I’ve been running my business through AI every day for a year now.
Not experimenting. Not dabbling. Actually running it — client work, content, code, operations. Every day, without exception.
Here’s what I learned: the prompts are about 30% of it. The other 70% is building the context that makes prompts unnecessary.
The Four-Layer Stack
Over that year, an architecture emerged. Not designed upfront — it evolved from the work:
A wiki for context — the things that don’t change. My philosophy, my positioning, my methodology. The stuff that makes the system understand me instead of giving generic answers.
A database for state — the things that change daily. Client records, projects, deadlines. The operational reality of what’s actually happening right now.
My judgment — twenty-six years of pattern recognition. The “this feels wrong” that no AI can replicate.
The system for velocity — the execution speed. The grunt work that used to fill my afternoons.
Each layer does what it’s good at. Nothing more, nothing less.
The Problem I Was Solving
Early on, I kept having the same experience. I’d open a conversation, explain my business, explain my tone, explain who the client was, explain what I needed. Get something back. Edit it heavily. Use maybe 30% of the output.
Next week, same thing. From scratch. No memory. No context.
Every conversation started at zero.
So I started writing things down. Not for documentation’s sake — because I was tired of repeating myself. The wiki grew from that frustration. The database grew from needing live data. The judgment was always there. The system got better as the context around it accumulated.
The Compound Effect
Here’s what happened over time:
Week 1: I was explaining everything. Every prompt was long. Every output needed heavy editing.
Week 10: Common tasks just worked. The context was built. I started to feel the velocity.
Week 50: High-level instructions triggered cascades of understanding. “Prepare for the client meeting” pulls context, checks the database, drafts the agenda, and identifies issues — because all the pieces exist.
The architecture compounds. Individual prompts don’t.
Read the full breakdown: The Five-Layer Stack: How to Work With AI
Why Thirty Clients, Not Three Hundred
I also published something this week about a decision that shapes everything I do.
Every musician I’ve ever loved made the same choice. Janiva Magness plays rooms of a hundred people. There are five hundred outside who can’t get in. She doesn’t look at that queue and think I need a bigger venue. She looks at the room and thinks this is where the music works.
Prince could fill any stadium on earth. He chose to do residencies in clubs. Because the small room is where you hear every note.
I’m building the same thing. Not a consultancy that serves everyone passably, but one that serves thirty people properly. Where I know the business, know the industry, know what’s working and what isn’t.
The small room lets you play the song the way it was written. No compromise. No translation for the cheap seats.
Read the full piece: The Small Room: Why I Chose Thirty Clients, Not Three Hundred
Where to Start
You don’t need to build what I’ve built. But you need something.
Start simple:
- One document that describes how you talk to clients and what you stand for
- One spreadsheet that tracks your operational state
- Your judgment about what matters
- The system to execute once the first three exist
The document grows. The spreadsheet gets more useful. Your judgment gets augmented by data you can actually see. The system gets better as context accumulates.
And if you want someone to build the web presence while you focus on the business — that’s what I do. Thirty covers. Real work. No filler.
Tony Cooper
We Build Stores
tony.cooper@webuildstores.co.uk
07963 242210
P.S. The Four-Layer Stack works with any AI — ChatGPT, Claude, Gemini. It’s the architecture that matters, not which model you plug in. If AI keeps giving you generic answers, the fix isn’t a better prompt. It’s better context.
