How I built my client’ s brain in 3 hours
I just delivered the highest-leverage coaching session of my career.
I'm not so good at the whole coachy questions thing.
My forte is actually doing things.
I built Rens a personal AI that sounds like him — and will get smarter every day for the rest of his life.
The raw material
Rens is a big dog exec coach. Published in the Harvard Business Review. But also a humble, down-to-earth guy. A dear friend.
I've been coaching him for over a year. During that time, without either of us quite planning it, we'd accumulated an enormous library of material about him.
Coaching calls auto-transcribed by Gemini. Summaries generated automatically. Over months, patterns surface — what he avoids, where he lights up, where he goes flat. Things neither of us would notice in a single conversation become obvious across twenty.
On top of that, he'd done every exercise I give my coaching clients. Vision documents describing his ideal life in five years, cascading down into annual, quarterly, and weekly intentions. Daily habit tracking. Deep journalling. Inner inventory work — fears, desires, shadow, the things you don't put on your professional bio.
Then there's everything he's written professionally. HBR articles. LinkedIn posts.
All of it just sitting there. Waiting.
The three documents
I took everything and leveraged hundreds of hours of AI experimentation — courses, two years of daily use, thousands of conversations with Claude where I've refined what works and what produces generic drivel.
All of that experience condensed into a few hours of focused work.
I created a Claude project for Rens and built three documents:
Comprehensive profile. Everything we know about him, inside and out. Background. Career. Pains. Patterns. Ambitions. Blindspots. What makes him tick. Fascinating insights I hadn't personally spotted.
Custom AI instructions. How Claude should respond to Rens specifically. His communication preferences. What he finds annoying. Where he needs to be challenged. How much he values tough love versus sympathy. The equivalent of the detailed custom instructions I've built for myself over two years — tailored to his psychology and working patterns.
Two voice style guides. One for his personal writing, built on the inner work we've done together. One for his professional writing, built on his LinkedIn and HBR articles.
Most people think of "my voice" as one thing. It's not. Your reflective journalling voice and your Harvard Business Review voice are different animals. Both deserve training.
The reveal
When Rens saw the voice style guide on our call, he chuckled. "Man, this is really on point."
Then he reached the section on language he'd never use: "aggressive metaphors, spiritual vocabulary, beast mode discipline language, excessive superlatives, corporate buzzwords used vacuously."
"It's quite fun what it picks up on."
The results were remarkable. He's now using this system daily.
Three hours of my time. A permanent shift in how he operates.
Why it compounds
This isn't a one-off trick. It's infrastructure that gets smarter.
I call the mechanism the Compounding Corpus Loop. You write. AI analyses your patterns and produces drafts that sound more like you. You refine and feed back what's working. AI adjusts. The next output is closer. Every piece of writing — every journal entry, every coaching call, every email — feeds the system. Every word compounds.
This is why journalling has become funner. The inner work and the AI work become the same work.
This article is proof.
It started as a ten-minute rambling voice note.
I fed the transcript to Claude — which already knows my writing style, my background, my voice, and has detailed instructions on how not to sound like AI. I've threatened it, saying I will have an affair with Grok if it produces "let me provide you with a comprehensive overview."
I have template playbooks for different formats. LinkedIn posts. Newsletter articles. Claude knows the structure, the rhythm, the word count. It knows all too well that I am a terrible British snob who curls his lip at any Americanisms.
Claude produced a draft. I edit. I cut, sharpen, add — the jokes, in particular. My goal is for AI to remove all the boring stuff so I can operate primarily as a court jester.
Then I'll repurpose this into a LinkedIn post. I've made a playbook that knows how to optimise for LinkedIn — which I've stripped of anything that would make me hate myself for posting.
AI handles architecture. I provide soul.
The ten minutes of rambling became a publishable article in under an hour. Without the system, this would have taken three hours. Or — more to the point — it would have joined the graveyard of things I meant to write about but never did.
Why you won't build this yourself
You could, though.
If you sat down in a perfectly quiet room with no meetings, no inbox, no children, and committed a full week to it — you could probably build something close to what I built for Rens.
That's a big "if," right?
You know this. You've read the articles. You've seen what AI can do. You've probably experimented with custom instructions and thought "this could be so much better if I actually invested the time."
But that week doesn't exist. And "probably figure it out" is not the same as doing it.
This is the pattern I see with every high-performing client. The gap between knowing how and actually doing it. It's not an intelligence problem. It's an implementation problem.
And the systematic, obsessive tinkering required to build these systems properly is a particular kind of madness that most people don't have — and shouldn't need to have.
I happen to have it.
I should confess something. I am blessed and cursed with the compulsive optimiser gene. As a kid playing Age of Empires, I spent 90% of my time designing the battlefield and 10% actually playing. Football Manager? Hours planning transfers and tactics. I didn't actually play the game. These were some of the early signs that convinced the psychiatrist to give me an autistic certificate that enabled me to live mask-free during Covid (mwahaha).
AI scratches the same itch. I spend hours building, cultivating, and refining systems for every aspect of my life and my clients' lives. The prompts. The architecture. The feedback loops. The custom instructions that get updated every time something doesn't land right.
And I've come to believe this is itself a creative act. A genuinely complex, demanding field of creativity that most people haven't clocked yet.
Some aspects of creativity are less important than they were before AI. Boo-hoo. Get over it. New ones have arrived.
What you can steal from this
You don't need my particular brand of madness to start. Here's what actually moves the needle:
Train AI on how to talk to you. Most people skip this entirely. Tell it your communication preferences, what you hate, what you value. My free exercise 10 Minutes to Make AI Less Annoying walks you through this — it's step one and it's the highest-leverage ten minutes you'll spend.
Separate your voices. You don't write the same way in a journal as you do in a board memo. Create different projects or instruction sets for different contexts. Even a rough separation immediately improves output quality.
Feed it everything you've already written. Emails, presentations, journal entries, LinkedIn posts — you already have a corpus. You just haven't fed it to anything yet. Upload your best writing and ask AI to identify your patterns. You'll be surprised what it sees.
Give it a "kill list." Tell AI what you never want to see. Specific phrases. Tones. Habits. Rens's AI knows never to use "beast mode" language or vacuous corporate buzzwords. Mine knows that if it says "sacred transmission" or "dive into," I'm filing for divorce.
Stop correcting the same mistake twice. When AI does something you don't like, don't just fix it in the moment. Update the instructions so it never does it again. This is the compounding part. Most people skip it, and then wonder why AI keeps disappointing them.
Use AI to find your blindspots. Feed it enough of your writing and ask: "What patterns do you notice? What do I avoid? Where do I hedge?" It'll spot things about you that you can't see yourself. That's what happened with Rens — the profile surfaced insights I hadn't personally spotted in a year of coaching.
Remember: AI handles architecture. You provide soul. The goal isn't to outsource your thinking. It's to remove the friction between your thinking and your expression. You still provide the raw material — the ideas, the experiences, the taste. AI just stops it from dying in the gap between "I should write about that" and actually writing it.
Inside Deep Writing — fear not, the doors are already closed, so I won't attempt to hoodwink you into giving me money — I break this entire process into steps over 100 days. Personalised. Hands-on. I do the fiddly architectural work so my clients can focus on the writing.
The current cohort is full. If this resonates, there will be more.
