The gap in AI travel that nobody talks about
Ask ChatGPT to plan a trip to Barcelona and it will suggest an itinerary in seconds. Ask it, through the Expedia app, to find a hotel and it will show you live prices and give you a booking button. Ask Booking.com the same, or ask TripAdvisor for the top-rated restaurants near your hotel, and you’ll get a clean answer.
Now ask ChatGPT to do something else. Your flight tomorrow just got cancelled. Rebook you on the next available flight to the same destination. Or: you booked a hotel three months ago and want to shift your check-in from Friday to Saturday. Or: the fare on your March flight to Rome just dropped by $200, request a refund of the difference from the airline.
The AI will not do any of that. Not because the models can’t understand the request. They can. They just can’t act on it.
This is the gap in AI travel that gets very little press and almost no consumer coverage: the industry has shipped AI that can find and book a trip. It has not shipped AI that can change a booking once it exists. And the reasons this is hard are worth understanding, because they explain a lot about where AI in travel is really going.
What “agentic booking” actually means
The term “agentic AI” is having a moment. Search volume for the phrase itself is enormous. The word “agentic” in a business context usually means an AI system that can take actions on your behalf, not just answer questions.
In travel, agentic AI splits neatly into two use cases:
- Agentic booking (search and reserve). The AI finds options, applies your preferences, and makes a reservation. This is largely shipped. Expedia does this inside ChatGPT today. Booking.com and TripAdvisor have similar integrations. Kayak and others have their own AI agents.
- Agentic booking adjustment (change what already exists). The AI takes an existing reservation and modifies it. Rebook a cancelled flight. Change a seat. Move a hotel night. Request a refund. Cancel a car rental. This is what the industry has not shipped.
Almost every current headline about “AI travel agents” is about case 1. The much harder problem is case 2, and the reasons why are structural.
Why AI can book a flight but not rebook one
Four hard problems stand between “AI can book” and “AI can rebook.”
1. Authentication and identity.
Booking a flight from scratch through Expedia’s ChatGPT app is a session that starts fresh. You authorize Expedia to hold your payment method. You confirm the purchase. Expedia’s system knows you’re a first-time booker on this itinerary, and the transaction is a straightforward one.
Adjusting an existing booking is different. The airline’s system needs to know that the person requesting the change is authorized to make it. That authorization is tied to the traveler’s account, not to a third-party AI agent. Passing that authorization through an AI intermediary raises real security questions the industry has not solved: how does the airline verify that the person behind the chat is you, and not someone who has access to your account? What happens when an AI acting on cached credentials moves your seat while you’re not looking?
2. Fare rules, change fees, and contractual complexity.
Airline fare rules are notoriously baroque. Whether you can change a flight, at what cost, with what fare differential, subject to what routing restrictions, is governed by fare basis codes that expand into contracts hundreds of lines long. Hotels and car rental companies have similar systems. Non-refundable rate. Prepaid. Modifiable up to 24 hours out. Cancelable but not changeable. Changeable but subject to a rebooking fee equal to the fare difference plus $200.
A human at a call center can read those rules, explain them to you, and decide with you whether the change is worth it. An AI that just changes the booking without walking you through the rules is a customer-service disaster waiting to happen. Building an AI that reliably surfaces the right rule at the right time, gets your consent to the fees, and only then commits the change, is a much harder product problem than booking from a clean slate.
3. Payment authorization for fare differences.
A rebooking almost always involves money moving. Sometimes you owe a fare difference. Sometimes you’re owed a refund. Sometimes you’re owed a travel credit that expires in twelve months. Regulations in most countries require explicit user consent for each payment authorization, and that consent has to be traceable and revocable.
An AI that automatically authorizes a $340 fare difference without an explicit user confirmation is legally exposed. An AI that pauses to ask for confirmation is functionally not much faster than the airline’s own app or website. The value of automation collapses when the automation has to stop and ask.
4. Liability when something goes wrong.
This is the quiet blocker nobody wants to name. When an AI rebooks your flight and it doesn’t clear, or the refund never arrives, or the new booking gets you to the wrong airport, who is responsible? The airline that processed the transaction? The booking site whose AI initiated it? The AI provider whose model made the routing decision? The MCP developer whose tool wrapped the API call? The consumer protection frameworks for this are not written yet.
