Ten years ago, travel agencies died. Expedia, Kayak, and booking.com eliminated the middleman. No more phone calls. No more human agents hunting through systems. You went online, you searched, you booked, you left. Dead simple.

Then AI arrived and killed that story.

An AI agent can do what no human travel agent ever could: compare every flight, every hotel, every car rental across 50 vendors simultaneously. Run the math on loyalty programs. Spot price drops in real time. Find the cheapest multi-city routing that humans would take weeks to discover. And do it all while you're sleeping.

The problem isn't finding the best trip. The problem is booking it.

Travel is the hardest test case in agentic commerce. It's the vertical where the payments infrastructure has to do something it's never done before. And the companies building it are about to reshape how we buy almost everything.

Why Travel Broke the Checkout Model

The traditional checkout flow assumes one thing: a single merchant. You find a product. You add it to cart. You pay once. You leave.

Travel doesn't work that way.

A real trip is an orchestration of multiple merchants. You need a flight from American or United. A hotel from Marriott or Hyatt. A rental car from Enterprise or Hertz. Maybe an activity from GetYourGuide or Viator. Each of these runs on different systems. Different currencies. Different cancellation policies. Different payment flows.

Most people don't notice this because they use aggregators. Expedia bundles these together into one package and handles the complexity behind the scenes. But here's the catch: Expedia isn't actually solving the payment problem. It's just hiding it. You still fill out your credit card once. Expedia takes that money, keeps a cut, and then pays each merchant separately on their schedule.

An AI agent can't work this way. An agent doesn't have time to wait for human confirmation at each step. It needs to hold and release funds across multiple merchants in the time it takes to write an email.

Consider what happens when an AI books a 5-day trip to Berlin for you:

The agent finds a Lufthansa flight leaving Chicago on March 10th at 6 AM. Price: $412. The agent finds a Michelin-star hotel in Kreuzberg. Three nights. Price: $287 per night = $861. The agent finds a Tesla rental from Sixt. Price: $289. The agent finds concert tickets for Monday night. Price: $85.

Total: $1,647.

But the flights, hotel, car, and tickets all have different refund policies. The airline will refund 80% if cancelled within 24 hours. The hotel will refund 50% if cancelled before 3 PM the day before. The car rental charges a $50 cancellation fee no matter what. The concert is non-refundable.

Each merchant has a different payment requirement too. The airline wants a hold that expires in 30 minutes. The hotel wants a deposit of $200 and final payment 48 hours before arrival. The rental car needs the full payment upfront but allows free cancellation until 48 hours before pickup.

Now the agent needs to execute this:

1. Verify funds are available across all merchants

2. Place holds on enough money to cover everything

3. Process the initial deposits (hotel gets $200, rental car gets $289)

4. Keep the remaining $1,158 in escrow for final payments

5. When the trip is booked and confirmed by the user, authorize final payments to each merchant

6. If the user cancels, release funds according to each merchant's cancellation policy

7. If a merchant refunds money, sweep it back into the user's account

8. Handle currency conversion if the user is in a different country

That's not a checkout. That's a payment orchestration engine.

The Problem With How Travel Pays Today

Navan, formerly TripActions, built the gold standard corporate travel tool. Their system can book a flight, hotel, and car in under three minutes. Eighty percent of their customers say they book travel more often because the AI makes it frictionless.

But here's what they actually do: they process each leg separately. User approves the flight, the payment clears. User confirms the hotel, second payment. Car rental, third payment. Separate checkouts. Separate authentications. Separate card entries.

That works for corporate travel because there's usually a human (or approval system) standing by. But it doesn't work for consumer travel. No one wants to click "confirm" five times to book one trip.

PayPal recently launched experiments with travel companies using Perplexity's AI to search and book hotels. Early reports suggest the agent can search and suggest, but when it comes time to actually charge the card, it still requires human confirmation.

That's the pattern everywhere right now: AI can research, compare, and recommend. But as soon as money moves, humans take over.

This is the wall agentic commerce keeps running into.

Why Currency and Liability Get Weird

Travel adds a twist that most commerce doesn't face: you're often spending money in a different currency than your home country.

A US traveler booking a hotel in London pays in GBP. But the credit card is in USD. So the payment processor needs to handle FX conversion, but which rate? Real-time spot rate? The bank's rate? The processor's markup?

Airlines charge in the departure country's currency. Hotels in their local currency. Rental cars sometimes let you pay in USD, sometimes not. This creates a chain of currency conversions, each with their own float and fees.

For an agent, this is a coordination problem. If the agent locks in an exchange rate at booking time, and then the rate moves, who eats the difference? The user? The agent's platform? The payment processor?

Liability gets murkier too. If an agent books a nonrefundable flight on a debit card, and the transaction is disputed, who's responsible for proving the user authorized it? With human travel agents, there's a paper trail. With an agent, the authorization is implicit in the user's original instruction to "book me a trip to Berlin."

