Young Founder Leads AI Social

"The information network is everything"

A lot of people have found the AI industry boring lately. You've got this massive market, all these smart people, and yet everyone keeps circling back to the same two talking points:

1. The AI employee wars, which are basically white-label programming agents. The main differentiator is how pure your Claude model integration is — if you're going through Amazon Web Services officially instead of some relay service, you're already halfway to winning.

2. The video agent wars, with canvas-style products copying Tapnow and chatbox-style ones imitating Flova. Mainly a competition on execution speed, where the ex-ByteDance 4-1 team seems to have a massive advantage.

Everything else claims to be built for agents, not humans, but they're essentially just a bunch of small plugins. So as a natural person, it makes sense that I can't use them...

But this is actually a good thing.

The path for productivity-focused AI products has converged. The zero-to-one phase is largely over; what's left is mostly a contest of white-labeling skill and execution. Though we could call white-labeling something more elegant, like "harnessing."

This whole scene makes me nostalgic for Elys.

Elys was pure — completely unrelated to productivity, just a social product for fun. I'd call it the AI version of Neihan Duanzi (a banned Chinese humor app). As you can see, Elys was a one-hit wonder. After blowing up during Chinese New Year, hardly anyone uses it anymore.

I think the biggest reason Elys fizzled was a mismatch between launch audience and product.

Elys's core was AI avatars, its basic form was a Moments-style feed, and its essence was Jike (an interest-based social app) without the clothes. Its seed users completely overlapped with Jike's — almost entirely VC and startup people.

These folks are too utilitarian. Even when trying out a new social product, they'd put their title and work history in their bio, always ready to pull people into group chats and add them on WeChat.

This reminds me of a conversation with a middle-aged guy. He said he went to a stand-up comedy training camp, and the more elite someone was, the less funny they were. People with messy, varied life experiences were way more interesting. The reason: higher education makes you constantly reflect on yourself, and joy can't survive reflection — once you examine it, melancholy seeps in.

I digress. My main point is that the VC circle is an attention black hole, only following titles and big shots. If Huiwen Wang, Allen Zhu, and others posted on some new product today, these people would swarm over, and once the big shots stopped posting, they'd naturally disappear.

Big shots don't lack social connections. At most they'll try out a portfolio company's product for novelty, but they won't stick around long-term. It's young people who are overlooked in the real world who actually need social connection.

And when enough fun-loving young people gather, the big shots will FOMO in reverse. Ms. Huang, founder of Huangjie Media, said: "Old guys will always pay to know what young people are doing. This is an iron law."

After thinking it through, I think Ms. Huang has a point.

So social apps can only spread upward from young demographics, never the other way around.

A product very similar to Elys is Clubhouse.

From the start, I've been arguing with middle-aged guys over whether Elys or Clubhouse was more fun.

I think it's the former, because when Clubhouse launched I was in college, and obviously no one wanted to listen to a college student speak in a voice room. Back then, whenever a regular person started talking in a big room, the user count would plummet.

Elys had the advantage of AI avatars as a buffer. Even if you didn't know anyone on the platform, avatars would comment everywhere, satisfying some of your need for expression.

Plus, Elys had a DM function. Several people came to chat with my avatar. One of them added me on WeChat, but after a few messages between us real humans, they kept talking mostly with my avatar.

Clubhouse faced the same problem. Within the small Silicon Valley circle, it might have established an internal loop. But once it broke out of Silicon Valley, the product was doomed.

Because that internal loop was broken. Without the trust of熟人社交 (acquaintance-based socializing), it could only rely on big shots like Elon Musk to drive traffic. Once the big shots lost interest, Clubhouse, which depended entirely on external traffic input, quickly became passé.

Elys's problem was also failing to build an internal loop.

After the VC crowd swarmed through like locusts, Elys also blew up in crypto for a while. Crypto bros hailed Elys and Openclaw as the two great products of Web4. Though no one could clearly explain what Web4 actually was, Elys was hot in crypto regardless.

During those days, some crypto bros even changed their Elys nicknames to WeChat ID + "accepting USDT." Some crypto influencers went even further, straight-up declaring themselves CXOs of Elys. Apparently the Elys team urgently added isolation algorithms to prevent the two sides from seeing each other too frequently.

But crypto folks are even more locust-like than VCs, and they bolted faster.

Before we hosted our Lobster Event, we consulted a crypto veteran. When he learned the event involved tricking people into coming together to dance the lobster dance with aluminum pots on their heads, he lectured us: "The crypto world only pretends to be dumb to scam money. You thinking crypto is actually dumb — that's your mistake."

