Me: My Denver to Boston flight just got canceled. Get me home before noon tomorrow; nothing through Chicago.

Agent: Saw it 40 seconds ago. Your airline can put you on the 6:05 a.m. nonstop, 11 seats left, no fare difference.

Me: Other choices?

Agent: A 7:30 a.m. through Charlotte, and two other-carrier flights at roughly $214 more. If you decline everything, you are owed a refund. I can hold the 6:05 for 15 minutes.

Me: Hold it. Then book it.

Agent: Booked. Your Boston rental pickup now moves to 9:50 a.m. Want me to look at an airport hotel for tonight?

That exchange took a little under six minutes, start to finish, and it is a real one from February 2026. I have spent the last four years building and testing agents like this one for travel companies, and the lesson I keep repeating is simple. An AI travel agent can already run the rebooking triage faster than you can, but only when it is connected to the airline’s systems and has a spending limit you set in advance.

Without those two things, it is a confident search box that will describe a perfect reroute and then fail to book it.




Can it actually rebook me?


Yes, if it is wired in. Otherwise, it only talks.

The difference is plumbing. A working agent needs a live connection to airline inventory (a direct API or the newer NDC channel), a global distribution system for flights on other carriers, a payment method with a cap, and access to your loyalty account. Airlines and booking brands rarely build all of that on their own. They hire specialists in AI agent development to connect it, then run the result under their own name. I have sat in those scoping meetings. Picking the model takes ten minutes. Arguing about whether the agent may issue a refund without a human clicking approve takes three weeks.

Let me slow the Denver night down, because the order of operations is what separates a useful agent from a toy. The agent had been polling my flight status every two minutes since check-in. It saw the cancellation about 40 seconds before my phone buzzed, because it was reading the operational feed and not waiting in the push notification queue. Next it checked my own airline’s free reaccommodation options, since those cost nothing. Only then did it run the same Boston-by-noon search across other carriers, looking for anything a human agent could have booked under an interline agreement. It found the 6:05 a.m. nonstop and a 7:30 a.m. through Charlotte, read my fare rules, and held the seat. I tapped yes.

Here is that loop next to what you would do by hand.

Task after a cancellation What you do manually What a wired-in agent does What it still needs you for

 

Learn about the cancellation Wait for the push alert or gate announcement Reads the operational feed, acts in seconds Nothing
Find alternatives Open three apps, call the hotline Searches your airline and cross-carrier inventory in one pass Your preferences, like no red-eyes
Check rules and rights Google it at the gate, often wrong Matches your fare to DOT or EU rules Refund versus reroute is your call
Rebook or hold Line or hold music, 45 to 90 minutes Holds a seat under a preset limit The final tap, and spend above the limit
Fix hotel and car Separate calls the next morning Shifts or rebooks through partner connections Approval for new charges
Claim compensation Most people never file Drafts the claim with timestamps and the rule quoted Your signature and some patience

 

A general chatbot with no travel connections cannot do any of the middle rows. Paste your confirmation into one and it will hand you a tidy list of flights, a few of which stopped running years ago. I have watched people do this in an airport at midnight, feeling clever, and then join the line anyway 20 minutes later.


Is it allowed to?


Not unless you say so, and the airline agrees.

There are two permissions in play. Yours is easy: a cap on unapproved spend (mine is $300) and a tap for anything above it. I also tell the agent which things it may never do alone, such as buying a ticket on a different airline or accepting a downgrade in cabin. Early on, one of our test builds happily accepted a middle seat in the last row because nobody had told it that was a deal breaker. It did what it was permitted to do. We had written the permissions badly.

The airline’s permission is harder, and I will come back to it in the failures below, because some carriers now treat outside agents as bots.


Will it know my rights?


It will if it looks them up. Models that answer from memory get this wrong.

In the US, the Department of Transportation’s automatic refund rule means airlines must issue a cash refund within seven business days when they cancel and you choose not to take the alternative. A domestic arrival that slips three hours or more counts as a significant change you can also refuse. On a European itinerary, Regulation 261/2004 adds a second layer: a cancellation with under 14 days’ notice can be worth 250, 400, or 600 euros depending on distance, on top of rerouting or a refund.

Here is a case where the rule decides everything. If your replacement flight lands four hours later than the original, a human at the gate may offer a voucher and hope you take it. The refund rule says cash is yours if you decline, and an agent who has read the rule will put that sentence on the screen before it shows you the voucher. I have seen travelers take the voucher simply because nobody told them the other door was open.

