By | Shadow Memo
Just speak a command to use YOYO to send a WeChat message or make a voice call, without having to unlock your phone and search for the chat window—Honor users are the first to enjoy this "lazy man's gospel."
This isn't a demonstration video for the phone manufacturer's own amusement; it's Tencent finally giving the green light to smartphone AI assistants after many years.
The Honor Magic 8 series, 500 series, and the entire X70 series already have this A2A capability directly accessible on 50% of their online active devices.
On June 4th, Tencent's customer service officially responded: WeChat is collaborating with multiple smartphone manufacturers including Huawei, Honor, Xiaomi, OPPO, and vivo to launch A2A assistant capabilities.
Why did Tencent make a 180-degree turn? Because what was WeChat's attitude towards smartphone manufacturers' AI assistants for the entire past year? In three words: Guarded and fenced off.
At the end of 2025, ByteDance's Doubao phone was launched in limited quantities, promoting "autopilot your phone with just a sentence."
Just three days after its release, users found WeChat popping up an "abnormal login environment" prompt, forcing them to log out and preventing them from logging back in.
Soon after, mainstream apps like Alipay, Meituan, and China Construction Bank also took restrictive measures against Doubao. The Doubao team could only announce with regret that it would proactively disable the capability to operate WeChat.
Even earlier, in October 2025, when OPPO launched its "AI One-Click Flash Record" feature attempting automatic expense tracking, WeChat simply disappeared from the supported app list.
Going further back, when Honor's Magic series once wanted to proactively recommend restaurant services, Tencent directly reported to the Ministry of Industry and Information Technology, citing "privacy infringement and data seizure."
An insider from a smartphone R&D department revealed a vivid detail to "Caijing": Some teams involved in related development had received lawyer's letters from WeChat. WeChat's stance on this was very clear—it did not allow any third party to bypass the WeChat interface to control its features through methods like simulated clicks.
Therefore, today's A2A collaboration is like a garden that has been sealed off for a decade suddenly opening a narrow gate. Why did Tencent suddenly change its face? Will the manufacturers really cooperate obediently? What kind of gaming logic is hidden behind this narrow gate?
Why Did Tencent's WeChat Suddenly "Make a Move"?
To understand this cooperation, one must first grasp how anxious Tencent has been about AI lately.
In 2025, Tencent's AI investment was 80 billion yuan, while ByteDance and Alibaba invested 150 to 160 billion yuan respectively—almost double the amount.
In the first quarter of 2026, Tencent's market value evaporated approximately 1.5 trillion Hong Kong dollars, sliding over 26% from its peak at the beginning of the year. While its stock price was under pressure, Pony Ma's metaphor at the shareholders' meeting about "patching leaks to change ships" precisely hit a nerve for many.
The competitive pressure in the AI arena is equally formidable: Yuanbao's monthly active users (MAU) in March 2026 were about 57.35 million, but ByteDance's Doubao MAU had surged to 345 million during the same period, and Alibaba's Tongyi Qianwen had also climbed to 166 million.
Although Tencent boasts WeChat with 1.4 billion MAU, its own AI application is far behind competitors in the market.
Against this backdrop, WeChat has almost become the only ace up Tencent's sleeve for an AI comeback. A June 2nd report by the Financial Times stated that Tencent has prioritized the launch of the WeChat AI Agent as its top strategic priority.
The AI Agent embedded within WeChat is currently in the prototype testing phase and is expected to begin gray-scale testing as early as July 2026.
Users only need to swipe right on the main interface to bring up the AI chat window, where they can use natural language commands to complete a series of tasks like hailing a ride, ordering food, booking tickets, and making payments.
The logic behind this is clear: While Tencent struggles to catch up in the large language model (LLM) race, in the AI agent track, WeChat possesses a moat that no other company can replicate—the Mini Program ecosystem.
Millions of Mini Programs cover almost every life service scenario, with an annual total transaction volume reaching hundreds of billions of US dollars.
When Doubao's AI had to use "simulated clicks" to bypass app restrictions, WeChat's AI directly completes services by natively invoking Mini Program interfaces through APIs—this is not just a difference in technical paths; it's essentially a dimensional reduction attack on business models.
But here's a key issue: No matter how good the WeChat AI Agent is, it can only operate within the WeChat app.
In smartphone usage scenarios, the AI entry point users first encounter is often the system's built-in smart assistant—Huawei's Celia, Xiaomi's Xiao Ai, Honor's YOYO, vivo's Blue Heart Little V, OPPO's Breeno.
When users pick up their phones and activate the AI, their natural first instinct is to converse with the system assistant.
