How I use AI to think, not to think for me

AI can be an extension of your thinking or a substitute for it. The difference lies in how you use it, not in the tool itself.

I’ve spent the last few months trying to find an honest workflow: one that uses AI without losing the thread of my own ideas. What follows is what works for me, limitations included.

The problem with AI as a shortcut

When you use AI to generate the final text, the definitive analysis or the argument’s structure, you skip the process where real learning happens. Thinking isn’t in the result — it’s in the journey.

Writing slowly is thinking. Delegating the writing is delegating the thinking.

This doesn’t mean AI is bad. It means using it as an output generator rather than an interlocutor fundamentally changes what you get from it. And what you lose too.

The biggest risk isn’t that AI writes badly. It’s that it writes well, and you don’t realise you haven’t thought anything at all.

My real workflow

I write first. An ugly draft, full of gaps, with raw ideas. Unpolished, unstructured, without thinking about whether anyone will read it. Just ideas on the page.

Only then do I open the chat.

I use it for three specific things:

Asking questions I hadn’t asked myself. I give it the draft and ask it to identify the assumptions I’m taking for granted. It usually finds three or four I’d overlooked.

Identifying weak points. I ask it to argue against my thesis. Not to convince me, but to see where my argument is fragile. Sometimes it’s right. Sometimes it helps me see why I’m right.

Exploring counter-perspectives. If I’m writing about productivity, I ask it to give me the best argument from someone who believes all productivity culture is counterproductive. It broadens the frame of what I’m thinking.

In no case do I ask it to write the final text. That part is always mine.

The prompts I use

Some I use constantly:

“What questions should you ask yourself before publishing this?” — My favourite. It usually returns three to five points I’d missed.

“What assumptions am I taking for granted in this text?” — To see the water I’m swimming in.

“Argue against me. Give me the best counter-case.” — To strengthen the argument before publishing.

“Where is this text weakest? Where is it most vague?” — To refine, not to delegate.

None of these prompts ask it to think for me. They ask me to think better.

What I never delegate

The conclusion. The voice. The position.

If I don’t know what I think about something, AI can’t know for me. And if it tells me, what I have is its position, not mine, even if it’s written in the first person.

AI is an extraordinary interlocutor. It’s patient, available at two in the morning, and doesn’t tire of you asking the same question five different ways. But interlocutor is not the same as author.

The difference matters.