3%.

That was Tencent Holdings' late-session gain on the Hong Kong Stock Exchange on June 30. The catalyst: news that WeChat AI Agent "Xiaowei" had begun grey testing, spreading through both tech and investment circles simultaneously. According to beta testers, Xiaowei can access contacts, messages, and Moments; send messages; and invoke Mini Programs -- simply put, WeChat finally has an AI that can "operate WeChat for you."

Close on its heels, WeCom (WeChat Work) was also reported to be launching an AI Agent called "Dayuan", currently in internal beta. Guosheng Securities research noted: "For Agents, ecosystem building is one of the core competitive moats."

WeChat + WeCom dual-track Agent deployment isn't an isolated event. Huawei released HarmonyOS 7 at HDC 2026 on June 12, with Richard Yu declaring it the "first operating system to complete AI-native transformation," formally moving from "Internet of Everything" to the "Agent era," with Xiaoyi becoming a cross-ecosystem super agent. Apple's new Siri AI gained system-level permissions with multimodal screen perception capabilities. Doubao denied rumors of social feature testing but acknowledged collaboration attempts with Feishu in enterprise office scenarios.

WeChat's 1.6 billion monthly active users, plus system-level Agent-native transformations on HarmonyOS, iOS, and Android -- an endgame war for AI entry points is being waged on the screen you unlock dozens of times a day.

Why "Xiaowei" Matters: A WeChat Agent Is a Different Species

There are already plenty of AI assistants on the market, but an Agent inside WeChat is an entirely different species.

The reason is that WeChat isn't an App -- it's an OS-level ecosystem. 1.6 billion users' social connections, payments, Mini Programs, Official Accounts, Video Accounts, and enterprise services are all inside. An Agent that can operate freely within the WeChat ecosystem means it can reply to messages, post Moments, order food, handle work approvals, manage group chats, and invoke Mini Programs for services -- all without jumping to other apps.

A user who tested Tencent's ad-placement Agent put it well: "Regular LLMs can only answer questions; Agents solve problems." The key to WeChat Xiaowei isn't "smarter chat" -- it's actually doing work in real business systems. When an Agent can call Mini Programs, complete payments, and operate WeCom approval workflows, it transforms from a toy into a genuine productivity tool.

WeCom "Dayuan's" logic is even more direct: in B2B scenarios, Agent value is easier to quantify. Auto-replying to customer inquiries, auto-generating meeting minutes, auto-approving routine workflows, auto-analyzing sales data -- these are all scenarios where ROI is directly measurable.

OS Agent-Native Transformation: Who Controls the Entry Controls the AI Era

WeChat building Agents is application-layer entry competition; Huawei HarmonyOS 7 and Apple Siri are system-layer entry competition.

HarmonyOS 7's "Agent-friendly system architecture" treats Agents as first-class citizens -- Xiaoyi is no longer a voice assistant but a super agent capable of cross-device, cross-app orchestration. Tell Xiaoyi "organize yesterday's meeting PPT and send it to Mr. Zhang," and it automatically finds the PPT app, opens the file, extracts key points, and sends via email or WeChat. This requires system-level permissions and cross-app scheduling capabilities that independent apps can't achieve.

Apple's strategy is similar but more closed. The new Siri AI gained system-level multimodal screen perception and can complete system-level operations directly. Apple raised prices due to AI investment -- 16-inch MacBook Pro up $300, iPad Air from $599 to $749 -- Tim Cook said the price increase was "inevitable." Consumers are paying for AI entry points.

Endgame of the Agent Wars: From Chat Interface to Action Interface

Looking at WeChat Xiaowei, WeCom Dayuan, HarmonyOS Xiaoyi, Apple Siri, and Doubao-Feishu collaboration together, a clear trend emerges: AI's interaction interface is shifting from "chat interface" to "action interface."

ChatGPT represents the "chat interface" paradigm: you ask, AI answers, all actions require you to complete them yourself. WeChat Xiaowei represents the "action interface" paradigm: you say, AI does, completing operations directly within your ecosystem. This is a fundamental interaction paradigm shift.

This shift has profound implications for all AI companies. If your AI product is just a chat box, you face a pincer attack from super-apps and operating systems -- they have users, scenarios, data, and system permissions; standalone chat apps will struggle to compete. But if you build vertical Agent capabilities (like helping Xiaowei better handle medical consultations, or helping Dayuan do better sales analytics), you have a chance to become a "plugin" in the Agent ecosystem.

Behind Tencent's single-day market cap swings of tens of billions of HKD is capital's revaluation of Agent entry-point value. The last time WeChat triggered this level of market cap reaction was when Video Accounts rose to prominence. Agents may be the biggest variable in the WeChat ecosystem since Video Accounts.

1.6 billion people's screens are becoming conversation interfaces. And whoever gets the first word in that conversation interface holds the ticket to the AI era.