Booking sites and airlines are conservative about liability for a reason. Every one of them has watched competitors get sued or fined for automation gone wrong. Shipping an AI that can quietly change bookings, without a bulletproof answer to “who is liable when it goes wrong,” is a risk almost none of them are willing to take yet.
Where the industry actually is right now
Given all that, what’s the honest state of agentic booking-adjustment in September 2026?
- The major booking platforms have shipped agentic booking. Expedia’s ChatGPT app, Booking.com’s ChatGPT integration, TripAdvisor’s, and their equivalents at Kayak, Priceline, and others all let you search, filter, and book inside a conversation.
- The major booking platforms have not shipped agentic booking adjustment. As of publish date, no consumer AI product will reliably rebook a cancelled flight, change a hotel night, or request a fare-difference refund on your behalf. The airline apps, hotel apps, and booking-site apps still require a human to click through the change flow.
- Enterprise travel is slightly ahead. Corporate travel management platforms like SAP Concur and TripActions/Navan have limited agentic adjustment for business travelers, because the corporate account provides a cleaner authorization envelope. Consumer travel is behind.
- The specialized players are betting on watch-and-alert. Products like SlickTrip, which watch bookings and alert you when something changes, are shipping today. They deliberately don’t take the adjustment action. They tell you when an action is worth taking and let you take it.
Will AI replace travel agents?
This question gets asked a lot, and the honest answer is nuanced.
AI will absolutely replace travel agents for a specific slice of work: search, compare, book a straightforward itinerary. That work is already being replaced, and by 2028 most simple leisure bookings will happen inside a chat interface with no human agent involved.
AI is not replacing travel agents for the work that human agents actually make money on. Complex itineraries with ten touchpoints. Cruise-plus-air packages with tight connection windows. Corporate travel with policy compliance. Disruption recovery, the exact moment when the flight is cancelled, the hotel is oversold, and someone needs to solve three problems at once with the phone rep who has actual system access. Those jobs are still human, and they will stay human until agentic booking-adjustment ships reliably.
For consumer travelers who used to Google their own flights, AI is a real upgrade. For travelers with complicated needs or trips going sideways in real time, a human agent is still the right answer.
What has to change for agentic booking-adjustment to arrive?
Three things must change:
1. Delegated authorization standards. The industry needs a clean, standardized way for a consumer to authorize an AI agent to make specific kinds of changes on specific bookings, with a clear expiration and a clear revocation path. Something like the OAuth 2.0 framework for financial-account delegation, adapted for travel. There are working groups on this. There is no consumer product yet.
2. Machine-readable fare rules and change policies. The current fare-rules infrastructure is designed to be read by human GDS operators, not by AI agents. Rewriting fare rules and change policies into machine-parseable form, with unambiguous consent requirements, is a huge industry-wide effort. It’s happening slowly.
3. A liability framework. Someone, probably a combination of the airlines, the booking sites, the AI providers, and consumer regulators, needs to decide who is responsible when an agentic booking change fails. Until that framework exists, the risk-averse move is to keep humans in the loop for changes, which is where every major platform is today.
Timeline estimate: 2027 for early enterprise agentic adjustment products (with tight scope, on specific carriers, for specific booking types). 2028 to 2029 for consumer products that can reliably rebook a cancelled flight without human intervention. Sooner for narrow verticals (hotel-only changes, for example, which are contractually simpler than flight changes). Later for the truly complex cases.
Where SlickTrip fits in this picture
SlickTrip is not building agentic booking-adjustment, and that is a deliberate choice. When the SlickTrip MCP launches inside ChatGPT, it will watch your bookings and alert you when something changes: a preferred seat opens up, a fare drops below your target, a deal appears on a route you’re watching. It will not change the booking on your behalf.
That’s the honest current-state answer for what an AI travel tool can safely do today. Watch and alert, without acting. The traveler makes the actual change, on the airline’s site, with their own credentials, under their own authority. That model works now. Agentic adjustment does not.
When the industry solves the four hard problems above, authentication, fare-rules machine-readability, payment consent, and liability, the products that can safely act on your bookings will ship. Until then, the responsible product is one that gives you the information you need to act and gets out of your way.
For a look at what SlickTrip’s watch-and-alert MCP will do inside ChatGPT when it launches, see the pre-launch overview: SlickTrip Is Coming to ChatGPT: Better Seat, Fare Alerts, and Your Bucket List in the Conversation