Hi everyone, I’m Shengxiong (Leo). A 35+ practitioner who transitioned from the FMCG industry into hands-on AI work, now focused on “AI-driven new-media customer acquisition.” Tap to follow me and see real-world AI new-media experience.
This morning, I finally got my AI content creation system working end to end.
I say “finally” because I’d been working on this for a while, on and off, trying plenty of dead ends. Today, for the first time, it ran cleanly from my input all the way to a publishable article without getting stuck — and that felt almost unreal. So this is what “smooth” feels like.
Broken down, though, it isn’t that complicated.
My Two Tools: Obsidian as the Brain, WorkBuddy as the Hands
The whole system has just two roles.
Obsidian is my knowledge base. It holds the 100-plus articles I’ve written over the years. They’re not just “published work” — they’re my corpus, the real “me” that AI can draw on when it writes.
WorkBuddy is where the actual output happens. I open an Agent in WorkBuddy and give it a few simple sentences about what I want to write. For example, in the morning I just copy over the “takeaways” section from my daily journal. That single line of input is all it takes; the rest is handed off to it.
About three minutes later, a WeChat article is ready.
Note that those three minutes aren’t spent staring at the screen. I go get a glass of water, reply to a message, and by the time I’m back the article is there. In the past, starting from a blank document, just deciding “what to write today and how to open” could eat up half an hour.
Machine-Written Text Shouldn’t Be Published As-Is
But one thing needs to be clear: this is entirely machine-written.
Of course it draws on things I’ve written before — after all, my old articles live in Obsidian. As you read, you’ll notice some words aren’t ones I’d use, some phrasing isn’t how I normally talk. So after generation, I always do a manual pass:
First, I swap out the words I wouldn’t use for my own phrasing.
Second, I do fact-checking. Ordinary new-media articles usually don’t need it, but anything touching finance or news, I have the AI verify the data and claims line by line — making sure nothing is fabricated, conflated, or misattributed — before I proceed.
After editing, I send the article back to WorkBuddy, and it gets archived in the system. More importantly, the AI learns from my edits — what I deleted, what words I swapped, and why — so the next time it writes, it’s a little closer to me.
This “generate → I edit → it learns” loop is, to me, the part of the system that’s genuinely valuable. It’s not a one-off ghostwriter; it’s a collaborator that becomes more like me the more I use it.
Creation Is No Longer the Barrier — Efficiency Is
Once it was working, the clearest conclusion in my mind was this: for today’s AIGC creators, creation itself is no longer the barrier.
In the past, writing one article could take me three or four hours front to back. And finishing wasn’t the end — the next day I’d often do another round of edits, because things always look off after a night’s distance.
Now? Move a bit fast and three minutes gets an article done. Add a cover image and it’s ready to publish within ten minutes.
From three or four hours to three minutes. The gap isn’t “I got better” — it’s that the process changed. My energy shifted from “getting the words out” to “judging what’s worth writing, whether the edits are right, whether the argument holds up.”
I’m increasingly convinced that what separates people going forward isn’t who can write better, but whose system runs smoother, who hands the repetitive labor to AI. Everyone can use the tools; the difference is in how you wire them into your daily work.
In Closing
So this article is both a record and a conclusion:
Everyone trying to pivot into new media really needs to use this Agent-style approach, to free their time and creativity from repetitive labor. It’s not about letting AI think for you — it’s about letting AI do the things you’ve long known how to do but aren’t worth spending three hours on again.
My system just got working; it’s still immature. But I can see the direction clearly. I originally built this system on Trae, and now it runs just as smoothly on WorkBuddy. Some friends might ask why not use Codex or Claude Code?
Because transitioning into a one-person company (OPC) isn’t easy, and every penny has to be spent where it counts — and Trae and WorkBuddy are both free right now. For daily article writing, free is more than enough.
What’s your article-writing workflow these days? Feel free to share in the comments.
By the way, I recently started an “AI new-media buddy” group — it’s full of people doing AI new media together, troubleshooting and solving problems as they come up. We’re already 20 members in. If you’re interested, you can join me; the details are in this article:
I Decided to Start an AI New-Media Buddy Group — First Cohort Is Just 5 Yuan
