Think before you ask

I built a tool that asks me questions because I was getting uncomfortable with how much I could ask a machine to do for me.

Writing was the obvious one. An AI could give me an argument, a structure and a clean paragraph before I’d properly decided what I thought. The result could look finished while I was still at the beginning.

I don’t have evidence that this was damaging my brain. I have a simpler concern: I could get very good at choosing between other people’s sentences and spend less time learning to make my own.

That distinction matters when the point of writing is to work something out. The awkward first attempt contains information. Where I hesitate. Which example I can actually explain. Where I’m repeating something I’ve heard without understanding it. A polished answer can step over all of that.

Ponder puts the order the other way round. I get a question and answer it out loud. Sometimes the answer is messy. Sometimes it’s short. Sometimes I realise the question is wrong. Then the system can organise what I said, challenge it or connect it to something I’ve written before.

I still get help. I just have to produce something for the help to work on.

One of the questions asked what I was outsourcing that I should still be practising. My answer was writing, and then reflection. Thinking about things outside the businesses as well as inside them. That was useful precisely because nobody had supplied the answer first.

There’s a contradiction here I want to keep visible. I build automation because I enjoy getting tedious work out of the way. If a tool can sort my notes or find the relevant passage, I’m pleased to hand it over. I don’t need to suffer through every task to deserve the result.

But some tasks contain the practice I wanted. If I ask for the conclusion before I’ve made an attempt, I can remove the difficulty and the reason for doing it at the same time.

The distinction depends on the job. When I need a transcript organised, speed is useful. When I’m trying to understand why I want to move somewhere, the pause is useful. I’m not looking for an elegant answer as quickly as possible. I’m trying to discover which answer I can stand behind.

I don’t always hold the line. Sometimes it’s easier to ask for a draft and edit it. Sometimes that’s exactly the right thing to do. What I want to notice is when convenience becomes the default for a skill I still claim to value.

So I’ve kept a small requirement: before asking the model what I think, I say what I think. A paragraph, a voice note, even a few uncertain sentences. Enough to expose the gaps.

Then I can ask it to disagree. Ask where the argument fails. Ask which claim has no evidence. That conversation is much more useful when there’s a position of mine inside it.

The question can come from the machine. The first attempt stays with me.