In June 2026, news that WeChat’s embedded AI agent “Xiaowei” had entered gray-scale testing immediately ignited capital market enthusiasm — Tencent Holdings surged 10.46% in a single day, with market capitalization increasing overnight by more than HKD 400 billion (approximately USD 51.28 billion). Before this, ByteDance’s Doubao had surpassed 300 million monthly active users, Alibaba’s Qianwen was steadily capturing share in the AI services track, and Tencent’s own standalone AI product Yuanbao had consistently struggled to break out in terms of user scale, with the market at one point labeling Tencent’s AI positioning as “falling behind.”
With the rollout of the small green eye icon on WeChat’s homepage and the swipe-right gesture to summon the intelligent dialogue window, everyone has finally seen Tencent’s new strategy clearly: abandoning the race among standalone AI applications, and instead deploying WeChat — the national-level super app with 1.4 billion monthly active users — to go all-in on the AI Agent track. This is not simply the launch of a chatbot. It is a foundational restructuring of WeChat from “people actively seeking services” to “AI handling chores on people’s behalf.” It is a super-entrance defense war aimed at ByteDance and Alibaba, and even more so, an industry-wide reshuffling that will determine the fate of millions of mini-program developers.
A Major Strategic Pivot: Moving Beyond Yuanbao’s Solo Struggle, WeChat Takes Up the AI Main Battlefield
As early as 2023, after ChatGPT ignited a global AI wave, the industry had continuously anticipated WeChat’s AI answer. For a long stretch of time, WeChat maintained restraint, only lightly embedding Yuanbao traffic referrals on its search and contacts pages, without launching an AI tool deeply integrated into its native ecosystem. The group poured resources into supporting Yuanbao — Spring Festival red envelopes, full-platform traffic allocation, one wave after another — yet the data gap remained difficult to close. March 2026 data showed Doubao with 345 million monthly active users and Qianwen with 167 million, while Yuanbao had only 57.35 million — a stark gap that directly gave rise to what Pony Ma himself called “AI anxiety.”
Pony Ma has openly acknowledged that Tencent’s early AI deployment hit roadblocks — substantial resources were invested without successfully building a complete service closed loop. Standalone AI applications carry an inherent fatal weakness: users must actively download and deliberately open them, vendors must continuously spend on marketing to cultivate user habits, and the “use once and leave” nature makes it difficult to accumulate long-term service transactions.
Standing in stark contrast is the product philosophy Allen Zhang has maintained for years: WeChat does not build standalone AI software; AI should function like “Scan,” seamlessly integrated into scenarios, used as needed, and quietly exiting once the need is resolved. Based on this thinking, WeChat secretly launched an Agent project internally codenamed “Xiaowei,” designated as the group’s highest priority, refined over nearly a year. Product functionality and multi-scenario scheduling pathways have already been fully validated, awaiting only the completion of regulatory compliance approval before a phased gray-scale rollout.
The birth of Xiaowei marks a complete change of direction in Tencent’s AI strategy. Rather than relying on Yuanbao alone to break through, Tencent is transforming WeChat — the super infrastructure that has accumulated social connections, mini-programs, and payments — into a native AI operating system. An application opened with high frequency every day by 1.4 billion users inherently possesses a traffic advantage requiring no additional customer acquisition cost; AI is naturally embedded across chats, group chats, Channels, and mini-programs in every scenario, cultivating user mindshare imperceptibly — a moat that no standalone AI product can replicate.
The A2A Exclusive Technical Barrier: A Decisive Edge Over the Industry’s Mainstream Screen-Reading GUI Approach
Mainstream smartphone manufacturers and ByteDance’s Doubao phone have adopted a GUI graphical agent route. WeChat has chosen a different path, building a proprietary A2A (Agent to Agent) institutionalized collaboration protocol, constructing an encrypted dialogue channel between machines. Once a user issues a request to Xiaowei, intent instructions flow and execute natively within WeChat itself, without needing to capture the external screen — all service invocation permissions remain entirely under WeChat’s control. This model balances the interests of multiple parties: phone manufacturers gain a system-level AI experience selling point, merchants retain control over their own business interfaces, and user privacy and security are protected to the greatest extent possible. This is also the core reason why leading service providers such as Meituan, JD.com, Trip.com, and Didi have proactively connected to the Xiaowei ecosystem, while collectively resisting screen-reading-style AI.
