Alibaba-owned travel platform Fliggy is pushing AI travel assistants beyond the familiar “here are five hotels you might like” stage. The new Fliggy agentic AI travel assistant is designed to offer more advanced and personalised travel planning.
Its latest agentic AI travel assistant, powered by Alibaba’s Qwen model, is designed to actually do things.
That distinction matters.
Instead of stopping at itineraries and recommendations, Fliggy says the upgraded assistant can handle bookings, changes, check-ins, seat selections, cancellations and more complicated requests that may involve several steps.
It is a glimpse of where consumer AI may be heading: less chat, more execution.
Fliggy Wants AI to Handle the Messy Parts of Travel
Travel planning sounds simple until it isn’t.
Finding a flight is easy enough. Changing it after the itinerary shifts, finding a replacement hotel, checking availability, arranging a room upgrade and keeping track of everything afterward? That is where the friction starts.
Fliggy’s new assistant is built around those situations.
A traveler can explain what they want using normal language through text or voice. The assistant can then work with Fliggy’s transaction systems and travel inventory to carry out the request.
One example given by the company is a hotel room upgrade. Rather than simply suggesting that the traveler contact the hotel, the AI can communicate with the property, check availability, coordinate the request and return with updates.
That is a very different product from a chatbot that generates a travel checklist.
Planning Is Only One Part of the Assistant
Fliggy has divided the experience into three broad functions: Go Think, Go Book and Go Sort.
Go Think handles the early stages — destination ideas, routes, budgets and itinerary creation based on available inventory.
Go Book moves closer to the transaction. It can compare flights, hotels and tours before creating orders.
Then comes Go Sort, which deals with the parts travelers often discover are harder to automate: check-ins, reservation changes and other services during or after a booking.
The point is not to force travelers into an AI-only interface either.
Fliggy says the assistant works alongside its existing visual booking experience. Someone who wants to manually browse hotel listings can keep doing that. Someone who would rather tell the AI what needs to happen can delegate the task.
That hybrid approach may prove important. People tend to be comfortable asking AI for ideas. Handing over control of a paid reservation is another matter entirely.
Memory Could Make the Travel Agent More Personal
Fliggy has also given the assistant stronger memory capabilities.
Rather than treating every trip as an isolated conversation, the system is designed to combine longer-term preferences with whatever the traveler needs at that moment.
A person who consistently prefers certain hotel types, flight schedules or travel styles could theoretically spend less time repeating those preferences with every new trip.
That sounds small. In practice, memory may be one of the features that separates useful personal agents from glorified search boxes.
Travel is unusually suited to this because preferences repeat. A traveler might prefer aisle seats, avoid early departures, stay near public transportation or consistently choose hotels within a particular price range.
Remembering those patterns gives an AI agent more context before it starts searching.
Qwen Sits Underneath the Travel-Specific System
The assistant runs on Alibaba’s Qwen AI model, but Fliggy is not simply placing a general chatbot on top of a travel website.
The company says the system also uses its real-time transaction infrastructure and proprietary travel data. It has been further tuned for travel scenarios and uses agentic reinforcement learning alongside a dynamic evaluation system.
That evaluation layer is particularly relevant.
A travel recommendation is not useful if the hotel no longer has a room or the flight cannot actually be booked. Fliggy says the system checks the real-world availability of results while also evaluating whether the requested task was successfully completed.
Its architecture has also been streamlined into a multi-agent setup that combines specialized agents with an expanding set of tools.
This is becoming a recurring theme in agentic AI. The language model may understand what someone wants, but useful agents also need access to tools, live data and systems where they can take action.
Fliggy Claims a Significant Jump in Usability
According to Fliggy’s own internal data, the usability score for results generated by the upgraded assistant improved by more than 70% compared with its previous-generation system.
Average completion time for comparable tasks reportedly fell by nearly 10%.
Those numbers come from Fliggy, so independent comparisons will matter as the product reaches more travelers.
The more interesting test may be less measurable.
Will users trust an AI agent enough to modify an expensive international flight? Cancel a hotel? Choose between competing travel options without inspecting every screen themselves?
Agentic AI gets much harder when an incorrect answer stops being text on a screen and becomes an actual transaction.
Alibaba Is Building Toward AI That Acts
Fliggy’s upgrade also fits neatly into Alibaba’s wider AI strategy.
Earlier in 2026, Alibaba expanded its Qwen App with agentic capabilities connected to services including Fliggy, Taobao, Alipay and Amap. The broader idea was straightforward: move Qwen from an AI that answers requests to one capable of completing them.
Travel is a natural place to test that idea.
There are flights to search, rooms to reserve, payments to make, schedules to manage and unexpected changes that create chains of smaller tasks.
Alibaba has also been extending Qwen beyond its own ecosystem. An integration with China Eastern Airlines, for example, was designed to let users handle activities including ticket searches, booking, seat selection and check-in through conversational interactions.
Fliggy now takes that concept deeper into the travel platform itself.
The Bigger Story Is What Happens After the Chat
For the last few years, much of consumer generative AI has revolved around producing things: answers, summaries, images, plans, emails.
Agentic systems introduce a more consequential question.
What happens when the AI gets permission to act?
Fliggy’s assistant is interesting because its usefulness is not really measured by how impressive its travel advice sounds. The real test is whether the hotel gets changed correctly. The reservation also needs to actually exist. Seat selection has to work. Most importantly, the traveler should arrive without discovering that the AI misunderstood something three steps earlier.
That is a much tougher standard.
It is also probably where the next phase of consumer AI will be fought.
The chatbot era taught AI to talk.
Companies such as Alibaba are now trying to teach it to finish the job.
Sources
- Breaking Travel News — “Fliggy unveils its next-generation agentic AI travel assistant”
- Alibaba Group — “Qwen App Advances Agentic AI Strategy by Turning Core Ecosystem Services into Executable AI Capabilities”
- Alibaba Group — Qwen App and China Eastern Airlines agentic travel integration
- Fliggy AI Open Platform — Travel AI capabilities and real-time travel search
