WeChat Rolls Out Biggest Update Ever as Alipay Simultaneously Unveils Major Redesign
N.R. Finch
WeChat's "Xiaowei" assistant and AI-powered Alipay both entered beta testing in June, as China's two super-apps upgrade AI from a Q&A tool into a service-dispatch layer — rewriting how mini-programs are distributed, how ads work, and how developers compete.
Why are both moving at the same time?
In the AI Agent era, the ceiling on user experience is no longer the model itself — it is how many real-world services the model can reach.
WeChat spans messaging, social, content, mini-programs, and payments, with 973 million monthly active mini-program users; Alipay's mini-programs reach 644 million.
This means → both own a service-orchestration space that standalone AI firms cannot replicate. A powerful model with nothing to dispatch is just a chatbox.
How do their product approaches differ?
WeChat's "Xiaowei" is tucked behind the top-left corner of the home screen — swipe right to open it — staying true to WeChat's "don't disturb the user" philosophy. Payment steps still require user confirmation.
Alipay calls its revamp "the biggest redesign in its history": the interface collapses into just two pages — "A-Bao" and "Assets" — and the company may eventually let users set the AI version as default.
In plain terms = WeChat added a side door next to the existing house; Alipay gutted the entire building.
Which models are under the hood?
Xiaowei primarily runs on WeChat's in-house WeLM large language model, with some answers routed to DeepSeek.
Notably, WeChat did not adopt Tencent's flagship in-house model, Hunyuan.
This reflects a pragmatic "controllability + scene fit" stance — the team picks models by use-case match, not by internal corporate hierarchy.
What does "generate a mini-tool in one sentence" mean?
In the beta, Xiaowei lets users describe a need in plain language and instantly generates a working mini-program prototype, with multi-turn dialogue to adjust the design. For now it is personal-use only.
Huawei's device division recently unveiled a similar feature — "generate an Atomic Service in one sentence" (Atomic Services are HarmonyOS's version of mini-programs) — aimed at lowering the development barrier.
This means → mini-program creation is shifting from "a developer writes code" to "an ordinary user says one sentence." The decentralization of *building* power may arrive before the shake-up in *distribution* power.
How will distribution logic be reshaped?
Previously, mini-programs relied on users actively searching and tapping. Under the Agent model, the platform understands the need first, then filters services on the user's behalf.
Developers' competitive focus shifts from fighting for on-screen ad slots to whether AI calls them first — the more standardized the data, the more stable the API, the higher the ranking.
In plain terms = the old game was who had the flashiest billboard; the new game is who has the cleanest back-end API. Users may never even see a second option.
What variables remain unresolved?
User side: How many people will hand purchase decisions to AI rather than browse and choose themselves?
Developer side: The new distribution system concentrates platform power further, but mini-program developers have historically accepted this — the ecosystem was built on a semi-managed "platform distributes, developer delivers" model from day one.
Business-model side: Traditional ads rest on display logic, but AI naturally compresses user choice steps. What platforms may sell in the future is not impressions — it is recommendation authority.
Content is for reference only, not financial advice.