The ritual of making dinner plans is about to get surreal.

It used to go like this: you text your friend "what do you feel like eating tonight?" → they reply "up to you" → you say "that new hot pot place?" → they say "kinda far" → five or six rounds later you finally lock it in. The whole thing takes over ten minutes, not counting the classic back-and-forth of "whatever you want, you pick."

WeChat's in-testing "Xiaowei AI Social" changes this flow. You tell your own Xiaowei "dinner with Zhang San tonight, he's been craving spicy food, around 7pm," and Xiaowei goes off to find Zhang San's Xiaowei to talk it out — cuisine, time, location, all of it. The two AIs hash it out between themselves, then bring the result back to you for confirmation. It only interrupts you when a decision needs your sign-off. Otherwise it just keeps chatting in the background.

This sounds like a small feature. The structural shift behind it? Not small at all.

The Dividing Line: From "Message-Sending Proxy" to "Two AIs Negotiate First"

Most people's first reaction is "isn't this just auto-sending WeChat messages?" It's not.

Auto-sending follows the logic of "you say something, it types it out and sends it." Xiaowei AI Social follows the logic of "you give it a goal, it goes talk to the other person's AI, and brings back the result." The former is a tool. The latter is an agent.

A key detail from reporter testing: AI social content doesn't appear in personal chat windows — it only happens inside "Xiaowei." In other words, your chat history with a friend stays clean. The two AIs' conversation takes place in a separate parallel space. You never see what the other person's Xiaowei said, how many rounds they went, or what tone they used — you only see the final result.

This design is thought-provoking. It means WeChat isn't bolting an AI social feature onto the existing chat box. It's building a separate "agent layer." Person-to-person chatting happens up top; agent-to-agent negotiation happens down below. The two layers don't intermix.

WeChat's A2A Vision: 1.4 Billion People's Social Life Gets an AI Layer

"Before you even talk to your friend, two Xiaoweis have already chatted" — it sounds a little sci-fi, but it's happening.

WeChat doing this makes perfect sense. It already has the Xiaowei agent entry point and millions of mini-program service entries. Now it's adding agent-to-agent capability to the person-to-person social graph.

Imagine future scenarios: you want to organize a gathering, Xiaowei handles the invites; you want to book a restaurant, Xiaowei books it; you want to send a client a gift, Xiaowei picks it out. All these things used to require opening five or six apps and five or six rounds of back-and-forth with people. Now you can hand them to Xiaowei to run in the background.

This isn't a "chatbot." It's a "social operating system." WeChat is turning itself from chat software into an agent orchestration platform. The people are still the same people. But more and more of the interaction between them is being handled first by AI.

Of course, there's another side to this. When two AIs socialize for you, what becomes of the meaning of "socializing" itself? The warmth between friends, the negotiation, the ambiguous warmth of "whatever you want, you pick" — will all that get efficiency'd away? Nobody has the answer yet. But one thing is certain: WeChat is already quietly running this experiment, and 1.4 billion users' social habits have always been defined by the product.

See you tomorrow.