Yesterday

Order Me a Room

Fliggy's open-source "flyai" travel skill is now live in Huawei, Xiaomi, OPPO and Honor native skill stores, enabling AI-driven bookings without a user ever opening the Fliggy app, just as Chinese AI orders surged 800% during Spring Festival.

By Terence Ronson, Founder & Managing Director, Pertlink Limited

China

Photo by Pertlink Limited

TL;DR

Fliggy’s open-source “flyai” travel skill (flights, hotels, attractions, ground transport, packages — no account needed) is moving from a GitHub developer tool into the native skill stores of Huawei, Xiaomi, OPPO and Honor.
It’s already transacting, not just planning: AI-driven orders on Fliggy surged 800% during Spring Festival, with conversion rates now at parity with traditional OTA channels — and this lands two weeks before China’s “Super Golden Week” (27 Sept–12 Oct).
It sits inside an unresolved three-way fight for control between AI model makers, phone/OS makers and service platforms — and Fliggy’s move is a hedge across all three rather than a bet on one, the strategically literate play while that fight stays unsettled.
Implication for this region: any hotel, DMO, or OTA relationship assuming a Chinese guest arrives via a familiar app search is now standing on a shrinking assumption — the same shift “Never Mind the Search” flagged for Google, just further along and one layer deeper.
Action headline: audit whether your property has any machine-readable presence beyond OTA listings (schema, MCP-compatible feed, clean API) — “we’re on Fliggy and Ctrip” is an app-era answer to a skill-store-era question.

This desk has tracked Fliggy’s shift from online travel agency to “omni-intelligent travel agent” since it was first announced, and it would be easy to read this latest move as more of the same. It isn’t. There is a real difference between a developer skill sitting on GitHub and the same skill sitting in the phone's default skill store in your guest’s pocket, and last week that difference became a fact.

THE THING THAT ACTUALLY CHANGED

Fliggy’s “flyai” — an open-source travel skill built on Fliggy’s own Model Context Protocol standard, covering flights, hotels, attractions, ground transport, leisure activities, passes and packages, requiring no account setup — launched in March. Within days, it had passed 100 GitHub stars and ranked among the top travel skills by install volume. It already integrates with the current wave of Chinese agent platforms and assistants — Kimi, QClaw, MaxClaw, ArkClaw, QoderWork, Wukong. What’s new is the distribution: Fliggy is now rolling flyai out to the native skill stores of Huawei, Xiaomi, OPPO, and Honor — the four device makers that, between them, put an AI agent into the hands of most of the Chinese smartphone market.

That is the sentence to sit with. Alibaba no longer needs a traveler to open the Fliggy app or even think of Fliggy at all. It needs the phone’s own assistant to reach for flyai when the guest says, in whatever language and however casually, “book me a room.”

WHY THIS ISN’T HYPOTHETICAL

Two numbers make this concrete rather than aspirational.

First, the transaction layer already works: Fliggy’s AI interface, powered by Alibaba’s Qwen App, now processes real bookings, and during the last Spring Festival its AI-driven orders surged 800% year-on-year, with attraction-ticket bookings up more than 24-fold. Conversion rates for air tickets, train tickets, and attractions are now reported on par with the traditional OTA interface.

Second, the timing: China’s next travel peak — the “Super Golden Week” running roughly 27 September to 12 October this year — is two weeks away, and Ctrip is already forecasting inter-provincial travel orders up 45% year-on-year, with Tuniu reporting a 200% jump in searches for “least crowded places” during the holiday.

Put those together, and you get something worth naming plainly: the same infrastructure now being used to discover where to go is being used, on the same platform, to complete the transaction that gets someone there. This closes a loop this desk described from the demand side in “The Klook Signal,” and in Klook co-founder Ethan Lin’s recent argument that overtourism is a concentration problem AI can help solve by redistributing discovery. Fliggy’s flyai is the fulfillment half of exactly that mechanism — and it now sits one voice command away on the device, during the exact week when Chinese travelers are actively trying to avoid the crowds.

THE THREE-WAY FIGHT THIS SITS INSIDE

None of this is happening in isolation, and it’s worth being honest about how unsettled the layer underneath it still is. Chinese commentary on the smartphone AI agent market has settled on a useful frame: a three-way fight for control between the model providers (Doubao, StepFun, Qwen), the hardware and OS makers (Huawei, Xiaomi, OPPO, Honor, Nubia), and the service platforms that actually fulfill a request (WeChat, Douyin, the payment rails underneath both). No single company currently controls all three layers, and the illustrative example doing the rounds is a coffee order: when a user says “order me a coffee,” the agent understands the instruction — but does it route to Luckin, to Starbucks, or to the independent café down the street? Whoever controls that routing captures the distribution fee and the customer relationship. Swap “coffee” for “hotel room,” and you have this desk’s entire distribution thesis in one sentence.

