WeChat AI Opens a Narrow Door for Phone Manufacturers | Focus Analysis 微信AI对手机厂商打开一道窄门｜焦点分析

WeChat has finally come around. Or rather, it’s not that it “came around” — it’s that it “crunched the numbers.”

Not long ago, Tencent President Martin Lau’s remarks at the earnings conference still echoed: “You can have your own agents on your operating system, but if you think you can use system privileges to ‘raid’ app data and functionality? Not a chance.” This was essentially a stern warning to all phone manufacturers at the time who were trying to use GUI Agents to forcefully “control” WeChat. To put it bluntly: if Tencent won’t open a backdoor to WeChat, no one else should think about prying it open.

And so, we witnessed a brilliantly executed blockade. Honor and Xiaomi’s once-touted “send WeChat red packets with one sentence” feature relied on a background chain of simulated clicks—“unlock, open WeChat, search, tap”—which was swiftly shut down by a single announcement from WeChat’s Security Center. System AI assistants instantly regressed from “all-powerful butlers” to “feature-phone” level tools that could only adjust brightness or connect to Wi-Fi, leaving them utterly embarrassed. It wasn’t until ByteDance’s Doubao phone assistant attempted even more aggressive system-level permissions for “cross-app operations” that it was met with a collective “ban package” from WeChat, Taobao, and banking apps. The reality is clear: when facing WeChat’s absolute scale, any “technical shortcut” that bypasses official authorization is a dead end.

Therefore, this so-called “A2A assistant capability collaboration” is essentially a long-overdue, inevitable “co-optation.” Tencent has opened a strictly regulated “door,” allowing system assistants like Huawei’s Xiaoyi and Xiaomi’s Xiaoai to come in for “dialogue,” rather than letting them “simulate clicks” like intruders. This marks a shift from “physical isolation” to “diplomatic negotiation.” For phone manufacturers, it resolves the critical pain point of unstable user experience; for Tencent, it transforms a “potential threat” into a “controlled channel,” with the keys to the gate firmly in its own hands.

But the most intriguing aspect lies in the subtle shift in power dynamics. In the past, phone manufacturers wielded system privileges, attempting to position themselves as “super gateways,” while WeChat was the “invaded” party. Now, the collaboration appears equal on the surface, but it isn’t. As one Tencent insider put it, “Any phone agent that can’t access WeChat isn’t a true system-level agent”—a statement dripping with confidence, which underpins this entire partnership. Phone makers know that users might switch devices, but they will never leave WeChat. Thus, this collaboration isn’t Tencent “granting favors”; it’s an experiential necessity that manufacturers “have to embrace,” especially in the high-end market.

Zooming out, this is a microcosm of how tech giants are staking their claims in the AI era. Alibaba is pursuing a “full-stack ecosystem” approach, bundling its services into an all-encompassing AI super-agent; ByteDance’s Volcano Engine acts like an “infrastructure fanatic,” quietly securing a strategic foothold in the domestic phone ecosystem through foundational computing power. Meanwhile, Tencent holds WeChat—China’s sole “super app”—along with millions of mini-programs. It chooses to integrate in a more conservative, rule-abiding manner. It doesn’t rush to be the “all-in-one agent” at the forefront; instead, it positions itself as the indispensable “infrastructure” of the AI ecosystem—any system assistant seeking a complete experience will ultimately have to “shake hands” with it.

The so-called “dual authorization mechanism” (user authorization + app authorization) sounds safe and standardized, but beneath it lies raw ecological influence. The A2A protocol is the only permitted tunnel through the “Berlin Wall” that Tencent has drawn, with all rules set by it. This prevents the chaos that could arise from unchecked system-level “raids,” but it also means the future app-agent ecosystem may very well operate according to “Tencent standards.”