A good agent does not choose for you. It lays out the free option, the paid option, and what you are owed whichever way you go. A bad one invents a “72-hour rebooking window,” which is exactly what one of our early builds did before we forced every policy statement to quote a retrieved document or say it could not verify the claim. Answers got a little blunter. Trust went up a lot.

One more detail from the Denver night. Three days later, the agent drafted my claim for the meal voucher I was owed, with flight numbers, timestamps, and the rule quoted. I edited one sentence and sent it. That is where I think agents quietly earn their keep, because most people never file at all.


What does it do about the hotel and the rental car?

It nudges the rental car before you remember it exists. In my case, it pushed the pickup back five hours by messaging the rental company. Had I been stuck overnight, it would have shortlisted hotels within a 15-minute shuttle ride and asked before booking, since hotels sat above my cap.



Where does it still fail?


Mostly at the edges that matter on a bad night. I have shipped some of these failures myself, so I will not pretend they are rare.

Booking authority comes first. Most consumer agents can read and recommend; some can hold, very few can pay. Even Delta’s Concierge, which rolled out to all SkyMiles members in August 2026 with rebooking and refund requests inside the chat, still carries a beta label and can cancel only a whole itinerary. That limit bit a member the same month. She asked it to cancel just her outbound flight; it told her the return “remains booked,” and the system refunded the entire award. Delta called it a bug and restored the trip. Notice the shape of it, though. The agent said one thing and the backend did another, and I would not call that a language problem.

Then there is the bouncer at the door. In an October 2026 report, Delta said it has no partnership letting outside agents book flights, and United pointed to terms of use that bar automated tools without written permission. So the assistant you bring from outside may get bounced at the login page, while the airline’s own agent walks right in.

Groups are a quieter failure. Moving one person onto an open seat is easy. Moving a family of five, two kids under 12 and a wheelchair request included, means keeping seats adjacent and service codes intact across a reroute, and most booking write paths were never built for a machine to do that unattended. Every agent I have tested either hands that to a human or quietly splits the family.

One more thing, and it is my read rather than anyone’s data. Agents fail in bursts. The storm that cancels 800 flights also hammers the systems the agent depends on, which is the one night you want a phone number written on paper.


Which agents exist right now?


Fewer than the marketing suggests.

Expedia and Booking.com launched apps inside ChatGPT on October 6, 2025. Good for planning a trip, not what I would open at 9:40 p.m. with a dead flight. Google said on November 17, 2025, that it is building agentic flight and hotel booking into AI Mode, with Booking.com, Expedia, Marriott, IHG, Choice, and Wyndham as partners, and its own lead described the work as just starting. A year on, I have seen restaurant and ticket booking go live and hotel demos, but nothing that will rescue a canceled connection.

Delta’s Concierge is the most capable airline-native option in the US today, and the reason is boring. It sits closest to its own reservation system, so there is the least distance between what it says and what it can do. The pattern holds everywhere I look: the nearer an agent is to one airline’s records, the more it can do, and the less it can do for you on a different carrier.

Cancellations are common enough to care about. The DOT’s full-year figures show 1.4% of US flights canceled in 2024, up from 1.3% in 2023, with 78.1% arriving on time. For 2025 the rate rose again to 1.5%, and December 2025 alone reached 1.6%. Roughly one flight in 65 does not go. The 22% that arrive late are where most missed connections live. Fly 16 segments a year, and you should expect to meet at least one cancellation every four years or so, which is my own back-of-the-envelope math, not the DOT’s. That is why I stopped treating disruption as a freak event when I design these tools.


Can I trust it with my card?


Today, only a little, and only with a cap.

Most travelers agree with that caution. An Expedia Group survey of 5,700 adults in the US, UK, and India, reported in April 2026, found only 8% would be comfortable booking through an AI platform, and 66% would not trust an AI assistant to book for them. Yet Phocuswright reported in March 2026 that 56% of US leisure travelers now use AI for travel, mostly for research.

Here is my inference, not the surveys. That gap between 56% and 8% looks like a systems problem wearing a trust costume. Planning is read-only, so a wrong answer costs nothing. Booking is where a mistake needs a person to undo it, and nobody has made that person easy to find yet. If an agent reports what the system actually did, every time, and shows a visible spending limit, I expect the 8% to climb. I could be wrong about how fast.

