36Kr has learned from Tencent sources that WeChat is working with smartphone makers, including Huawei, Xiaomi, Honor, Oppo, and Vivo, to launch agent-to-agent (A2A) capabilities. Several manufacturers have already completed integration, according to the sources.
The feature will allow users to start WeChat audio or video calls, or send messages to specified contacts, through the artificial intelligence assistant built into their phone system. Tencent said the feature will maintain data security and privacy through a dual authorization structure.
That technical route differs from the approach typically used by system-level AI assistants. Those assistants often rely on graphical user interface (GUI) agents that read the screen as a user would and simulate taps to operate apps such as WeChat.
In this case, WeChat opens a controlled access point that lets system agents, such as Huawei’s Xiaoyi, Xiaomi’s Xiao Ai, Honor’s Yoyo, Vivo’s V, and Meizu’s Aicy, communicate directly with WeChat’s internal agent. After the system agent parses a user’s intent, it sends instructions through an encrypted and controlled protocol. WeChat then executes the action in the background and returns the result.
In essence, the structure requires dual authorization: user authorization and app authorization.
At Tencent’s first-quarter earnings call in May, an analyst asked Tencent president Martin Lau how he viewed the long-term potential and possible disruption of operating-system-level agents from iOS, Android, or smartphone makers.
Lau’s response separated true operating systems, such as iOS and Android, from apps that try to behave like operating systems. If an OS wants to serve users through an agent, he said, it must protect the ecosystem, curate it carefully, and obtain reasonable permissions from apps. Otherwise, the OS is effectively taking from apps rather than managing the platform neutrally.
His point is now clearer: agents can operate apps from the OS layer, but not without app authorization. Tencent may not accept GUI agents, but it can accept A2A integrations.
Phone makers have tried to force their way through GUI automation before. In 2023 and 2024, manufacturers promoted phone assistants that claimed to be “fully autonomous.” Honor’s Yoyo pitched WeChat “red packets” and food orders, while Xiaomi promoted Xiao Ai’s ability to place WeChat calls automatically.
The method relied on system-level macros or GUI automation. When a user said, “Send someone a RMB 10 (USD 1.5) red packet,” the assistant would unlock the phone, open WeChat, search for the contact, tap the plus sign, select the red packet, enter the amount, and trigger payment.
WeChat soon blocked this behavior across the board. In April 2025, the platform warned users not to install or use third-party tools that could access local chat records or bypass WeChat’s security measures. System-level AI agents were then pushed back into smaller tasks, such as testing smoothness, adjusting screen brightness, or connecting to Wi-Fi.
Phone makers have continued trying to use accessibility workarounds, but that approach is unstable. Once WeChat changes its interface or controls, AI assistants may tap the wrong person or the wrong button.
The Doubao phone appears to have tested app providers’ limits again. In December, ByteDance and ZTE released a technical preview of the Doubao phone assistant on the Nubia M153 engineering prototype, with system-level cross-app AI operation as its primary feature. The product was later blocked by app providers. That month, many users reported forced WeChat logouts and abnormal login environment warnings, while Taobao, Alipay, Agricultural Bank of China, and China Construction Bank also began restricting related capabilities.
Tencent said the Doubao phone assistant had used system-level permissions to simulate user actions, triggering WeChat’s existing security risk controls and violating rules against third-party plug-ins and automated operations.
On June 2, the Financial Times, citing two people familiar with the matter, reported that Tencent had completed prototype testing of its AI agent and could begin compliance approvals this month before a public launch.
After the report, Tencent shares closed up 10.5% that day, marking the stock’s biggest single-day gain since January 25, 2021.
The Financial Times also reported, citing a person who had seen an early demo, that the AI agent could be accessed by swiping right from WeChat’s main screen. Users could enter instructions allowing the agent to access WeChat’s millions of mini programs and complete tasks, such as finding a cafe that matches their price and taste preferences, then ordering a drink.
In the AI era, major Chinese tech companies are leaning into their strongest positions. Alibaba’s Qwen app has connected to dozens of Alibaba ecosystem agents for maps, ride-hailing, shopping, and quick commerce, while also opening to third-party agents and skills. ByteDance has the Doubao app, related large models, and early cooperation with smartphone makers. According to the source article, nine of the world’s top ten smartphone makers, excluding Apple, have already connected to Doubao models through Volcano Engine.
Tencent’s greatest advantage remains WeChat, China’s dominant superapp.
One Tencent insider said any phone agent that cannot call WeChat is not a true system-level agent. Tencent will open this channel, the person said, but only on its own timing.
Phone makers may pay infrastructure platforms such as Volcano Engine for AI computing services, but at the app layer, they know users can switch phones more easily than they can give up WeChat. A phone that can place a WeChat call by voice command could have a meaningful experience advantage in the high-end market. Cooperation with Tencent is something phone makers can wait for, but cannot skip.
Tencent’s A2A cooperation with smartphone makers opens a controlled path between device makers and app providers while balancing both sides’ interests. Foreseeably, more Alibaba and ByteDance ecosystem apps may eventually connect to system agents through the same route.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Wang Yuchan for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.79 = USD 1 based on estimates as of June 9, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.