Education & AI for Africa

FROM THE ARCHIVE

Think on Paper First

Published Monday, 3 August 2026 · 3-minute read

Don't hand AI a complex task you haven't thought through. Yes, you read that correctly. Why? I'm glad you asked. Coffee?

AI works best with direct asks. Vague prompts get vague results. If you're lucky, it will be close but not quite.

AI needs you to say, Build me a 3-page website for our secondary school in Mushin. Not just, Build me a 3-page website.

AI needs you to ask for, 3 critical questions that parents of a primary school may ask me at my seminar tomorrow on raising AI-literate children. Not, Tell me what I should say at my seminar tomorrow.

AI needs you to be articulate about what you want it to do for you. Which means first, you have to be clear about what you want. The more complex the ask, the clearer you need to be.

I wear many hats. I run my grandfather's schools and I manage multiple estates. I do strategy, administration, financial management, government liaison, grant chasing, alumni relations, and the daily business of parents, students, teachers and staff. I also do IT development.

My past lives include software developer, business analyst, program manager, and business development. So my past lives have helped and directed how I handle my current responsibilities. Other aspects I have had to learn on the go.

After jumping on the AI bandwagon thinking AI would solve all my challenges, I am realising that is not so. AI is a very competent assistant and it is extremely efficient. But I have to know what I want it to do for me, and sometimes how.

I am currently trying to design a complete management software system for all my roles. But what does that even look like?

I made the mistake of just jumping in and asking AI for what I was thinking as I was thinking it. I told it not to code anything, that we should discuss what I wanted first. We went back and forth till we had something.

Or so I thought. It had the information. It was complex. It had the keywords and the sections. I used it. I mean I tried to use it. I logged one or two things. Tagged one or two things. I thought it had my use cases and I understood the workflow.

It took me two days to admit that while it was multilayered and complex, it just wasn't what I wanted. Two days.

I went back to AI. I listed all that was wrong. AI agreed. It asked me what I wanted to do. Try and fix, or start again.

Start again. I'd rather get it right from the ground up.

You'd think between me and AI we would have learnt a few lessons. We had. Maybe that's why the second attempt took me three days to say no. Aaargh.

The third time I tried Claude Design. It worked quite nicely, actually. This time I was in love. But this third time was not a charm.

Three different approaches failed. The problem was never the tool. The problem was I did not know what I wanted yet, so how could AI build it?

So on my fourth attempt, I leaned on one of my past lives and did it the good old fashioned way. I went traditional before going AI. I sat down and thought and designed on paper first. I sketched entities, workflows, inputs and outputs, and the reports I actually needed. I used mind maps, flow diagrams and use cases. I used anything and everything to clarify my own thinking first.

When I finally had a complete design of what I wanted, how it should work, how I wanted to use it, and so on, then and only then did I ask AI to build it for me.

I haven't given it a complete pass yet but I am still using it. And AI still has to update a few things as I use it.

The truth is, even a competent assistant still needs clear instructions.

What do you think? Will you think on paper before handing AI complex tasks? What do you use AI for, and what do you keep off AI?

Tell us in the comments. Or get in touch and let's discuss.