Fliggy’s answer to that unsettled fight is structurally smart, and worth naming as strategy rather than luck: rather than betting on which model or which OS wins, flyai is being made available across all of them — every major agent platform, now every major device skill store. Alibaba doesn’t have to win the phone war. It only has to be the travel answer every side of that war reaches for. That is TCPG’s migration logic taken to its logical endpoint: the cost — and the value — of serving a guest keeps moving toward wherever the decision actually gets made, and the decision is now being made a layer earlier than the OTA search, earlier even than opening an app at all.

It would be a mistake, though, to write this up as inevitable and stop there — this is exactly the overclaiming this desk has argued against elsewhere. The device-level agent market itself is genuinely fragile right now. China’s smartphone shipments fell 4.3–6.7% year-on-year in the second quarter. OEM-native assistants show 755 million monthly active users but declining engagement against standalone apps (51.4 monthly opens versus 92.7). Analysts are openly skeptical that “core capabilities have yet to deliver a major leap,” that on-device models and compute remain insufficient, and that walled-garden ecosystems mean the mass-market case isn’t proven yet. Huawei’s HarmonyOS 7, by contrast, is a genuine counterpoint to that skepticism — over 2,000 specialized agents, a claimed 90% success rate on multi-step tasks, and a Mate 80 flagship arriving in late October to press the advantage against Apple just as HarmonyOS’s China OS share (19%) has pulled ahead of iOS (16%) for the first time. The picture is uneven, not settled — which is precisely why positioning across every layer, as Fliggy is doing, is the more defensible bet than picking a side.

WHERE THE REST OF THE MARKET SITS

Fliggy isn’t moving alone, and the competitive picture matters for how much runway this has. Ctrip has focused its AI investment on multilingual customer service ahead of the inbound season. Meituan has launched an AI agent, internally called “Jibai,” aimed at hotel operational efficiency rather than the discovery layer. JD.com is running an integrated “AI + Robot + Eat + Stay” model, closer to “The Second Turndown”’s physical-AI thesis than to a distribution play. Fliggy has also partnered with Amap’s travel assistant “Teacher Xiao Gao” for personalized food, hotel, and attraction recommendations, and has its own “Ask Me” multimodal assistant that layers cultural storytelling on top of the transaction. None of the other major Chinese platforms are yet making the specific move flyai represents — putting the travel-transaction skill directly into the operating system layer, ahead of and independent of their own app. That is a genuine first-mover position on the exact layer this desk has been arguing will matter most.

WHAT THIS MEANS BEYOND CHINA

Two reasons this belongs in an Asia-Pacific hospitality briefing rather than a China-desk footnote.

First, source-market exposure: China remains one of the largest outbound travel markets for the region, and any property, DMO, or OTA relationship built around the assumption that a Chinese guest arrives via a familiar OTA search is now built on a shrinking assumption — flyai’s own use cases explicitly include corporate travel management agents and lightweight consumer apps built by third parties with no payment infrastructure of their own, which is precisely how this kind of skill ends up embedded in travel products your hotel has never heard of.

Second, the pattern itself is portable. Google’s agentic hotel booking inside AI Mode, covered in this desk’s “Never Mind the Search,” is the same shift happening one operating system over. The device-level agent race in China is simply further along, and further down the stack, than its Western equivalent — which makes it the leading indicator, not a separate story.

ACTION POINTS

FOR MARKETERS AND REVENUE LEADERS: audit whether your property has any structured, machine-readable presence beyond your OTA listings — clean schema, an MCP-compatible feed, an API a third-party agent-builder could actually plug into. “We’re on Fliggy and Ctrip” is a channel strategy for the app era, not the skill-store era.

FOR C-LEVEL AND DISTRIBUTION TEAMS WITH CHINA-OUTBOUND EXPOSURE: treat the next four weeks — Golden Week itself — as a live test rather than a slow news period. Watch where AI-driven bookings originate, if your systems can even tell you that yet; most can’t, which is the finding itself.

FOR BOARDS AND INVESTORS: read flyai’s multi-platform hedge as the strategically literate move in an unsettled market, and read the OEM engagement numbers (755 million users, falling opens) as the reason not to over-rotate capital into any single device partnership yet. The right posture is presence across the layer, not a bet on which corner of it wins.

FOR GOVERNANCE-MINDED OPERATORS: this is a Decision Envelope question before it is a marketing one. If a device-native agent can complete a booking on a guest’s behalf with no human in the loop on either side, your own AI Override Rate discipline needs to extend to bookings you never see coming through a channel you didn’t build.

The intelligence may be artificial. But the experience is human.

Made with the help of various AI tools — but always with a HITL

WHERE THIS CONNECTS

The Alibaba/Fliggy Viewpoint papers and “The AI’s Have It” — the omni-intelligent travel agent thread this continues directly
“Never Mind the Search — You Need to Understand the Context” — the same decision-layer migration, one operating system over
“The Klook Signal” and Ethan Lin’s overtourism-as-discovery-problem argument — the demand-side half of the loop; flyai now closes on the supply side
The TCPG series — cost- and value-migration logic applied to its logical endpoint: the device layer
“The Second Turndown” — JD.com’s “AI + Robot + Eat + Stay” model as a physical-AI comparator
“Who Controls the Controls?” and the Decision Envelope / AI Override Rate frameworks — governance implications of bookings completed with no human in the loop