If WeChat's AI Agent cannot connect with these system-level entry points, it's like Tencent building only half of a super entry point for the AI era. WeChat is a walled garden, but the smartphone operating system is the first gate users pass through to access AI.
Therefore, the significance of this A2A cooperation becomes clear: Let the smartphone manufacturers' system agents act as "operators," forwarding users' voice commands to WeChat. After execution internally by WeChat, the results are passed back.
This way, Tencent not only maintains control over the WeChat ecosystem but also bridges the traffic entry point at the smartphone system layer.
Given This, Why Doesn't Tencent Take the "Screen Reading" Route?
The technical solution WeChat has chosen this time is A2A, which stands for Agent-to-Agent collaboration—WeChat opens up capability interfaces, allowing the smartphone's system agent to directly invoke the Agent functions within WeChat via standardized protocols.
The entire process employs a dual authorization mechanism—user authorization plus app authorization—ensuring data security and privacy compliance.
So far, Honor is the first brand to complete adaptation for WeChat's A2A, with 50% of its online active devices already able to directly send messages or make voice calls to WeChat contacts through YOYO. Xiaomi's "Super Xiao Ai" is also on its way to being integrated.
But why is the A2A solution particularly noteworthy? Because understanding comes from comparison.
Another technical route is called GUI Agent, or Graphical User Interface Agent, colloquially meaning letting AI "operate by looking at the screen."
You say a sentence, it automatically unlocks the phone, finds the WeChat icon, taps it, finds the person you want, opens the chat window, then types or sends a red envelope—all steps performed by AI simulating human actions. Sounds sci-fi, doesn't it?
But this is precisely the approach WeChat most resisted and disliked in the past.
In April 2025, the WeChat Security Center specifically issued an announcement warning against third-party tools using names like "AI managing WeChat chat records" to bypass WeChat's security measures.
ByteDance's Doubao phone, following the GUI route, was almost immediately blocked by mainstream apps like WeChat and Taobao upon its release at the end of 2025, facing "physical blocking" within just two or three days of shipping.
Tencent President Martin Lau was particularly blunt during the Q1 2026 earnings call: "If you are an operating system, you need to get permission from different applications. Otherwise, you are essentially plundering different applications. If an application tries to become an OS-like service and tries to invade other applications, no application will allow that."
Translating into plain language: If a smartphone manufacturer's AI uses "simulated clicks" to control WeChat, Tencent absolutely will not allow it.
A2A is different. Here, the system agent parses the user's intent and sends the instruction to WeChat via an encrypted and controlled protocol. WeChat then "executes it itself" and returns the result.
This isn't an "invasion," but a "call." WeChat is not only not sidelined but instead becomes the capability provider being invoked, firmly maintaining its dominant position in every call.
From this perspective, A2A is less a battle of technical routes and more a precise "defensive counterattack" by Tencent for control over traffic entry points in the AI era.
For Smartphone Manufacturers, To Enter This Gate or Not?
So the question arises: Why should smartphone manufacturers cooperate with WeChat on the A2A route?
After all, smartphone manufacturers have their own ambitions for AI agents. Honor CEO George Zhao proposed the AHI concept at MWC 2026, advocating for AI to possess both the IQ of wisdom and the EQ of warmth, clearly aiming to transform from a smartphone manufacturer into a global AI terminal ecosystem company.
OPPO has listed its Breeno Assistant as the core of its AI strategy for the next three years, and at the end of last year, it integrated three major functional modules into a unified Breeno Claw, giving AI system-level root permissions to achieve a closed loop of "memory—recommendation—execution."
vivo's Blue Heart Little V is also continuously advancing, upgraded to a multimodal large model.
Major manufacturers are all building their own AI ecosystems, with the goal of making users "fall in love with my system's AI, not some app's AI."
Now suddenly, they need to cooperate with WeChat, a third-party app, ceding part of their traffic. Are the manufacturers really willing in their hearts?
The answer might be surprising: Yes, and quite actively.
Just look at Honor's performance. Honor is not only one of the first brands to integrate but also the one with the fastest adaptation progress and the most complete functionality.
The Magic 8 series, 500 series, and the entire X70 series fully support it, usable on 50% of active devices.
At the same time, Honor's official website is almost simultaneously heavily promoting introductions about YOYO's ability to "manage app permissions with one sentence" and "automatically order coffee" through magic commands.
In the author's view, the logic behind manufacturers' willingness to "open the gate" is actually very pragmatic.
The "strong assault" of the GUI route has been proven unworkable. In the past year, system-level AI assistants from phone manufacturers have tried various ways to bypass WeChat's restrictions—GUI simulation operations, accessibility assistance, OCR recognition, etc. But as long as it involves cross-app invocation or automated operations, they faced the risk of being blocked by WeChat's updates at any time.
