AI is getting surprisingly good at planning trips.
Give it a destination, a budget, and a few days, and it can generate an itinerary in seconds.
Five days in Tokyo?
Shibuya on Day 1.
Asakusa on Day 2.
Tokyo Tower on Day 3.
Ginza on Day 4.
TeamLab on Day 5.
It looks perfect.
The problem is…
I might hate it.
I don't like crowded places.
I wake up late.
I care more about food than landmarks.
I don't want to spend half my trip rushing between “must-see” attractions.
And I would happily spend $250 on one amazing dinner instead of visiting five popular tourist spots.
The itinerary isn't wrong.
It just isn't mine.
And I think this reveals one of the biggest challenges for AI travel.
It's knowing how I make decisions.
For a long time, personalization in travel has mostly meant collecting preferences. Beach or mountains?
Budget or luxury?
Business or leisure?
Window seat or aisle?
But human travel decisions are much messier than that.
Two people can have exactly the same destination, budget, and travel dates—and still want completely different trips.
One person might want to stay in the center because they want to walk everywhere.
Another might prefer a quiet neighborhood and take a taxi whenever necessary.
One traveler wants to see everything.
Another wants to do absolutely nothing before noon.
One person sees a $300 hotel as expensive.
Another sees it as a bargain if it means waking up next to the beach.
The difference isn't simply preference.
It's trade-offs.
And that's where I think AI travel agents still have a lot to learn.
It's the one that understands your priorities.
Think about what a great human travel agent does.
You tell them:
“I'm going to Tokyo.”
They don't immediately send you a list of 20 hotels.
They ask questions.
“Is this your first time?”
“Are you traveling with kids?”
“Do you care about nightlife?”
“Do you mind taking public transportation?”
“Would you rather save money on the hotel and spend more on food?”
And sometimes they learn something even more important:
“I know you said you want to visit five places, but based on how you usually travel, I think you'll hate that schedule.”
That's valuable.
Because travel isn't about finding the objectively “best” option.
There usually isn't one.
It's about finding the option that is best for you.
Today, most AI travel experiences are still recommendation engines.
Ask a question.
Get an answer.
Ask for a hotel.
Get a list.
Ask for an itinerary.
Get a plan.
But agentic AI is changing the relationship.
The next generation of AI travel agents won't just tell you what to do.
They'll search, compare, make decisions, book hotels, arrange activities, and potentially manage the trip afterward.
That means the quality of the agent won't just depend on how well it can generate text.
It will depend on how well it can make decisions on your behalf.
And that's a much harder problem.
Imagine an AI agent searching for hotels for me.
There might be 500 hotels that match my destination and budget.
Which one should it choose?
The cheapest?
The highest rated?
The closest to the station?
The most popular?
The hotel with the best cancellation policy?
There is no universally correct answer.
The right answer depends on me.
If the agent knows that I hate crowded areas, love food, sleep late, and don't mind paying more for convenience, its definition of “best hotel” changes completely. This is why I think the future of AI travel won't be about generating more recommendations.
It will be about understanding the person behind the request.
An AI can create a beautiful itinerary in seconds.
But hotels have real-time availability.
Prices change.
Rooms sell out.
Cancellation policies differ.
Room types matter.
A recommendation that was perfect five minutes ago might no longer be bookable.
This creates a huge gap between:
“I found a great hotel for you.”
and
“I found a great hotel for you, and you can actually book it right now.”
For an AI travel agent, that distinction is everything. Because once AI starts making decisions rather than simply giving suggestions, access to reliable travel infrastructure becomes just as important as intelligence.
A brilliant agent with outdated inventory isn't very useful.
A personalized itinerary with no bookable room isn't a completed trip.
The intelligence has to connect to the real world.
The first phase of AI travel was about answers.
“What should I do in Tokyo?”
The second phase is about recommendations.
“Which hotel is best for me?”
The next phase is about actions.
“Book it.”
And eventually, the most interesting part may be the combination of all three:
Understand me → Make a decision → Take action.
That's what makes an AI agent fundamentally different from a search engine.
A search engine helps me find options.
An agent is supposed to help me choose.
And eventually, act.
We keep asking:
“Can AI plan the perfect trip?”
I'm not sure that's the goal.
I don't want a perfect trip according to the internet.
I want a trip that feels like me.
I want an AI that knows I would rather walk through a quiet neighborhood than visit another crowded attraction.
That I care about a great restaurant more than checking another landmark off a list.
That sometimes the best recommendation is the one that saves me time.
And perhaps most importantly:
I want an AI that knows what I don't want.
Because sometimes knowing what I hate is more useful than knowing what I like.
This is also why I find the development of AI travel infrastructure so interesting.
As AI agents become better at understanding travelers, they need access to increasingly reliable, real-world travel data and actions.
That's one of the problems we're working on at RollingGo.
Through our Hotel MCP, AI agents can access hotel inventory and information across 2M+ properties worldwide.
But the bigger idea isn't simply “more hotels.”
It's giving AI agents the infrastructure they need to move from talking about travel to actually doing travel.
Because ultimately, I don't think the future traveler will want an AI that gives them 30 hotel options.
They'll want one that says:
“I know what you like.
I know what you hate.
I know what matters to you.
And I found the one that makes the most sense.”
That's a very different kind of travel agent.
And perhaps that's where AI travel is really going.
Not toward the perfect trip.
Toward the trip that feels like yours.
What would you rather have?
An AI that knows everything about a destination—
or one that knows everything about you?