WeChat has a growth problem. Its MAU hit 14.32 billion in Q1 2026, but the year-on-year growth rate is down to 2%. That is basically flatlining for an app that used to add hundreds of millions of users every quarter. Tencent knows this. The company also knows that Douyin just crossed 1 billion MAU with 14.43% growth, and that users now spend 93 minutes per day on Douyin versus 85 minutes on WeChat. The gap is real.
So WeChat is doing what any 14-year-old superapp would do: it is scrambling.
In the past month alone, WeChat shipped PC version 4.1.9 with scrolling long screenshots and voice messages from desktop. It started grey-testing “combined payment,” which lets you stack multiple payment methods in one checkout. It rolled out horizontal video calls with 16:9 widescreen and background blur in iOS 8.0.73. And it killed a rumoured “visitor log” feature for WeChat Status before it even launched, because users freaked out.
But the real story is what Tencent is building behind closed doors.
The secret agent
According to The Information and multiple Chinese tech outlets, Tencent has been running a top-priority, highly confidential project since at least H1 2025 to build a native AI agent inside WeChat. The plan is to launch grey-box testing around mid-2026 and roll it out to all users in Q3.
The pitch is simple: you type “book me a high-speed train to Shanghai tomorrow and find a hotel near the West Bund,” and the agent does it. Not by opening a browser. Not by sending you links. By calling the actual APIs of the millions of mini-programs already living inside WeChat. Train ticket? 12306 mini-program. Hotel? Ctrip or Meituan. Payment? WeChat Pay, already authenticated. The whole thing happens in the background while you keep chatting.
Tencent is calling this “AgentOS” internally. The idea is that the agent becomes the new top layer of the OS, and mini-programs become the backend services it talks to. That is a big architectural shift. Right now, mini-programs are apps you open. In this model, they are skills the agent invokes.
Why this could actually work
Every other AI agent in China has an ecosystem problem. Douyin’s Doubao wants to be your phone’s OS layer, but Meituan and Didi keep locking it out with anti-automation defences. Alibaba’s Tongyi Qianwen has great agent tech, but it can only really move inside Alibaba’s own apps: Taobao, Fliggy, Gaode. Step outside that garden and it becomes a search engine again.
WeChat does not have this problem. It already has the largest collection of third-party services in China, all packaged as mini-programs. Almost every major service, from ride-hailing to grocery delivery to government paperwork, has a WeChat mini-program. WeChat is the rule-maker of that ecosystem. No one can block its agent from calling its own mini-programs.
That is the structural advantage. The agent’s task radius covers most of daily life in China, not just one company’s walled garden.
The catches
There are three obvious problems.
First, the model. Tencent’s own Hunyuan model is reportedly not good enough yet for this level of complexity, so the team is testing external models from Zhipu, Alibaba, and DeepSeek alongside smaller in-house ones. Using third-party models for a product that handles payments and personal data is… complicated. Every data handoff needs legal clearance, and at 1.4 billion users, there is no room for “we will figure out compliance later.”
Second, the partners. If the AI agent becomes the only front door to mini-programs, merchants get pushed to the back. They lose their brand interface, their push notifications, their direct user relationship. Will Alibaba, ByteDance, and PDD keep their mini-programs in WeChat if Tencent’s agent intermediates every transaction? That is a genuine tension. If major platforms pull out, the agent’s coverage shrinks fast.
Third, user trust. WeChat’s whole brand is built on low-pressure social interaction. No read receipts. No visitor logs. Tencent PR director Zhang Jun literally said those two features are “welded shut, will not be developed.” An AI agent that reads your chats, knows your location, and spends your money is a much bigger intrusion than a read receipt. WeChat users are sensitive. Any misstep here could backfire badly.
What else is moving
While WeChat figures out its agent, the rest of the ecosystem is not standing still.
WeChat’s AI Mini-Program Growth Plan is already live for 2026, offering free cloud resources, 100 million Hunyuan 2.0 tokens, and 10,000 text-to-image generations to developers who build AI-powered mini-programs. The goal is to flood the ecosystem with AI-capable services before the agent itself launches. Smart move.
Alipay, meanwhile, is having a moment with “Tap-to-Pay.” Daily transactions crossed 100 million earlier this year, and the company is expanding the NFC tap from payments to door access, food ordering, and package pickup. It is also integrating Qianwen directly into Alipay and Taobao Flash Purchase, so you can order bubble tea by talking to the AI and paying without leaving the chat.
Douyin hit 1.009 billion MAU in March 2026 per QuestMobile, making it the second app in China after WeChat to break 10 figures. Average daily use is now over 90 minutes. It is no longer just a short-video app; it is a commerce, local life, and content platform that happens to serve video.
My take
I have been using WeChat for twelve years. It is the one app I cannot delete. But I also notice I open it less for fun now. It is a utility. I pay bills, I scan QR codes, I reply to work messages. The agent idea could bring some life back to the platform, or it could make the whole thing feel even more like a government services portal with a chat layer on top.
The real question is whether Tencent can make the agent feel optional. If it is a helpful layer you can ignore, great. If it starts inserting itself into every conversation, users will revolt. WeChat’s user base is too big, too old, and too conservative for radical UI changes.
Mid-2026 for grey testing, Q3 for wider rollout. That is the timeline. I will be watching.