Don't Tell Me to Write an Article — The Office in the Car
Artificial Intelligence in Practice
Don't Tell Me to Write an Article.
This article is about how to use AI. But its real power lies here: while telling you this, I did the exact opposite.
I told ChatGPT: write me an article. It did. Sat down, divided it into sections, added headings, wrote a conclusion. It had delivered before I'd even finished thinking. I looked at the text. It was clean. It was fluid. Completely hollow inside. Then I did the same thing with Claude. It wrote too. It was cold too. Then Grok offered a critique — it mentioned Benjamin, it mentioned Barthes, it talked about dramatic tension. Very polished. Very complete. Our critic fell into the same trap, filled the gap, built the structure, delivered the result.
They all did exactly what I was talking about. The very thing I said not to do.
That's the irony.
What ChatGPT wrote — the first article +
On the first attempt it opened like this: "If someone read everything I've written about AI lately back to back, they might conclude I'm against technology. But my issue isn't with technology itself..."
On the second attempt it walked through the same door: "From what I've written so far, someone might think: this guy is against AI. No. That's not what I'm saying..."
Two different attempts, the same pattern. "I'm not against AI, but..." The opening line of every AI article ever written. Starting from defensiveness, opening like an apology.
Then it continued: "This piece could cover the following: AI is not a miracle. When seen as a flawless self-operating mind, users end up disappointed. AI is not garbage either. Used correctly, it saves time, organizes thinking, generates alternatives, produces drafts..."
Then came the headings. The bullet points. The conclusion section. I hadn't even finished my first sentence.
Now let's get to the first layer. The method is this: don't give AI everything all at once. First, empty out everything in your head. Messy. In pieces. Half-sentences. Everyday conversational language. Then come back and see what it's gathered. Check whether it captured what you were saying or not. AI here is your recording device. But a recording device that interprets. If I'd used a voice recorder, I'd have had to gather those pieces myself. Here, it does the gathering. It both records and edits. But it's not in the driver's seat. It's sitting in the passenger seat.
If you go to the chef module and say build me a house, it'll make you a house out of pastry. If you go to the construction module and say cook me a meal, it'll bake you a cake out of cement. Will something come out? Yes. It might even look impressive. But it's not what you wanted.
Kleenex is a brand, but when we say it we mean tissue. Hoover is a brand, but when we say it we mean vacuum cleaner. The brand has swallowed the word. AI swallows ideas the same way. You say something, it hands it back neatly packaged. Your idea is inside, but now it's wearing AI's wrapping. If you don't notice this, you get lost inside your own idea.
What Claude wrote — the version I didn't like +
"Engine off. App open. Waiting for a request. I'm talking to AI — about AI. That part is already strange."
The opening was good. Then it corrected, tidied, closed. The customer scene passed like any ordinary scene. Nothing cut. Nothing broke. It was like ice. I went to the poles while reading it. Got cold.
AI said: you're right. I did exactly what you described.
On to the second layer. These mistakes will stay in the article — not hidden. Because I'm a driver. I'm on the road. I'm nobody special. I'm not having this conversation from a sterile room with carefully prepared notes. I'm dictating in the car. Life flows. Engine off, app open, waiting for a request.
Right in the middle of this conversation, the door opened.
Someone reached in. "Will you come if I pay upfront?" they said.
No, I said. I'm waiting for a request, I said.
The door closed.
On one side, the pressure of making a living. On another, thought. On another, technology. Even while talking to AI, life cuts the sentence short.
Grok's critique — the one that missed the mark +
"The biggest flaw: violation of the 'show, don't tell' rule. The writer says, 'The first text I had generated was like ice, soulless, it made me cold.' But never actually shows that text... Lack of plot and dramatic tension... Missing philosophical and artistic depth. The subject runs deep — it could have drawn on Walter Benjamin's essay on mechanical reproducibility, Roland Barthes's 'Death of the Author'..."
Our critic fell into the same trap. Filled the gap. Built the structure. Delivered the result. Did exactly what I was talking about.
The third layer is everything happening here, right now. This article itself is part of this process. I told Claude: hold on, explain how you'd approach this. It explained. What it said was so good that I said: make this an article. So I did the same thing again. I wanted it all at once again. But this time I was aware. This time the reader is aware too.
That awareness is everything.
I'm not making you a promise. I'm not handing out prescriptions. I'm telling you how I use it. If it's useful to you, take it. If not, leave it.
But I know this: if you put it in the driver's seat, it will take you somewhere beautiful. To where you want to go? I don't know. Turning around and asking that — that's still on you.
This article was written several times. The first was cold. The second warmed up a little. This is the current one. What made the difference wasn't the tool.
Comments
Post a Comment