"The more you add, the more you're taking away. Know when you need to stop."
A working document. Not a portfolio, not a blog. Practitioner notes from inside.
One operator. One codebase. One operating system that boots in seconds and maintains state across sessions. 26 years of digital craft compressed into infrastructure that makes a solo practitioner move at the speed of a team.
This is the practice of building with AI from the inside. Not tips. Not tutorials. The actual work — documented as it happens, proven by the commit history, refined by the corrections.
The model is a commodity. The situation is the edge.
The method is for hire at We Build Stores — this site is where it gets worked out.
Working notes and the longer arguments, collected: the writing →
Three days after writing that correctness is cheap now, I got the receipt. A corporation tax return prepared, checked and filed from the system that runs everything else here — a £57.95 bill paid the same afternoon, one catch at the entry screen that proves the method, and the strange discovery that paying the taxman now costs more in software than in tax.
I built my business on baked beans and the belief I couldn't afford to do things properly. That belief turned out to be two debts wearing one coat — knowing how to do the books, and the cash to survive — and only one was ever the hard part. Why doing things properly got cheap, why revenue didn't, and why a monthly retainer buys real compliance but never the hours that depth needs.
Nine and a half months into building an AI operating system on Claude Code, I asked it to inventory itself. It found a paying client the roster wasn't counting and fifty pages with no direct route in. Context drift arrives inside year one — and the fix is a discipline, not a tool.
AI Overviews gutted the how-to blog — a hundred successful sites lost 85% of their traffic. The pages that survived didn't describe something, they did something. That line is the whole game now.
Naming a principle after a literary character routes the instruction through the deepest channels in a language model's training data. Why 'Joe Gargery' carries weight in prompt engineering that 'be humble' never could.
AI dropped the cost of building software to a couple of sessions, and the old build vs buy question quietly broke. The story of the helpdesk I nearly built — and the four lines of code that turned out to be the answer.
There are two kinds of AI memory. Remembering your business between sessions is the one everyone works on. The system knowing what it has actually built is the one almost nobody is solving — and the gap is an entire missing layer.
This morning I read thirteen open-source repositories and pulled seven patterns worth borrowing — in one session. The lesson wasn't that borrowing is free. It's that the haul is a function of the substrate you've built, not the source you took it from.
Operators on X keep saying they're four turns into an AI session and the context window's full. The replies miss the actual problem — they've been taught the wrong physics.
Context engineering is the new buzzword. The skill underneath is older and simpler: people burn their AI allowance in two turns and blame the model, when the model is fine — they're paying for synthesis they could have externalised. That's context discipline, and it's the entire skill.
Government departments are funded to collect data, not to present it. The gap between what the state holds and what an ordinary person can use is structural, durable, and almost entirely unclaimed.