WeChat and Alipay both enter the scene: Agents have begun transforming the mini-program ecosystem.
```
Author | Huang Yu
In the past few years, the most attention in the wave of large models has always been on AI-native applications.
Chatbots, AI search, AI office tools, and AI browsers are emerging one after another. The market once believed that the next super entry point would arise from new applications, just like it did during the mobile internet era.
Now, the industry trend has shifted. Simple Q&A interaction and text generation can no longer sustain long-term competitiveness. The AI race has officially moved from "competing in dialogue ability" to "competing in practical application," and the richness of the service ecosystem has become one of the key decisive factors.
In this context, two major national-level applications have officially joined the Agent battlefield.
After the internal testing of the AI version of Alipay started on June 16, on June 20, WeChat's AI assistant Xiaowei also began limited gray-scale testing. Some iPhone users can already experience it.
At around the same time, WeChat and Alipay have both embedded AI assistants within their apps—not just offering search, Q&A, or an independent tool, but trying to insert AI into users' core usage paths: “Help users do work with one sentence.”
This means that AI is moving from standalone capability toward infrastructure. Relying on massive user data and service ecosystems, super apps are starting to first understand user needs and then organize services. More importantly, WeChat and Alipay are turning AI into a new service distribution layer.
When the most mature platforms actively restructure interaction methods, a new stage may have already begun.
01 AI Transformation of National Applications
Embedding AI into super apps comes down to how AI capabilities are integrated into existing ecosystems and how they change user interaction.
From a product positioning perspective, the AI assistants of WeChat and Alipay are both designed to be connectors within the original application ecosystem, invoking content and various service capabilities.
In terms of product design, compared with the “biggest overhaul in history” of Alipay, WeChat AI has remained as cautious as ever, with a philosophy of “not disturbing users.”
The entrance to WeChat's AI assistant Xiaowei is currently located in the top left corner of the main page, with an eye-like icon. You can click to enter or swipe right from the main WeChat page to directly chat with Xiaowei. The interface is simple. Xiaowei’s entry is also embedded in many core WeChat scenarios, such as a "Ask Xiaowei" shortcut in the chat dialog plus-sign sidebar, or long-pressing chat content or reading public articles to invoke "Ask Xiaowei."
These entry paths continue WeChat's product philosophy: changing user habits as little as possible, making AI a new layer of interaction.
Alipay currently retains the old interface, but when users swipe right or click the bottom left corner to open the AI version, almost all native Alipay functions are integrated. The interface is greatly simplified compared to the old version, with only two feature pages: Abao and Assets.
Therefore, it’s possible Alipay may support users setting the AI version as the default page in the future.
Despite some differences in product design, both WeChat and Alipay are testing whether users can shift from “clicking services themselves” to “letting AI get services for them.”
Currently, Xiaowei supports operations for native WeChat functions and calling mini-programs via text or voice conversations, such as sending messages to friends, checking Moments, invoking mini-programs for services, setting reminders, recommending music, etc., but purchases still require user confirmation and payment.
In the gray-scale test, Xiaowei has an intriguing feature: “One-sentence tool generation.” Users can describe their needs in natural language, and Xiaowei can directly generate a basic-function mini-program prototype, then further adjust it based on multi-round dialogue.
Currently, the generated mini-tools can only be used by the user and are not yet shareable with others.
This idea is also being promoted by Huawei devices. At the recent Huawei Developer Conference, AllWeatherTech learned that Huawei devices provide “one sentence generates meta-service” capabilities. Meta-services are mini-programs running on the HarmonyOS.
Huawei product personnel told AllWeatherTech that this is mainly to lower the threshold for developing mini-program products, so ordinary consumers can develop personalized needs that mainstream apps can’t satisfy, and it could also be prototyping for promising mini-programs.
It is noteworthy that Xiaowei’s main model did not pick Tencent’s previously promoted self-developed “Hunyuan” large model, but chose WeChat’s internally developed Chinese large language model “WeLM”; some answers use DeepSeek.
Thus, for embedding AI Agents into the WeChat ecosystem, the WeChat team maintains a pragmatic attitude of "controllable + scenario adaptation."
02 Competing for the New Distribution Rights
If the core of large model competition in the past two years was parameters, reasoning and generation capabilities, after entering the Agent stage, the rules are changing.
The industry has increasingly reached consensus: what determines the ultimate Agent experience is not just model capability, but how much of the real world is connected behind the model.
If a user says "Help me order a cup of milk tea with no extra sugar," it requires calling up delivery, payment, fulfillment, account, risk management, message notification, and many capabilities. If a user requests "Help me check this year’s social security records," then social security and identity capabilities are needed.
At this stage, the agent's upper limit is determined by the richness of the platform ecosystem. And when some AI manufacturers try to connect to national-level apps, they are often resisted because of data security and business interests.
This is why, as AI enters the real-world phase, national-level apps have regained their advantage.