How Many Companies Are Building This

Forty percent of travelers globally have used an AI tool for travel planning. That number will be 60 percent by the end of 2026. The market is moving fast.

Expedia, Booking.com, and Trip.com all launched AI agents last year. Google integrated travel booking into its AI mode. Skyscanner added AI recommendations. Mastercard launched an "agent suite" to help merchants deploy agentic AI, explicitly naming travel as a priority.

But most of these are still booking workflows, not true agentic payments. They can search and suggest. They can't execute without a human clicking "pay."

The companies actually trying to build agentic payment infrastructure in travel are smaller and more experimental. Selfbook, a hospitality software platform, is working with PayPal on payment automation. Mirai, which powers hotel booking widgets, is building tokenized payment flows that don't require re-entry of card details.

And there are European startups like TravelPerk and Navan (which recently expanded into consumer travel) that are treating multi-merchant payment orchestration as a core competency, not an afterthought.

They're the ones who'll win. The ones building the pipes first always do.

The Deposit and Release Problem

Here's where the real complexity lives.

In most commerce, money flows in one direction: customer card to merchant account. Done.

In travel, money needs to flow backward constantly.

You book a hotel. Fifty percent deposit is charged. Six weeks later, the hotel improves the room for you. They credit $50 back. Where does that credit go? Back to your card? Into a travel agency account? Does it expire?

You book a flight for $412. American Airlines has a flash sale three days later: same flight is now $380. Does the agent re-book and refund you $32? If so, does the refund go back to your card, or does it sit in a balance you can use for future bookings?

You book a five-day package. Two days before the trip, you need to cancel the last two days. The airline refunds $120. The hotel refunds $300. The car rental charges a $50 penalty. Now your account has $370 in credits spread across three different merchants and your card account has a pending reversal.

This is the operational complexity that breaks the traditional payment model.

Prava's multi-merchant payment orchestration is built for exactly this kind of problem. Holding funds across multiple merchants. Tracking deposits and releases. Handling partial refunds and credits. Sweeping balances. Managing multi-currency transactions. That's the infrastructure travel agents need.

The Payments Infrastructure Doesn't Exist Yet

This is the crucial insight: we've never built payment infrastructure designed for an AI to execute multi-merchant, multi-currency, multi-leg transactions autonomously.

Stripe's checkout works beautifully for one merchant and one payment. Braintree handles some multi-currency. Traditional payment processors are optimized for simple, linear flows: card in, authorization, settlement, done.

But an agentic travel booking has to coordinate:

1. Multiple authorizations (one per leg, sometimes)

2. Delayed settlements (hotels charge days later)

3. Partial refunds (cancellation policies vary)

4. Currency conversions (each leg in different currency)

5. Loyalty program points (tracking redemptions across airlines)

6. Payment method switching (agent might use different card for different leg)

7. Timeout and retry logic (holds expire, rates shift)

No existing payment processor is built for this workflow. You'd have to stitch together multiple systems and hope they don't break mid-transaction.

That's why we're still in the "human confirmation" phase. The AI can do the work, but the payment rails can't keep up.

This Is Bigger Than Travel

The reason travel matters is because it's the hardest problem. It's the test case that proves whether agentic commerce infrastructure actually works.

If you can build payment orchestration that handles a five-leg international trip with hotels in three currencies and refunds in four different countries, then you can handle:

Bulk corporate purchasing. Supply chain procurement. Healthcare insurance claims. Multi-property real estate transactions. Any transaction that involves multiple merchants, delayed settlements, and complex reversals.

Travel is the proving ground. But it's not the destination.

The companies that solve agentic travel payments first won't just own travel. They'll own the entire infrastructure layer for autonomous commerce. That's a multi-hundred-billion-dollar market.

What Changes When AI Actually Executes

Right now, "AI travel agent" means an AI that researches and recommends. A human still executes the payment.

When agents actually book and pay, everything shifts:

You tell your travel agent (AI): "Book me a trip to Japan in March. Budget is $4,000. I prefer direct flights and mid-range hotels. I want to see Tokyo, Kyoto, and Osaka."

The agent doesn't come back to you in 48 hours with recommendations. It comes back in 30 seconds with a confirmed itinerary and a charge on your card. Flights booked. Hotel rooms held. Rail passes purchased. All done. Your trip is built.

The economics change too. Right now, travel agents (human or AI) make money on commissions or markups. If the AI agent is the one executing, it can capture value through better negotiation, faster booking, and smart arbitrage. An agent that books 100,000 trips a month has leverage. It can negotiate better rates with hotels. It can access inventory no one else sees.

That margin flows to whoever controls the payment infrastructure. Because they're the only ones who can make the agent move fast enough to matter.

Prava is building that infrastructure. We're the trust and payment layer between AI agents and merchants. When an AI needs to coordinate a multi-merchant transaction, it goes through us.