The challenge
I was spending too much of every AI conversation re-explaining myself. Who I am, how I write, what I consider a finished piece of work, and which mistakes I can’t tolerate. The assistant would get close, and then I’d spend another twenty minutes correcting the same habits I had corrected the week before.
The problem wasn’t the model. It had no durable picture of the person it was working for.
The approach
I started treating my own working knowledge as something to document, the way a company documents a procedure. The system grew in layers.
The first layer is communication. It’s my writing system, released as version 1.0 in August 2026 and kept unchanged underneath everything else. It describes my voice, my editing workflow, and separate guidance for different kinds of writing.
The second layer covers how I work and decide. I wrote a personal operating manual that describes, as working hypotheses, how I tend to think, lead, learn, and make decisions. Alongside it sits a set of guardrails, which are practical checks the assistant should apply where I have predictable blind spots.
The third layer governs the whole thing. A master file sets the order of authority when instructions conflict. Law, safety, and factual accuracy come first. My current instruction for the task comes next. Style preferences come much further down the list, so a sentence is never made to sound better by weakening a fact.
How challenge works
One part of the design I’d point any organization toward is the idea of graduated challenge. The assistant uses the lightest form of pushback that fits the consequence of being wrong. Routine work gets done directly. A consequential or hard-to-reverse decision gets its assumptions, alternatives, and downside made explicit before a recommendation. The assistant stops and asks only when a missing fact would materially change the result.
That keeps the assistant from rubber-stamping my ideas and also keeps it from turning every small request into a debate.
The development
The deliverables
- A master operating standard with a defined authority order and file map
- A writing system with a style guide, a drafting workflow, voice profiles, and discipline guides
- A personal operating manual and an AI guardrails file
- A framework for approved gold-standard examples, used to show voice and judgment rather than to be copied
- A changelog, with every version preserved beside the one that replaced it