where these models are going. The agent will, in a few seconds, search all of the possibilities relevant to your prompt and return a small subset of options.

For a hotel in a specific pre-decided destination, I can imagine maybe three to five results:

• A direct booking link with the supplier—this seems sensible. Some people value it. 
• A link with maximum trust, maybe a major OTA booking link, where trust is still a valid factor (because humans will take some time to transfer their trust 100% to their agent).
• Maybe the lowest price, which might come with some conditions. This could be 2-3 options, with differing conditions, and maybe mixed risk profiles.

After that, what's the value of anything else—something more expensive, less trustworthy or not direct? I don’t know why an agent would ever display it.

Once I know the agent has searched all the sources just like a metasearch engine, I don't need to continue shopping. I just need assurance that there's no better option out there. Why would I, a human (tiny brain, limited time) think I could find a better result when my AI has already proven it's exhausted its search?

The Grace Period

There is a catch though. When it comes to the core accommodation transaction, top-tier OTAs are already remarkably close to a perfect friction-free experience. They have massive inventory breadth, built-in consumer trust, and lightning-fast checkout.

Will AI construct a better experience than this? Yes. But will that incremental improvement be enough to immediately break deep consumer habits and drag users out of native OTA apps? That buys the giants like Expedia, Trip.com and Booking some time.

Furthermore, Google isn't going to turn off its ad monetization machine overnight. It will keep selling ads across its surfaces, and humans will keep clicking them. Big OTAs still possess the capital and data scale required to optimize for those ad slots better than independent suppliers.

But if you are a second or third-tier OTA caught in the middle without that massive scale? Your piece of the travel transaction is about to become small to non-existent.