With five ecosystems including communication, social, content (public accounts, video accounts), mini-programs, payment and commerce, WeChat AI was seen as Tencent’s differentiated product in this AI competition even before release.
Thanks to its user relationship network and huge service supply network, when the Agent starts executing tasks, WeChat naturally has service orchestration space—it can understand demand and mobilize internal resources to act.
QuestMobile data shows that in March 2026, monthly active users for WeChat, Alipay, and Douyin Mini-programs were 973 million, 644 million, and 273 million respectively.
The mini-program ecosystem is indeed rich, but the first challenge for AI ecosystems built by national apps is whether they can attract enough developers.
WeChat has published clear guidelines for developers to connect to the WeChat AI ecosystem, giving the decision to developers and offering both auto and development modes for those who opt in.
WeChat specifically notes that these two modes are not mutually exclusive and can be enabled at the same time. If developers disable both modes, WeChat AI will not recommend or call that mini-program, but existing mini-program services are unaffected.
At this stage, mini-program developers are likely more willing by default to join the platform AI ecosystem than independent app developers are to grant general AI access.
The main reason is the difference in control. Independent apps have always operated under a “self-managed” logic. Once general AI is granted access, developers worry about three things: Who owns the user relationship? Who decides recommendation order? Who controls transaction data?
But the mini-program ecosystem was built on a different logic from day one.
Developers accept that the platform handles distribution, search, payment, and user acquisition. In a sense, mini-programs are a “semi-hosted ecosystem”. Developers relinquish some control in exchange for platform traffic, tool abilities, and lower customer acquisition costs.
When the platform launches Agent capabilities, mini-program developers’ psychological cost is lower. For long-tail developers, this could be an opportunity.
Of course, what truly determines developer willingness is whether the added value is real. Only if AI brings new demand, new calls, new conversions, will developers proactively connect.
Tencent president Liu Chiping said in the Q1 2026 earnings call that the WeChat platform naturally has multiple advantages to support AI Agents, but Tencent is still exploring the best way to present these capabilities and engage mini-program developers.
03 Challenges in Reshaping Mini-programs
Whether developers are willing to connect to the platform AI ecosystem is only the first barrier to Agent landing.
The real change happens in the distribution mechanism. Previously, users searched, clicked, and chose mini-programs. In Agent mode, the platform interprets needs and screens services for users.
So, when massive services enter the ecosystem, the key new problem for Alipay and WeChat in the Agent era is how to understand these services and dispatch the most suitable abilities to users.
A user saying “Find me the best restaurant tonight for a family dinner” actually involves a lot of hidden conditions: budget, distance, number of people, historical preferences, membership rights, fulfillment capability, inventory status.
If the platform cannot accurately understand these service capabilities, a typical problem arises—AI seems able to do anything, but its recommendations end up being untrustworthy.
This means the core ability in the mini-program era was connection. In the Agent era, the core ability becomes semantic understanding and service orchestration.
The platform needs to answer at least three things.
First, it must know exactly what each service can do.
Previously, developers just uploaded a page. In the future, a standardized description might be needed: supported scenarios, actions, inputs/outputs.
Second, the platform must establish a new scheduling mechanism.
Distribution used to rely on search ranking and recommendation algorithms. In the future, an Agent scheduling system is needed: which service to call when; how multiple services collaborate; how to fall back on failures; how to evaluate results.
As the platform takes on intent understanding, service matching, and decision making, the business model built by the mini-program ecosystem will change.
The developer ecosystem may be affected first. Previously, success for mini-programs was getting enough exposure; in the future, competition might shift to whether AI calls them first.
The more standardized the data structure, the fuller the service capability, the more stable the interface response—the easier it is to get entry.
The ad system may be redefined. Traditional ads are built on display logic, but AI naturally pursues fewer choices.
If users only see one AI-recommended result, traditional exposure slots shrink. Developers now compete for call ranking, not page position. Perhaps platforms will no longer sell exposure, but recommendation rights.
In this new round of ecosystem war, whoever establishes a reasonable scheduling mechanism first is more likely to keep both supply and demand within their ecosystem.
Third, the platform must solve the trust issue.
Before, users clicked into a page and made their own decisions. If the platform now chooses directly for users, it assumes partial decision responsibility. If the AI recommends wrongly, who’s liable—the platform, the model, or the developer?
Of course, whether users are willing to hand over decision rights, whether developers accept the new distribution system, and whether the platform finds a new business cycle—all remain variables.
This battle, on the surface, is WeChat and Alipay competing for the AI entry point. In reality, it's a contest over who qualifies to understand user needs, select services, and complete transactions for the user.
Risk Warning and DisclaimerThe market involves risk; investment should be cautious. This article does not constitute personal investment advice, nor does it take into account the unique investment targets, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article suit their specific circumstances. Invest accordingly at your own risk. ```