WeChat, the messaging app that runs the daily lives of more than a billion people in China, has opened its doors to AI agents. The company said it is letting mini-program developers build agents that act on users’ behalf, booking rides, ordering food, planning trips, with Didi, Meituan, Ctrip and Tongcheng among the first to plug in.
In the same release, WeChat added a search feature for Moments, its Facebook-like feed: users can now search the text, images and comments inside their friends’ posts. The two announcements, one about machines acting, one about people searching, amount to the biggest product push Tencent has made in the AI race.
The move turns WeChat’s 1.3 billion users into a distribution channel for AI agents, the largest such opening anywhere in the world. No Western platform has an equivalent: agents in the United States and Europe are still sold as apps to install, while WeChat’s agents will live inside a conversation that already has a payment system attached.
How it works follows the platform’s established logic. Developers build agents inside mini-programs, the lightweight apps that turned WeChat into an operating system a decade ago. Users invoke an agent the way they send a message, and the agent carries out the task, with payments running through WeChat Pay and receipts returning to the chat thread.
The design matters as much as the scale. An agent that books a hotel needs your identity, your payment, your location and your permission, and WeChat has all of those in one account, verified, trusted and already in daily use. That combination is the reason developers fought for early access, and the reason the first partners are the giants of Chinese services.
Tencent’s AI strategy has been cautious in public. Its Hunyuan models have been well regarded but quietly marketed, and the company has said it prefers to integrate AI where it already has distribution rather than chase headlines. The agent opening is the most concrete product of that philosophy, and it arrives with the scale of the company’s flagship app behind it.
The Moments search feature is the privacy-sensitive part. Searching text, images and comments inside friends’ posts touches the kind of data that regulators in other countries would scrutinize for years, and WeChat has framed the feature as user-controlled, with results limited to posts a user can already see. The framing does not change what the feature collects, and privacy researchers are watching.
For developers, the opening changes the economics of building. Mini-program developers have spent years optimizing for discovery inside WeChat’s search and recommendation flows; agents upend that model, because users will describe what they want instead of browsing for it. The developers who adapt fastest are the ones who build agents that answer first and surface options second.
The competitive response is already visible. Alibaba’s Alipay and ByteDance’s Douyin are both super-apps with their own payment rails, and both have announced agent plans of their own in recent weeks. The agent entry point is becoming the new battlefield of Chinese consumer technology, and WeChat’s head start is measured in months, not years.
The regulatory question hangs over the whole category. Chinese authorities have been building rules for generative AI for two years, and agents add a new layer: an AI that spends money, books travel and signs up for services on a user’s behalf is a different risk than a chatbot that answers questions. Beijing has signaled it wants the agent economy to grow, with guardrails, and WeChat’s rollout will define what the guardrails look like.
Tencent’s board is betting the agent layer deepens the moat rather than dilutes it. For a decade, WeChat’s genius was the mini-program: an entire internet inside an app. The agent opening is the same play one level up, an internet that answers instead of waiting to be browsed.
The first agents are utilitarian, ride-hailing, food delivery, hotel booking, and the early demos show the limits as well as the promise. An agent that books the wrong flight or orders from the wrong restaurant creates a dispute that lands in WeChat’s customer service, and the platform will learn, the way it learned with payments, that trust is built one resolved complaint at a time.
For investors, the opening is a reason to revalue Tencent’s AI position. The company has been treated as an AI laggard relative to its valuation, with its models seen as competent but not frontier. Distribution is the argument now: frontier models can be rented, but a billion users with payment rails and verified identities cannot.
If it works, Tencent won’t just own the app; it will own the conversation, and the conversation is where the agents live.
