Tencent's WeChat AI agent opens Mini Programs to 1.4 billion users

Tencent's WeChat will integrate Mini Programs into its AI system, letting 1.4 billion users order food delivery, book hotels and hail rides through a single AI agent — without leaving the chat app. The integration, announced June 8, marks the most significant step yet in Tencent's plan to build agentic services inside its super-app.

"This testing, together with continuous feedback from the Hunyuan Hy3 model, will provide valuable data for Tencent's next-generation model iteration," Citi analysts wrote in a June 9 report, maintaining a Buy rating and HKD 763 price target. The target implies 23 times 2026 forecast earnings and about 21.7 times 2027 earnings on a sum-of-the-parts basis.

The initial beta includes Meituan, Tongcheng Travel and Trip.com as core partners. Mini Programs that integrate become eligible for recommendation and invocation by WeChat AI; those that do not cannot be accessed. The platform offers two modes — automatic, which accesses source code during review, and development, which lets developers customize integration based on business requirements.

WeChat's 1.432 billion monthly active users and more than 900 million daily active Mini Program users represent the largest addressable market for any AI agent product globally. Citi expects a full rollout in the near term. Tencent shares surged 10.5 percent on June 2, adding more than HK$300 billion in market value in a single session, after reports that the company was testing an AI agent prototype within WeChat.

How WeChat AI Works

The agent opens through a swipe-right gesture on the main WeChat interface, displaying a chat window where users type commands. The AI then automatically triggers Mini Programs to filter options, place orders and complete payments. A user asking for "a coffee under 30 yuan, not too sweet, available for pickup nearby" would trigger the agent to search Mini Programs, match preferences and complete the transaction — all within the chat.

Behind the interface, Tencent developed UI-Oceanus, a world model trained on 5 million simulated interactions to predict what happens when the agent clicks a button, navigates a page or triggers a payment flow. Training directly in live Mini Programs was too slow and unstable, so the model learns in a virtual environment before transitioning to real-world scenarios.

Tencent also faces a cost challenge: triggering inference for every query across 1.4 billion users would be prohibitive. The company is pursuing a multi-model strategy, using smaller models for basic tasks and invoking more powerful models for complex requests. Its rumored 10 billion yuan investment in DeepSeek aligns with this approach — DeepSeek's low-cost inference capabilities suit WeChat's massive scale. On June 2, Tencent said DeepSeek-V4 pricing on its cloud platform matches official rates with no markup.

Why WeChat Holds the Advantage

Tencent's internal Co-Design mechanism funnels capabilities from multiple AI products into WeChat AI. Yuanbao trains multi-turn dialogue and intent understanding on real user queries. Ima builds search capabilities for transforming vague requests into precise queries. WorkBuddy contributes enterprise semantics for document and task understanding. Marvis handles task decomposition and multi-agent orchestration. All these trained skills transfer to WeChat AI through pre-training and post-training generalization.

The company chose the A2A protocol over GUI-based agent approaches, which it has actively blocked. WeChat banned third-party tools that circumvent its security measures in April 2025, and ByteDance's Doubao phone assistant triggered forced logouts when it attempted cross-app operations. Honor became the first smartphone brand to complete WeChat A2A integration, allowing its YOYO assistant to send messages and make calls through voice commands.

Tencent Chief AI Scientist Yao Shunyu, speaking at the Tencent Cloud AI Industry Application Conference on June 5, said AI is a long-term game, not a short-term window. "Practical value outweighs leaderboard value," he said, comparing the current phase to the personal computer era of the 1970s where new product opportunities will continue to emerge.

For investors, the question is whether WeChat AI can generate revenue. Doubao, with 345 million MAU, is reportedly planning to charge for certain features. WeChat's 1.4 billion user base presents even greater cost pressure but also a larger monetization opportunity — from individual users willing to pay for task completion to enterprise customers in automation, customer service and marketing. Tencent trades at about 23 times forward earnings, and a successful WeChat AI rollout could support a valuation re-rating by deepening platform stickiness and opening new service revenue streams.

This article is for informational purposes only and does not constitute investment advice.