And how I finally learned to write.

Writing can be a traumatic experience for me.

When I was a child I remember sitting at the desk my Dad made for me trying to write thank-you letters to family members and experiencing what I can only describe as pain. The effort of serialising thoughts into neat, linear sentences and then committing those thoughts to handwriting was enormous. It felt like having to do a gigantic jigsaw puzzle with your hands tied behind your back and using your nose and mouth instead.

Later, in secondary school, I had an English teacher called Mr Mitchell (my favourite, but don’t tell him that). I suspect I was a constant source of disappointment to him – he was consistently positive about my work but also relentlessly frustrated by the lack of it.

I remember him saying to me one day that the thing about my work was that I could say more in half a page than other students managed in three. He just couldn’t mark it properly because it was too short. That clear disappointment has stuck with me. I wish I could have done what he wanted.

My lack of output wasn’t laziness, lack of ideas, or indifference. It was dyslexia.

The realisation

Years later, a friend who was a teacher with experience of special educational needs helped me to see it. I have a lot of the standard symptoms but have developed a series of coping mechanisms which tend to mask them in most cases. I miss words out, sometimes entire sentences, get words in the wrong order and spelling is a constant challenge even with spell-check. However, with this understanding I’ve also realised I have an array of advantages.

They say that people with dyslexia can be good storytellers. I think it’s true for me, but what interests me more is why.

My brain works unusually. It thinks things through in long, complex trains of thought, with each element needing to connect in logical order before I can express it. It feels a bit like doing a huge three-dimensional jigsaw puzzle in your head. I collect the available pieces, spin them around in space to see how they fit together, and discard what’s not needed.

Generally I find it easiest to communicate these thought trains verbally, but if one of them gets interrupted halfway through, it can feel genuinely catastrophic. Anyone who knows me well will be familiar with the look of pain on my face when it happens. It’s not eye-rolling, it’s shock. The carriages have derailed, there are arms and legs all over the track, and I’m trying to work out how to clean up the mess.

All of this means that when I do write something, it tends to be well thought through but also very dense and compact. My brain tries to use the least words possible to get the point across. Ironically, this compression can appear almost machine-like, which means my writing tends to be flagged by AI detection tools even if I don’t use them.

Don’t get me wrong though, I use AI heavily, and I want to tell you about why and then how.

A desire to share

Many of you will know about my brush with death about eight years ago thanks to ‘the bad cancers’.

Needless to say, that experience had a profound impact on me, and one of the consequences has been a desperate need to try to capture the learnings of the last thirty years in text. I’m not entirely sure why it feels so important. It might be a desire to leave something, ego, or the sense that I learned some things the hard way and I’d prefer if other people didn’t have to. I suspect it’s all three.

However, thanks to my dyslexia, this process has been painful. I had the thoughts, felt the need, struggled with the transcription, and abandoned it half-written.

What changed wasn’t my thinking. It was generative AI.

Working with the robot

Like a lot of people in my orbit, I went deep on AI about two years ago. It was quickly obvious that something significant had happened. I ran experiment after experiment, refining my approach with each iteration and gradually improving the results.

Like most people I started simply, getting it to summarise things for me, analyse bits of text, and even draft some ideas which I then adapted.

The breakthrough, though, wasn’t asking AI to write for me – it was asking it to interview me.

The workflow

The first step was defining the role I needed it to take. I didn’t want a cheerleader or a summariser. I wanted an experienced journalist with deep knowledge of games, web, mobile and business. Someone who would take a working title and ask what triggered it. Someone who would keep asking clarifying questions until I explicitly told them to compile a draft.

That brief became the foundation of a dedicated project which I set up in ChatGPT and now treat like an editorial workspace.

I gave this project a series of examples of my writing and, before drafting anything new, asked it to analyse my existing work. It was to identify consistent tone, recurring themes and structural patterns, but also suggest areas of material improvement. We iterated until we had a usable style guide. That went into the project memory alongside the interviewer brief.

My first test was a piece on scarcity, abundance and lean. It didn’t work well – the AI rushed ahead and altered my thinking too much – so I treated that as part of the setup. I described exactly what behaviour needed altering and asked it to generate a clearer set of behavioural instructions: don’t standardise my phrasing, don’t resolve uncertainty too early, don’t draft until a concrete observation has been supplied, prioritise structure over polish.

I added those instructions to the project, started again, and it worked.

Since then I’ve written around twenty essays in this way. Each one still takes between one and four hours. This isn’t push-button output. The judgement remains mine. But the mechanical bottleneck has largely gone.

The key benefit of this process is that I don’t have to hold the entire train of thought in my head anymore. I can write a bit, submit it, leave it, and come back later to answer the next set of questions. The AI holds the place for me even when my attention doesn’t.

I guess you could say that generative AI has become a cognitive prosthetic for me. It doesn’t generate the insight, it stabilises the chain long enough for me to articulate it. It reduces the transcription cost without diluting the thinking. It acts less like a ghostwriter and more like an infinitely patient editor who keeps asking, “What do you actually mean?”

I understand why many people are sceptical or concerned about the impact of AI. Without a clear point of view to guide it, AI can generate convincing noise at speed, which is potentially dangerous for all sorts of reasons.

But here’s the distinction that matters.

For someone with compressed thinking, it can unlock backlog, remove the friction between knowing and expressing, and lower the cognitive cost of articulation. It doesn’t make me more intelligent, it just makes me more legible.

So, dear Mr Mitchell, I know you wanted more pages (sorry about that, Sir), but what I needed was a different interface.

---

Footnote

Tracing the production of this piece.

1. I opened a session stating that the next article would begin with an anecdote about my dyslexia.
2. I supplied the Mr Mitchell story in rough form, including the derailment metaphor and the near-death motivation.
3. The AI reflected back what it thought the piece was about and asked me to clarify the central claim.
4. I said I wasn’t sure of the claim, but described the practical outcome: roughly twenty essays completed in a month using an interview-style workflow.
5. The AI proposed a structural spine and asked whether to foreground the mortality driver. I said yes.
6. I then described in detail the workflow: defining the AI’s role, creating a project, uploading PDFs, analysing my style, generating a style guide, adding behavioural instructions, and restarting the scarcity piece when the first attempt failed.
7. A full draft was produced.
8. I flagged missing specifics about project setup and issues with rhythm.
9. I made significant structural and tonal edits to the draft, tightening sections and reframing the opening.
10. A light proofread was completed to address spelling and grammatical issues.
11. This version reflects that final pass. The thinking, judgement and final wording decisions remain mine.

P.S. Originality.ai rates this piece as being 56% original – one of my highest marks yet. Not totally surprising given how careful I was with its production but it’s still amusing given the subject matter.