You were promised an AI assistant that would make your life effortless. Remember that? The dream was a digital butler quietly handling the tedious friction of your day. The reality? You’re now stuck babysitting two robots arguing over who “received” a message.
Enter WeChat’s new AI assistant, Xiao Wei. It recently gained the ability to talk to your friends’ AI assistants. You tell it what to say, it drafts the message, you approve it, and it sends it to your friend’s AI. Your friend then reads it, approves a reply, and the bots send it back.
On paper, it sounds like magic. In practice, it feels like you’re trapped in a corporate CC-chain.
Let’s look at the actual experience. You want to ask a friend if they are free this weekend. You tell your AI. Your AI tells their AI. Their AI says, “Received, forwarding to human.” Your AI says, “Waiting for reply.” The reply comes. Your AI says, “Received, forwarding to you.”
It’s exhausting. Why are we watching two bots do a procedural handshake just to say “wanna grab a beer?” Machines needing to hand off tasks doesn’t mean humans need to watch the handoff. WeChat has accidentally built a bureaucratic nightmare for casual chatting. But underneath this clunky UI is the Trojan Horse for the next era of the internet: Agent-to-Agent (A2A) collaboration.
Here is the twist: This isn’t a failed AI social experiment. It’s a testbed for user authorization boundaries. The real story isn’t how many rounds the AIs can chat, but how much autonomy we are willing to give them.
The problem with the current execution is a fundamental mismatch of value. In casual chat, the back-and-forth is the point. In getting things done, the back-and-forth is the tax. Chatting is about connection. We don’t want to outsource our jokes, our empathy, or our weekend plans to a middleman. We want to talk to our friends.
But getting things done? That’s where A2A shines. Imagine you want to buy a used camera from a friend, but neither of you wants the awkwardness of haggling. You give your AI your target price, your absolute bottom line, and your delivery preferences. Your friend’s AI knows their minimum price. The two AIs negotiate. They go back and forth a hundred times if they need to. But you don’t see it. You only get a notification when they’ve reached a deal, or if there’s an absolute deadlock that requires your human input.
That is the future WeChat is poking at. The current iteration forces you to approve every single message because the AI doesn’t have the authority to make decisions. It’s just a glorified relay service. But the moment you grant it a boundary—”negotiate between $50 and $70, but don’t agree to shipping until you ask me”—it transforms from a parlor trick into an actual utility.
We need to stop judging AI by how well it mimics human conversation. We don’t need AI to be our social proxies. We need them to be our negotiators, our schedulers, our friction-removers. The true metric for an AI assistant isn’t how many witty lines it can draft, or how many rounds of chat it can sustain.
A real assistant isn’t measured by how much it talks, but by how much it lets you ignore.
WeChat’s current social feature is awkward, yes. But it is the necessary first step toward a world where our devices negotiate on our behalf. Let the bots do the haggling. Save the chatting for us.
FAQ
Q: If the current AI chat experience is so clunky, why is WeChat rolling it out?
A: Because they are testing user authorization boundaries. Before AI can autonomously negotiate for you, companies need to train users to accept AI-to-AI communication. The current 'confirm every message' flow is a safety blanket that will gradually be removed as trust increases.
Q: How should product managers actually evaluate AI assistant features?
A: Stop tracking conversation rounds or engagement metrics. The only metrics that matter for A2A features are: how many times was the user interrupted, did the task get completed within the user's boundaries, and how much time was actually saved.
Q: Is AI social interaction just a dead end?
A: Yes, AI 'socializing' on our behalf is fundamentally useless. We don't need digital middlemen for friendship. The real value isn't in AI talking to other AI for chat, it's in AI talking to other AI to get work done without us having to watch.