Stage 1: False Starts

I first picked up Gen AI about six months ago. At a public sector product meetup, a speaker showed an app they’d built with AI: a simple arrow that always points to the centre of the galaxy. It looked effortless. I wanted to try.

My own experiments were clumsy. I don’t code, and the outputs I got were hobbyist at best. I used ChatGPT to help me create an app to integrate with the Spotify API to build a better podcast search. Nothing I made was particularly useful.

But something else happened: I realised Gen AI didn’t have to be my app-builder. It could be my thinking partner. Where I brought experiences, it brought reflection. Where I had fragments, it helped me see concepts. That shift, from tool to partner, is what stuck.

Stage 2: Building a Mirror

Following my instincts, I tried something different. I didn’t want to plagiarise, and I didn’t want generic outputs. What I did want was to make more of what I’d already written over the years.

So I trained ChatGPT on my own publicly available writing. I asked it to zoom out: to look for themes, patterns, models, tone of voice. Then I asked what became the most powerful question of this whole journey:

“Permission to be honest, play devil’s advocate. What are the strengths and weaknesses in my ideas? What am I saying that just repeats others? And what do I offer that even you, ChatGPT, don’t already contain?”

That’s when something relly interested happened. I wasn’t just experimenting anymore. I was shaping a way of working. Over time it became clear that I was building what I can only describe as an AI constitution: a set of principles baked into my interaction with ChatGPT. It carried my unique perspectives, my ethics, my tone of voice.

Stage 3: From Notes to Guides (and a Book)

With a constitution in place, I started pointing Gen AI at new writing. I had a newsletter with a steady audience, but something nagged at me. Most of what I was publishing felt more like field notes (raw observations from the front line of product in the service sector) than field guides, the kind of structured, actionable guidance that helps others.

So I experimented. With ChatGPT, I studied my most guide-like outputs and noticed a pattern:

  • A real story — grounded in lived experience.
  • An abstracted model — the bigger system or principle behind the story.
  • A practical takeaway — something a reader could use the next day.

That became the structure. I brought raw notes and experiences; ChatGPT helped me find the unique angle, shape them, and make them usable. The result? Three monthly newsletters that doubled subscribers, tripled monthly reads, and, most importantly, sparked real engagement: comments, direct messages, invitations to speak.

I wasn’t outsourcing my thinking. I was sharpening it. I wasn’t getting generic content. I was creating ideas more uniquely my own.

With that confidence, I turned to my ultimate field guide: my book.

The draft was already mine: written, tested in talks, refined through 20+ early readers. But now I wanted to finish. I gave ChatGPT the manuscript and asked the same core questions I’d been refining for months:

  • Permission to be honest. Play devil’s advocate.
  • What models and patterns do you see?
  • What am I saying that’s already out there?
  • What is unique and valuable?
  • Where are the strengths and weaknesses in the narrative?

The feedback, layered with early reader comments and my own instincts, gave me what I needed. I took myself to an Airbnb for a weekend and finalised it.

By then, my constitution had grown symbols and shorthand, metaphors that acted as anchors. The most useful? The relationship between Tony Stark and JARVIS in the Marvel films. It isn’t perfect, but it mostly holds.

The Constitution in Practice (Stark ↔ JARVIS)

1. Stark sets the mission.

  • The human defines the purpose, direction, and ethics.

2. JARVIS scans the system.

  • AI zooms out, spots patterns, themes, and blind spots.

3. Stark brings instincts.

  • Judgment, lived experience, and emotional truth stay human.

4. JARVIS sharpens coherence.

  • AI structures ideas, stress-tests narratives, and keeps tone consistent.

5. Stark acts, JARVIS reflects.

  • Final responsibility, authorship, and delivery are mine.

6. Together, they amplify.

  • Not replacement, not plagiarism.
  • A co-creative loop where judgment × pattern-recognition creates clarity and movement.

Stage 4: Using AI Without Losing Yourself

Looking back, what started as a failed attempt to build an app has become a working practice. I don’t use Gen AI to think for me. I use it to help me think. The constitution I’ve built — part ethics, part tone, part shorthand — keeps me anchored in my own voice.

That’s the key lesson for me so far: AI can expand your practice without erasing your self. The danger isn’t plagiarism or automation. The danger is letting the machine’s voice become yours. My way through has been to bake myself into the process, so that what comes out is not generic, but more recognisably my own.

This field note is an example. The story is mine. The reflections are mine. The structure, the sharpening, the coherence: that’s where ChatGPT plays its part. Writing it this way enacts the very constitution I’ve been building: a loop of experience → reflection → coherence, without losing my own voice.

Use AI. Don’t lose yourself.