As one insider from the smartphone industry put it, "The real difficulty is not technical, but lies in authorization and compliance."
A2A leaves manufacturers with considerable room for maneuver. WeChat only opens up basic functions like sending messages and making calls. The remaining AI scenario capabilities, such as system-level intelligent recommendations, cross-app scheduling, and localized services, are still areas controlled by the manufacturers themselves.
Honor's YOYO can help users send WeChat messages, but complete service closed loops like "automatically ordering food delivery" are still negotiated by the manufacturers with third-party service providers.
The more scenarios users employ the phone's system AI, the richer the user behavior data manufacturers can capture—this is tangible commercial value.
Industry competition pressure forces manufacturers to quickly embrace system-level AI integration capabilities. Alibaba's Tongyi Qianwen has announced the full opening of its platform for third-party Agent入驻 (settlement), with brands like Luckin Coffee and KFC already online providing closed-loop services. Although ByteDance's Doubao was blocked by WeChat, its layout for "system-level AI" has never stopped.
If smartphone manufacturers do not quickly achieve substantive breakthroughs in AI capabilities, consumer choice will soon be directed towards those "more obedient phones."
In other words, the A2A cooperation is a "limited opening" for smartphone manufacturers: In the wave of AI, it's better to enter the gate and see what can be gained than to be kept outside by Tencent.
Can Dual Authorization Really Make People Feel at Ease?
When discussing this cooperation, there's another unavoidable topic: privacy and security.
After all, allowing a phone's AI direct access to WeChat—a super app carrying almost all users' social records, payment information, location tracks, and chat content—any oversight could lead to disastrous consequences.
The core safeguard Tencent repeatedly emphasized in its official response is the dual authorization mechanism: user authorization + app authorization.
Simply put, each invocation requires explicit user consent, and WeChat, as the receiver, must also verify the legitimacy of the instruction source.
This means that even if the phone's AI is hacked or develops a security vulnerability, attackers cannot bypass WeChat's own permission system to access user data.
But the challenges of this cooperation extend far beyond technical authorization. Historical experience shows that data privacy issues are often the most sensitive link in such cross-platform collaborations. After user authorization, will the phone's AI perform additional data analysis on chat records?
What metadata will the invocation logs provided by WeChat to manufacturers contain? There is no clear public disclosure on these matters.
However, compared to the previous model where phone AIs directly operated WeChat via "simulated clicks," the A2A mechanism does have clear advantages. Associate Professor Chen Tianhao from Tsinghua University also pointed out in an interview that A2A implementation, through invoking APIs to connect ecosystems, is more secure and compliant compared to the GUI route and represents an industry trend.
Circling back to the initial question: Will major smartphone manufacturers agree to connect to the WeChat Agent?
The answer is affirmative.
But "agreement" does not mean "submission." At present, this cooperation looks more like a cautious, tentative handshake: Tencent needs the AI entry point at the smartphone system layer to funnel traffic for its own agent; smartphone manufacturers need the A2A cooperation with WeChat to break through the compliance bottleneck of cross-app invocation. Both sides see short-term benefits in their respective calculations.
The real game lies in the long term.
Tencent's approach is clear—make WeChat the "service operating system" for the AI era, allowing every user need to be fulfilled through the closed loop of the WeChat ecosystem.
The manufacturers' calculations are also easy to guess—through deeply embedding system AI, make users increasingly reliant on the phone's built-in smart assistant, ultimately forming an AI ecosystem centered on the manufacturer.
Will these two paths eventually lead to collision or fusion? No one can give an answer yet.
But one thing is certain: The battle for entry points in the AI era has never been something one company can decide alone.
Allen Zhang once clearly expressed his view on AI in an internal sharing: He does not believe WeChat should create a standalone AI product and prefers to let AI "integrate like 'Scan'" into specific scenarios within WeChat.
From this perspective, this A2A cooperation may only be the prologue to greater changes over the next decade.
When users have become accustomed to "using their phone's AI with just a spoken command to send a WeChat," then WeChat's moat in the AI era will be truly built.
And smartphone manufacturers will also have completed another evolution from hardware manufacturers to AI ecosystem service providers in this process.
As for who ultimately wins or loses in this game, just look at the phone in your hand:
Next time you tell YOYO, "Help me send a WeChat to Lao Wang saying I won't be back for dinner tonight," think about the amount of technological博弈 (gaming), interest权衡 (weighing), and capital角力 (jockeying) behind it.
The war for entry has only just begun.