To accommodate its vast number of mini-programs, WeChat has launched two developer integration modes, which can be activated simultaneously. The zero-threshold automatic mode requires only one-click authorization on the backend; the platform automatically parses page logic to generate AI-callable skills, allowing millions of mini-programs to complete AI transformation quickly — suitable for check-in tools, queries, and simple lightweight applications. The high-controllability development mode requires developers to write standardized skill documentation and interfaces, requiring some development investment, but allows developers to independently define the boundaries of AI operation — suitable for high-compliance scenarios such as healthcare, finance, and online transactions.
A key rule in the official guidelines directly determines the shape of the industry: mini-programs that have not completed integration will be unable to be searched, scheduled, or recommended by Xiaowei. WeChat is using a standardized protocol to push the entire industry to pick a side, transforming all 4 million existing mini-programs into a service toolbox that AI agents can call upon at any time.
Short-Term Profit Pressure, Long-Term Defense of the Super-Entrance Moat
From a financial perspective, the rollout of Xiaowei will continue to weigh on Tencent’s profits in the short term. The group’s total capital expenditure for 2026 is expected to reach RMB 150 billion to 200 billion (approximately USD 20.83 billion to USD 27.78 billion); quarterly losses related to AI business are already close to RMB 9 billion (approximately USD 1.25 billion); computing power training and inference deployment will continue to drive up costs, and there is a risk that net profit growth in 2027 could stagnate or even decline. However, the market has already fully digested expectations of short-term losses — the core of the capital-market wager is whether this massive investment can translate into an unshakeable ecosystem moat.
Xiaowei’s core value lies not in direct monetization of itself, but in the spillover effect it generates across the entire ecosystem. AI-driven natural language distribution widens exposure channels for mini-programs, continuously driving WeChat’s advertising revenue; one-stop conversational transaction fulfillment increases the frequency of online transactions, steadily raising WeChat Pay’s commercial GMV; lower-tier markets and the silver-haired population will become a core source of incremental growth, as elderly users who struggle to navigate complex, multi-layered interfaces can simply state their needs to Xiaowei in a single sentence, unlocking large amounts of previously dormant consumer demand.
Looking deeper, Xiaowei is fundamentally an extremely costly but necessary defense of the entry point. Tencent’s true competitor has never been Doubao, whose user base remains limited and prone to “use once and leave”; it is Douyin, with daily usage time exceeding two hours and a continuous grip on user attention. If a competitor is first to establish a fully connected, full-domain AI service chain, the foundational base of service entry points that WeChat has accumulated over a decade would suffer a severe blow. Even if Xiaowei cannot generate substantial revenue in the short term, as long as it secures the interaction entry point for 1.4 billion users and blocks the possibility of competitors overtaking through a shortcut, this investment carries long-term value.
Of course, Xiaowei’s rollout still faces multiple real-world risks that cannot be avoided. A user base in the billions brings massive computing power pressure, and low-latency, high-concurrency stable deployment tests the limits of technical capability. Agent invocation requires multi-dimensional data including location, consumption, and social information, with privacy compliance red lines continuously constraining the ceiling of functionality. There is, as of yet, no globally successful precedent for a nationwide-scale agent handling everyday tasks for the entire population — model hallucinations and failures in multi-step tasks could very easily trigger large-scale negative public sentiment.
From social connections between people, to intelligent connections between people and services, every foundational iteration WeChat undergoes reshapes the rules of the internet industry. A decade ago, mini-programs bridged online and offline services; today, the AI Agent Xiaowei is restructuring the logic of traffic and interaction.
For Tencent, Xiaowei is the key lever for reversing its lagging position in AI. For millions of mini-program developers, this represents both an opportunity and a crisis in a coming reshuffle. For ordinary users, an AI butler will fundamentally simplify the tedious fingertip operations that fill their phones.
This all-out AI offensive led by WeChat has only just begun. Short-term computing power losses and compliance restrictions are temporary growing pains; in the long run, whoever controls the scheduling entry point from user demand to fulfillment will hold the power to define the next era of the internet. The window of opportunity closes before the full gray-scale rollout in Q3 2026 — those who integrate early will seize the advantage, while tool products that fail to establish differentiated barriers will ultimately be swept away by the AI wave.
[Disclaimer]: The above content reflects analysis of publicly available information, expert insights, and BCC research. It does not constitute investment advice. BCC is not responsible for any losses resulting from reliance on the views expressed herein. Investors should exercise caution.