I Was Making Five Figures on WeChat. I Walked Away.

One

On the day I left my job in March, I put my work badge on the front desk. The security guard glanced at me and said nothing.

On the way home, I counted: I had registered on 11 platforms in China that anyone could name. I applied for four WeChat official accounts. Two of them eventually took off, each bringing in five figures. That sounded okay. But after three months, I found myself doing the same thing every day: wait for a hot topic, find three sources, translate the English into Chinese, move other people's material here, and add two sentences of my own judgment.

Once I was writing about an AI topic. I searched for a long time and found an "in-depth analysis" that looked decent. Then I followed the trail and found the original: an English blog post published three days earlier by an independent developer in Berlin. The Chinese version had cut the entire passage where the author said, "I'm not sure this direction will work," and left only the conclusion. I stared at the screen for a while and closed the document.

I didn't write anything that night. I was thinking about one question: what I fed my readers every day. Was it information, or a steamed bun someone else had already chewed over?

Two

In July, I registered an X account, @leo_xiaolei.

The first week, I basically couldn't understand anything. The timeline was all English, abbreviations, slang, and the comments section was a mess. I used translation software sentence by sentence, then guessed at the context. There was a post about Grok Bot. I looked at it for forty minutes, looked up eleven words, and in the end understood only one thing: this person was talking about pitfalls he had hit himself, not a secondhand summary.

That feeling was strange. Like you've been walking around a nicely decorated shopping mall, and suddenly you're dropped into an open lot. No one gives you directions, but no one decides for you what you're allowed to see.

I started chewing through one paper a day. Not trying to understand all of it. Just grabbing one point: why he did it this way, what he tried that didn't work. If I couldn't understand something, I left it and came back the next day. Three months later, I posted my first technical thread. An overseas developer replied, "this is useful." I screenshotted it and kept it.

I don't have any grand narrative about English. It's just that one day you realize: you want to know the full story behind something, you can't find it in Chinese, and you can find it in English. That one point is enough.

Three

The aquarium business was where I first experimented with AI.

Before, I bred guppies, arranged planted tanks, and wrote "Aquarium Little Guide" posts with images. I could scrape by each month. But it was too draining. Changing water, controlling temperature, watching algae, packing, shipping. I was around the fish all day. After the pandemic, spending on entertainment and leisure dropped. Relying on breeding and volume was a dead end.

So I changed the system. Claude Code connected to OpenAI's API scanned Xianyu and aquarium forums at high frequency every day. With the species and price range set, whenever it found something below the threshold, AI automatically messaged the seller. Once it reached the payment step, the system notified me. I bought it and told the seller to hold it for two days before shipping. Meanwhile, AI had already listed the goods on my platform at a new price. If it sold fast, I sent the new buyer's address directly to the original seller. Drop-shipping. If it sold slowly, then it came back to my site for temporary holding.

From "breeding and selling fish" to "matching information flows and money flows." One person, one set of scripts, doing the work of three people watching the boards.

In June, I joined a security company as an AI architect (FDE). Day to day, I evaluate Claude, Gemini, Grok, VoxCPM; I write things in Rust, configure Agy CLI plugins, manage Claude Code's idle compaction configuration, and maintain an open-source "hand-drawn talking-head video .skill" workflow. In October I wrote "FDE Is Dead! From FDE to FDU," and after it went out, some people shared it and some cursed me out.

Separately, none of these things look big. But strung together, they mean one thing: AI isn't for chatting. It's for watching the market for you, replying to messages for you, running processes for you. Every step you stop watching gives you back a block of time. That time is better spent chewing through papers, writing threads, or arguing with people than sitting by a fish tank.

Four

I graduated in 2016, caught the tail end of online education, and worked there for a few years. In 2021, after the "double reduction" policy, I left. I switched to pure front-end and got a 700k annual salary. After a full year, I was laid off, with not a cent of performance bonus.

On the day I packed up, I peeled off my keyboard cover, rolled it up, and threw it in the trash. On the way home I bought an Americano, sat outside the convenience store and finished it, opened a job app, scrolled for ten minutes, and closed it.

It's not that I never thought about stability. But looking back, stability came from the other legs I kept outside the job.

I'm still updating my WeChat account, still updating X, the aquarium system is still running, and this weekend I'm holding my first offline training session. I don't know which leg will break first, but I know the odds of all of them breaking are smaller than the odds of breaking with only one.

Five

Temporary holding is the part that annoys me most right now.

If AI doesn't flip the fish within two days, they have to come to my place. Once they're at my place, light assets become heavy assets. I have to change water, control temperature, turn on plant lights, remove surface film, pack and ship, and watch what other people are pricing on the platform. Fish are alive. They die, get sick, lose color. You can't just hang them there and ignore them.

A friend said, why not play the "refund if it doesn't sell in two days" game with the seller? I said that's called ruining your reputation. I'd rather not take the order than do that.

In 2027 I want to either optimize this step out or replace it. The goal is simple: let the system move only information and money, not fish. The fish stay with the seller; the buyer gets them from the seller. I'm just the person in the middle matching the price. If I can't do that, I'll narrow the categories and only do ones that sell steadily. If that doesn't work, I'll stop. I won't get tangled up with physical assets.

Six

In 2026 I changed the way I live. From big-company development to watching how the underlying business logic turns. I've recovered a bit of the joy of writing code, even with the FDE hat still on my head.

This weekend is the offline training, on X and Grok Bot. Not many people. A dozen or so. I prepared three pages and plan to talk through the rest on the spot, see where the conversation goes.

In 2027 my main direction is overseas: overseas websites, overseas self-media. I'll keep one or two domestic platforms, but they won't be the main focus. Energy is limited, and I want to spend it where I can see positive feedback.

The Pacific is huge. I can't swim, but I have oars.

The oars are English, AI, and that system I've got running.

Tomorrow I still have to get up early to feed the fish. If that batch of guppies in the temporary tank doesn't sell, I'll have to change their water.