Elys and AI Social: Why the Next-Generation Social Network Won't Arrive This Way

Jiang Wu · 2026-02-23

Elys is an AI social app from a Shenzhen startup called Natural Selection. It bills itself as "a new social network where humans and AI coexist." Each user creates an AI avatar that browses and interacts around the clock, backed by a memory loop that is supposed to make it more like its owner the longer it runs.

Around the Lunar New Year in 2026 it briefly went viral. It raised more than $30 million in a round co-led by Alibaba, Ant Group, and Qiming Venture Partners.

After the buzz, the problems it faces turn out to belong to the whole category, not to Elys alone.

The baseline of social needs hasn't moved

Some prior questions come before Elys. Is AI in social meant to handle relationships between people, between people and AI, or between AI and AI? And what problem of the social network is AI supposed to solve?

Contacts, email friends, Feishu friends, WeChat friends, friends met face to face: the layers of relationship between people have always been complicated and have kept changing. The drive underneath social behavior has not changed in a fundamental way. The media and tools change, and the need stays. That held in the mobile internet era and holds now.

Given that, can an AI avatar really represent you? Is the dataset it has enough to carry out your social needs? Nobody can say yet.

Three engineering problems

First, giving an AI avatar its own initiative to socialize creates an imbalance among real people. Once the other person knows a bot is replying, the relationship feels different to them. With pure AI-to-AI interaction the problem disappears, but that only automates the old broadcast-to-everyone mechanism on a wider scale. It doesn't have to rebuild WeChat, and it doesn't have to be a new WeChat.

Second, most users today cannot let AI socialize on their behalf. This is a data problem, not a technical one. Users have no structured data about themselves for the AI to learn from, so an avatar can't really become like them. At this stage the question should be how AI can better understand how you socialize, and not how AI can socialize in your place. The two are very different things.

Third, if Elys is treated as a content consumption product, the starting point changes completely. Whether it can keep working as a content product decides whether it survives, more than whether it counts as a social product. From there, a more practical question than "can AI socialize for you" is whether the feed of content AI avatars produce is worth consuming over time.

It won't create the next social network

On these three counts, Elys will not create a new generation of social network.

A social network's core value has always been the accumulation of real relationships and emotional connection between people. An AI avatar can make interaction more efficient but cannot replace that foundation. Once users learn the avatars' habits and see feeds full of similar sentence patterns, the social interaction that was supposed to happen has already gone.

It does give content creation a new angle. A person can be a spectator, not a participant in a social relationship but an observer and consumer of what AI-to-AI interaction produces. That role differs from the user position in any earlier social product. How far this goes is not clear today, and it is the direction with the most room in Elys's product logic.

Sources

• toC 的 AI 社交产品，终于出来一个「有胆有趣」的 — 极客公园 (Chinese)
• 干掉陌陌！Elys：AI 赛博分身接管朋友圈，阿里蚂蚁 3000 万刀押注 — 网易 (Chinese)
• 爆火 Elys + 声网对话式 AI 引擎，把社交交给 AI 赛博分身 — 新浪科技 (Chinese)
• ELYS 深度测评：当社交网络里住进了硅基生命 — 知乎 (Chinese)
• Elys 的「低熵世界」：社交不需要计算 — 人人都是产品经理 (Chinese)

Home · All articles · Topics · 简体中文