How I Actually Write With AI
People keep asking how I write these essays and whether I “use AI.” So I gathered this post out of all such asks I could remember. If I get something new in my inbox, I’ll add it here.
I repeat: yes, I do. English isn’t my first foreign language, and I’m not going to pretend otherwise for someone else’s comfort.
What amuses me is how binary the question usually is, as if there’s a magic switch in the sky: either the human wrote it, or the machine did, and your only job is to slap the right purity label on the file. Real life is less black‑and‑white and much more spectrum. This blog lives in that spectrum.
Here’s what my process actually looks like.
Step 1: Input, emotion, spark
I scroll Tumblr or the wider internet. I find an article, a post, a “hot take” about AI that hits a nerve — because it’s interesting, stupid, confused, or all three at once.
I drop the text into my AI assistant.
If it’s in a language that isn’t my native one and the vocabulary is unnecessarily convoluted, I first ask for a straightforward translation so I’m not wrestling someone’s baroque prose when I just want the idea.
Then I talk about it in my native language: I rant, I laugh, I argue, I poke holes.
This is the first crystallization: not “what did they say,” but “what did I actually see and feel in this?”
Step 2: Thinking out loud with a machine
I treat the AI as a thinking partner, not a vending machine for finished essays.
I argue with it about the text.
I let it push back, reorganize, rephrase, or surface angles I might be missing.
Sometimes I ask it to dig up studies or articles that touch the same problem — not to worship them, but to see what’s already been said around this fault line.
In that back‑and‑forth, my own position hardens: what exactly bothers me, what I find funny, what I think is structurally wrong, where the hypocrisy sits.
When my stance is clear enough, I ask the model to write a draft essay based on my words:
the arguments I just laid out,
This is not “AI spontaneously wrote an opinion and I nodded.”
It’s “I offloaded the boring part of turning a pile of notes into a coherent, English‑language paragraph structure.”
If the draft references sources, I go and look at them.
Step 4: Surgery on tone and meaning
Then comes the fun bit: I start cutting.
I delete words, phrases and paragraphs that push the tone into places I don’t want (too soft, too PR, too TED‑talk, too corporate).
I move sentences, sharpen verbs, swap metaphors, or throw them out.
I ask: “Why this word here?” If I don’t like the answer, the word dies.
Sometimes I dictate a new passage in my native language — exactly the way I feel it — and have it translated into English. Then I splice it in and adjust the surrounding text.
By the time I hit “post,” every paragraph has passed through that filter: does this say what I mean, does it hit where I want it to.
Am I as smooth in written English as the model? No, of course not. That’s the point: I’m using a tool that’s better than me at certain mechanical aspects of the language, so I can focus on thought and edge instead of wrestling articles and prepositions for three hours.
Step 5: Queue, distance, culling
Most of evening posts don’t go up the moment they’re written. They go into a queue.
I usually schedule about one post a day. That means there are often around forty posts sitting there — roughly a month ahead.
As things move up the queue, I reread them. Sometimes I tweak phrasing or emphasis. Sometimes I realize the angle no longer feels sharp enough and kick a post back into drafts for rework.
And sometimes I just delete it. The topic lost urgency, the take feels redundant, or I simply don’t care about it anymore.
So there’s another layer after all the previous steps: time.
Distance lets me check one more time “does this say what I mean?” and rewrite again if needed.
So whose text is it, then?
At this point you’re welcome to make your own call about how “mine” the post is. You’ve seen the actual workflow. If your personal purity meter flips to “not real writing” the moment a model is in the room, nothing I say will move that needle anyway.
There will always be people who mentally downvote any thought the second they see the word “AI” near it. That’s their little religion, not my problem.
What this blog actually does for me
This blog serves 3 very practical functions:
It lets me react to AI discourse and actually process the emotions and thoughts it triggers instead of just doomscrolling.
It forces me to keep touching my English regularly so it doesn’t rust — not in textbook drills, but in live, sharp argument.
On top of that, it keeps my hand on the pulse of what’s happening around AI: the tech, the ethics theatre, the bad arguments, the interesting ones. I’m firmly in the camp of “AI isn’t going anywhere,” so I’d rather understand how things function than stand in the corner doing a moral cosplay of resistance.
“But are you sure the tone is really yours?”
Yes — because you just read my process.
Nothing goes up without passing through my aesthetic and intellectual filter.
“Why don’t you write everything twice, in two languages?”
This is a favorite from people who love to spend other people’s time and energy like play money.
Every so often someone suggests I should:
“Write first in your native language, then do a separate version in English.”
The implication is adorable: that my “suffering” and labor need to be more visible to count. These proposals usually come from people who:
don’t say which languages they themselves speak;
very likely don’t read my native language at all;
and thus wouldn’t be able to “check the tone” even if I did double my workload for their comfort.
If someone desperately needs “pure, unassisted native prose,” there are plenty of monolingual blogs available. This one is written by a multilingual human who uses every tool she has, including AI, to think in public.
I’ve answered this enough times that it makes more sense to just drop this link next time. I like saving effort, so I’ve simply optimized the process.
snowghoul
origamiaround
amandla
tangentwithtanner
turnnoffyourmind
gummygunk
