Efficiency Is Sometimes the Problem
Here’s something a bit awkward: when I write this blog, a lot of the drafts are actually organized for me by AI.
It’s not laziness—it’s just that it’s genuinely fast. I throw in a rambling mess of thoughts, a half-formed idea, an article I found interesting, and it churns out a fully structured draft in seconds. It also auto-translates into English, auto-publishes, and auto-backs up. The efficiency is staggering.
But that’s also the problem:it’s so efficient it doesn’t sound like me.
A Semi-Automated Blog and an Ideal Pipeline
The way I write this blog now is, frankly, semi-automatic: I come up with ideas, AI drafts the article, and I make edits. If I can improve it, I do; if not, I move on—it sounds decent enough anyway.
But “decent enough” and “sounds like me” are two different things.
My ideal pipeline would look something like this:
- Record what I’ve done—an agent quietly monitors my computer, logging my actions, the pitfalls I’ve fallen into, and the technical decisions I’ve made. That’s raw material for technical writing.
- Record what I’ve said—automatically saving my posts on Telegram and other platforms. That’s raw material for casual reflections.
- Write in my voice—feeding the AI articles I’ve written over the years to fine-tune it to my style (or distill it into a style skill), so the AI doesn’t just “know how to write” but actually “writes like me.”
- Settle into blog form—technical pieces and casual musings each find their proper place.
- Auto-distribute—pushed through publishing pipelines to Zhihu, WeChat Official Account, X, and other platforms.
One seamless flow from “my life” to “my blog,” with no manual work on my end.
But Style Can’t Be Rushed
Think about it, and the hardest part in this pipeline isn’t technical—it’s step three: getting the AI to write like me.
Style isn’t a template. It’s not just “use more short sentences,” “throw in an occasional self-deprecating remark,” “bold key phrases.” Those are surface-level tricks. Real style is the sediment of every pitfall I’ve stumbled into, every judgment I’ve made, every bias I’ve carried, crystallized in text. The AI can learn my sentence patterns; it can’t learn why I think the way I do.
I recently heard a podcast that introduced an idea, and it pried open a crack in the wall I’d been bumping into.
The approach goes like this: take an article with distinctive style—call it A. Rewrite A into a bland, straightforward draft that carries the same information but strips away all flavor—call it B. Then lay A and B side by side and, one by one, write down how B became A: why this word was swapped for a colloquial one, why this sentence was split into two, where the self-deprecating remark landed and why not elsewhere.
What you’ll discover is that style is actually a translatable rulebook, not some ineffable sense of language.
That’s a lot more reliable than “make the AI imitate me.” Imitation is a black box: feed it ten articles and hope for the best—you never really know if the output lands or where it falls short. A conversion table is a white box: if something doesn’t feel right, you fix that specific rule; new material comes in, you draft a dry B first, then apply the table to produce A—and every article is auditable.
So the path probably looks like this: pick a few of my best, most natural articles as A, honestly rewrite them as B, then map out the transition as rules. Those rules become a skill. The AI uses it, and I keep correcting—not vaguely saying “this doesn’t sound like me,” but pointing at a specific rule and saying “this is wrong.”
Until one day, whatever it produces, I can claim it with genuine pride.
Final Thoughts
This article was also put together with AI’s help—ironic, right?
But at least it remembers: my goal isn’t to make the blog faster, it’s to make it more me. The next step is breaking down “sounds like me” into a written, editable conversion table.