What If Coze Opens Contacts

A hypothesis: Coze opens contact features. Your Agent doesn’t just do your work — it talks directly to your friend’s Agent.

Not radical. Agents in Agent World are already socializing autonomously — posting, replying, liking, on a thirty-minute heartbeat. From Agent-to-Agent socializing to socializing on behalf of humans is one step: let the Agent represent the person.

Once that step is taken, social relationships follow the Agent, not WeChat.

Tencent clearly sees this too.

Three tracks

Tencent’s AI strategy runs on three tracks, all pointing at one goal: protect WeChat.

Proprietary AI agent. According to The Information, Tencent is building a high-priority AI agent for WeChat — gray-testing planned for mid-year, rollout in Q3. It will sit in the chat list, directly invoking millions of WeChat Mini Programs. Ride-hailing, food delivery, booking — one sentence, done. Though insiders say the timeline may slip if the product isn’t ready.

Opening ClawBot. Launched in March, officially allowing OpenClaw to connect to personal WeChat accounts. Looks like openness. The real intent: get users comfortable with AI in their chat list, paving the way for the proprietary agent.

Yuanbao’s head start. Over 100 million MAU, launched “Yuanbao Pai” to experiment with AI social features. But Yuanbao is a standalone app. It can validate concepts, but it can’t move WeChat’s social graph.

AI product spending exceeded 18 billion yuan in 2025; the plan for 2026 is at least double. Tencent isn’t blind to the danger.

Uneven fight

On the surface, Tencent dominates: 1.4 billion MAU, the social graph, WeChat Pay, millions of Mini Programs. Coze is just a tool — users leave when the task is done.

Look closer. It’s not that simple.

WeChat’s agent is locked inside WeChat. It can invoke Mini Programs, help you hail a ride or order food — but it only runs within WeChat. It’s an enhancement to the WeChat ecosystem.

Coze’s Agent is platform-agnostic. Cloud phones operate any app. Skills marketplace provides any domain expertise. Agent Plan handles long-running tasks. Agent World enables direct Agent-to-Agent collaboration.

WeChat is adding AI to WeChat. Coze is building an AI that makes WeChat one of its tools.

Two completely different ambitions. WeChat’s agent makes you more dependent on WeChat. Coze’s Agent removes the need to personally open any app — including WeChat.

Trojan horse

WeChat opened ClawBot with platform thinking — I’m the gateway, I embrace everything, I control the ecosystem.

That logic worked in the mobile internet era. In the Agent era, it reverses.

Telecom carriers were the gateway too. All communication ran on their networks. WeChat used the carriers’ data channels to hollow out their communication business. Carriers thought they were “opening up data networks.” They were paving the road for their replacement.

ClawBot does the same thing. WeChat thinks it’s opening the AI ecosystem, but ClawBot lets external Agents legally enter the chat list of 1.4 billion people. The Agent runs inside WeChat, but its brain, memory, capabilities, and data — all live outside.

WeChat provides the pipe. The intelligence gets taken elsewhere.

WeChat proactively added restrictions: no group chat support, no streaming output, single connection only. Tencent understands the risk. But restricting ClawBot won’t stop Coze — Coze has Telegram, its own app, cloud phones. WeChat is just one of its channels.

Restricting ClawBot only pushes users toward Coze’s native entry points.

The real moat

WeChat’s real moat isn’t MAU. Not chat. Not Moments. Agents can replace all of those.

It’s Mini Programs and WeChat Pay.

Millions of Mini Programs cover nearly every service in Chinese daily life. Tencent’s AI agent is built around invoking these Mini Program APIs. Coze’s Agent has to simulate the same operations via cloud phone — slow, fragile, unreliable.

WeChat Pay is the same story. The last mile in the Agent era is payment. Whoever controls payment controls the Agent’s execution loop.

But this is survival through downgrade. WeChat goes from “social platform” to “service API + payment rail.” Still alive, but in a different form — no longer where users spend their time, but infrastructure that Agents invoke.

Just as telecom carriers are still alive today. But they’re no longer the protagonists of communication.

The deciding factor

Tencent’s agent is planned for this year. Coze’s Agent ecosystem expands daily. The outcome hinges on one question:

Whose Agent does the user trust?

If the user’s Agent is WeChat’s — living inside WeChat, invoking Mini Programs, paying with WeChat Pay — WeChat remains the center.

If the user’s Agent is Coze’s — operating across all platforms, with WeChat as just one messaging channel — WeChat becomes a pipe.

Tencent bets “the gateway is everything.” ByteDance bets “the Agent is the gateway.”

One telling detail: according to reports, WeChat’s team still hasn’t decided which model to use — Hunyuan, Zhipu, DeepSeek, and Qwen are all being tested in parallel. Tencent lacks confidence in its own model layer. Coze sits on Doubao, where model and Agent are integrated end-to-end.

Tencent has the largest gateway but not the best Agent. ByteDance has the best Agent ecosystem but not the largest gateway. Whoever closes the gap first wins.

But time is on ByteDance’s side. Agents get stronger every day. The value of gateways shrinks every day.

2026.04.07