There is also a practical wrinkle for anyone using an outside agent. Log in through it a week before the trip. You want to learn about bot defenses on your couch and not at the gate. And write down the airline’s elite line, the international callback number, and the club location on paper. I have used that paper twice.

My own test is cheap. Before a trip, I ask the agent to change a seat or add a bag, then I compare its confirmation with the airline’s email. That one habit would have caught the Delta partial-cancellation bug before it mattered.


Verdict

Yes, with one condition, and the answer changes depending on whose agent it is. An airline’s own assistant can already save a canceled trip, mostly because it sits next to the reservation system and has the permission to act. An outside app is not yet, since the airline may block it and you cannot be sure what it really booked. In both cases, the condition I would hold to is the same: the agent must show you the airline’s own confirmation for what it did, and you must have set a spending cap before the trip began.


Today, only a little, and only with a cap.

Most travelers agree with that caution. An Expedia Group survey of 5,700 adults in the US, UK, and India, reported in April 2026, found only 8% would be comfortable booking through an AI platform, and 66% would not trust an AI assistant to book for them. Yet Phocuswright reported in March 2026 that 56% of US leisure travelers now use AI for travel, mostly for research.

Here is my inference, not the surveys. That gap between 56% and 8% looks like a systems problem wearing a trust costume. Planning is read-only, so a wrong answer costs nothing. Booking is where a mistake needs a person to undo it, and nobody has made that person easy to find yet. If an agent reports what the system actually did, every time, and shows a visible spending limit, I expect the 8% to climb. I could be wrong about how fast.

There is also a practical wrinkle for anyone using an outside agent. Log in through it a week before the trip. You want to learn about bot defenses on your couch and not at the gate. And write down the airline’s elite line, the international callback number, and the club location on paper. I have used that paper twice.

My own test is cheap. Before a trip, I ask the agent to change a seat or add a bag, then I compare its confirmation with the airline’s email. That one habit would have caught the Delta partial-cancellation bug before it mattered.


Lavinia Dsouza

Award-winning Analyst, multi-nominated digital content creator and photographer Lavina Dsouza's words capture stories about culture and tradition mainly through its food and people. She has written and contributed to publications such as The Washington Post, Lonely Planet and Matador Network, to name a few. She is the editor of UntraditionalHumans.com, a non-profit created to share inspiring stories from women of colour who break free from traditions and choose happiness.

She's also a speaker passionate about DEI and champions solo travel. She has collaborated with numerous renowned brands such as Intrepid Travel, TripAdvisor, Travel and Leisure and Adobe, to name a few.

She can be found on Twitter and Instagram.


PIN FOR LATER!

Stay in the loop

Slow travel.

In your inbox.

Curated wellness escapes, honest beach guides, and real itineraries —delivered thoughtfully. No spam, just good trips.

Newsletter


Curated wellness escapes, honest beach guides, and real itineraries,
delivered thoughtfully. No spam, just good trips.

Continent Hop pairs beautiful photography with honest guides that include meals you won’t want to rush,
sunsets at sea that linger in memory long after the trip ends,
& spa days that help reset the mood – written by Lavinia, not a robot!

Continent Hop pairs beautiful photography with honest guides that include meals you won’t want to rush,
sunsets at sea that linger in memory long after the trip ends,
& spa days that help reset the mood – written by Lavinia, not a robot!

Newsletter


Curated wellness escapes, honest beach guides, and real itineraries,
delivered thoughtfully. No spam, just good trips.

Continent Hop pairs beautiful photography with honest guides that include meals you won’t want to rush,
sunsets at sea that linger in memory long after the trip ends,
& spa days that help reset the mood – written by Lavinia, not a robot!

Connect

HOME    WORK WITH ME    DISCLAIMER + PRIVACY POLICY

 

Please note that some links on our blog are affiliate partners. If you choose to purchase through these links, we may receive a small commission at no extra cost to you. By using them, you are directly supporting Continent Hop to remain an independent travel blog, and to create free guides to help you travel more, travel better.

© COPYRIGHT 2015 – 2026 LAVINA @ A CAPRICIOUS COMPASS / CONTINENT HOP ¦ ALL RIGHTS RESERVED CONTENT (PHOTOGRAPHY + WRITTEN) CANNOT BE USED WITHOUT WRITTEN PERMISSION

Privacy Preference Center