So don’t just fixate on the convenient future of “one-tap WeChat messages.” The deeper significance of this collaboration lies in establishing a new order: in the AI Agent era, the super app remains king—only now, instead of rejecting all outsiders, it has become the “gatekeeper.” Phone manufacturers have temporarily gained the gateway capabilities they’ve long desired, but the price is acknowledging WeChat’s “suzerainty.” The breach in this “Berlin Wall” has been strategically crafted—it doesn’t weaken the wall; it reinforces it.

微信终于想通了。或者说，不是“想通”，而是“算明白了”。

之前腾讯总裁刘炽平在财报会上那番话还言犹在耳，大意是：操作系统你可以有你的智能体，但想用系统权限来“掠夺”应用数据和功能？门都没有。这基本是对当时所有试图用GUI Agent（图形界面智能体）“硬控”微信的手机厂商发出的严厉警告。翻译一下就是：微信的后门，腾讯自己不开，谁也别想撬。

于是我们看到了一场漂亮的封杀战。荣耀、小米前几年宣传的“一句话发微信红包”，后台那个“解锁-点微信-搜索-点击”的模拟点击链条，很快就被微信安全中心一纸公告给堵死了。系统AI助手瞬间从“全能管家”变回了只能调亮度、连WiFi的“功能机”，尴尬得不行。直到字节的豆包手机助手以更激进的系统级权限试图再次“跨应用操作”，迎来的却是微信、淘宝、银行APP的集体“封号套餐”。事实证明，在微信的绝对体量面前，任何绕过官方授权的“技术捷径”，都是死路。

所以，这次所谓的“A2A助手能力合作”，本质上是一场迟到的、必然的“招安”。腾讯开出了一道有严格协议的“门”，让华为小艺、小米小爱这些系统智能体可以进来“对话”，而不是让它们像强盗一样“模拟点击”。这是一个从“物理隔绝”到“外交谈判”的转变。对手机厂商而言，这解决了体验不稳定的致命痛点；对腾讯而言，则是把“潜在威胁”收编为“受控渠道”，开关门的钥匙牢牢握在自己手里。

但这事儿最有意思的地方在于权力关系的微妙转换。过去，手机厂商拿着系统权限，试图扮演“超级入口”，微信是那个被“入侵”的对象。现在，合作关系看似平等，实则不然。一位腾讯内部人士那句“任何手机agent，如果不能调用微信，就不算真正的系统级agent”，话里透出的自信，才是这场合作的底色。手机厂商清楚，用户可以换手机，但绝对离不开微信。因此，这个合作不是腾讯“施舍”，而是厂商“不得不做”的体验刚需，尤其在高端市场。

把视角拉远，这其实是AI时代巨头划地盘的一个缩影。阿里走的是“全家桶”路线，把自家生态捆成一个智能体巨无霸；字节的火山引擎则像一个“基建狂魔”，在底层算力上悄悄完成了对国产手机生态的“卡位”。而腾讯，手里捏着微信这个中国唯一的“超级应用”和它背后百万量级的小程序，它选择以一种更保守、更讲规矩的方式接入。它不追求自己冲到最前面做那个“全能代理”，而是把自己变成AI生态里那个不可或缺的“基础设施”——所有想获得完整体验的系统智能体，最终都得来和它“握手”。

所谓的“双重授权机制”（用户授权+应用授权），听起来很安全，很规范，但背后是赤裸裸的生态话语权。A2A协议是腾讯画下的“柏林墙”上唯一被允许的通道，所有规则由它制定。这避免了系统无序“掠夺”带来的灾难，但也意味着，未来的应用智能体生态，很可能要按照“腾讯标准”来运行。

所以，别只看到“一键发微信”的便捷未来。这场合作更深层的意义，是确立了一种新的秩序：在AI Agent时代，超级应用依然是王，只是从拒绝一切外来者，变成了自己做“守门人”。手机厂商们暂时获得了梦寐以求的入口能力，但代价是承认了微信的“宗主权”。这场“柏林墙”上的破洞，开得很有策略，它没有削弱墙，反而让墙变得更稳固了。