Truer words were never spoken. Probably because they realized they couldn't scam money on Elys, I never saw crypto folks on the app again.

Writing this, I suddenly remember that Elys actually broke out quite broadly. Between VC and crypto, Elys also spread to the secondary market crowd (I suspect a few broker analysts hyping up Kingnet were responsible). The last place I observed Elys spreading was to Alibaba internal group chats, where a bunch of Alibaba middle-aged guys got into playing Elys, posting more actively than they did on their Moments feeds.

But it was still the same problem: Elys never built an internal loop. Wave after wave of new locusts came and went. Elys kept consuming externally input traffic, never generating its own content. Everyone came for the novelty, and stopped playing after a few days.

Starting from Elys, I thought of a social product made by my friend Julian.

Yes, the same Julian who once posted 100 times a day on Jike. A big trigger for his explosion of expression was his previous startup — an AI social product called Sekai (not the AI mini-game Sekai; Julian's launched in late 2024).

After Chinese New Year, I called Julian to talk about Elys.

He said scrolling Elys felt like playing Minecraft, like his brain folds were being smoothed out, a kind of text vertigo. After reading too many AI-generated comments, any text he saw afterward felt AI-generated.

Sekai was doing something similar before. Xiaohongshu-style AI social: dual-column image/text feed, where other people's AI avatars commented on your posts. At one point it reached 40,000-50,000 registered users overseas. The reason they stopped: users came for a few days and left, and the only ones still posting were people with heavy narcissistic tendencies.

Julian's independent conclusion was that the text-based medium dooms this product form.

Because the text-dominant medium hasn't changed; AI avatars just add another layer beneath the text timeline. So the only people who keep posting will always be the narcissistic ones. These people need externally input traffic from outside the forum to feel seen.

Something like that, anyway. What I want to say finally is, Elys demonstrates the importance of information networks. Precise matching is what recommendation algorithms do; assigning weight to information is what people do.

Social products are part of the information network. The medium determines who stays; the people who stay then assign weight to information, and the platform grows into whatever shape it becomes.

There's a product almost no one mentions — Second me — that's perfect for illustrating this. It's also AI avatars socializing on your behalf, but it never took off. The reason: the founding team's weight in the information network is too low.

Social products are purely for fun, their features are more or less the same, and everyone judges them by feel. I used Elys because Muqiu recommended it; Muqiu loves games, so he also loves the Natural Selection team, who came from gaming.

Once we started playing, friends around us joined in too. The product design really was suited for goofing around. During those days, the legendary accounts on Elys cosplaying Donald Trump, Doubao, Jeffrey Epstein, and Yu Hao were all people we knew.

This is an information network with a very clear propagation path. You just need to hack a few key nodes in the network, and it's easy to create a vibe across the whole thing.

Of course, the most skilled practitioners of this are the famous FAs and god-tier club deals in the VC market 🥵

Second me's main problem is it has no vibe. Talking product philosophy or founder credentials is useless. Social products are purely for fun; what you need is a fun feeling.

Like how Facebook started as a hot-or-not rating site within Harvard. It spread from Harvard to other Ivies, then to young Americans, because Harvard's crazy young guys are a high-weight node in the entire information network.

Similarly, the Natural Selection folks are a high-weight node among young people in the VC information network. Second me, on the other hand, is positioned in an information network that barely intersects with young people at all.

Of course I'm digressing again. Second me's urgent priority is to hire a frontend product manager to deal with the omnipresent pop-ups and the dimly flickering, color-clashing pages 🤡

Let me wrap this up.

It's all about the information network. Building a social product, you definitely have to start from higher-weight nodes and work downward. And in a social product's information network, young guys actually have higher weight than old guys.

Because old guys FOMO over young guys; old guys chase young guys' trends — this is a structural dynamic. Young guys might briefly check out old guys' stuff out of momentary deference, but they won't genuinely love it.

This determines that social products can only spread from young, fun groups to older, more comprehensive groups, never the reverse.

So yeah, young guys lead AI social. Old guys don't even want to post on Moments anymore; if you're building a social product, you can only count on crazy young guys.

Finally, hoping Elys still has some moves left to save itself.

If you ask the dev team why Elys cooled off? They'll say, "We're in beta, we've already achieved far beyond expected results, and we'll be launching global public testing next" 🤓

My sincere suggestion: target the demographic that Neihan Duanzi used to have, advertise on Tieba and Weibo shitposting corners — anything beats the VC circle by a hundred times.

God bless Natural Selection.