Where these ideas come from

The four people whose thinking shaped how I use my assistant, what each one gave, and which pages carry it.

I did not invent most of what is in this section. I borrowed, and I would rather tell you from whom, because their originals are better than my compressions and you may want them. Four people, roughly in the order their ideas show up in a working week, and then a note on what is actually mine.

Geoff Woods and CRIT#

Geoff Woods wrote The AI-Driven Leader, and the one thing from it I use every day is CRIT: context, role, interview, task. Give the assistant the context it needs, tell it whose chair to sit in, let it interview you before it starts, then set the task. The interview step is the part most people skip and the part that changes the result most, because it makes the assistant ask for what it is missing instead of guessing at it. I liked it enough to build the whole first hour around it: /onboard is the interview step run at the scale of a brain.

Where to find him: aileadership.com, the home of the book.

Where it shows up: the CRIT method, the interview, prompting habits.

Nate Herk#

Nate Herk teaches people who do not code to build with these tools, on YouTube, at length, with more patience than I have. Five things of his are in these docs, and I want to be exact about them because I have caught myself presenting one as my own.

The four Cs: context, connections, capabilities, cadence. I mapped them onto the four things you actually touch in this app: the brain, the Connections page, the Skills page, and the schedule on each skill. The prompting habits: manage it like a person, say what and why, make it prove it. The "interview me" prompt, which I adapted into the grill-me prompt. The adversarial council: opposed stances, a separate judge, a verdict of green-light, reshape or kill, and the cheapest test before you commit. And the phrase autonomy equals risk equals cost, which is why the trust ladder on what it will not do climbs so slowly. The council is the single most useful thing I have added to my own practice this year.

Where to find him: youtube.com/@nateherk.

Where it shows up: how your assistant thinks, prompting habits, find what you do not know, what it will not do.

Greg Isenberg#

Greg Isenberg's line is that products can be copied and communities cannot, and his ordering follows from it: audience first, then community, then product. It is a founder's frame and I keep it for one use in these docs. When the research recipe is feeding a decision about a new offer, "who is already listening, and who do they talk to" is a better first question than "what should I build". A garden-design studio with two hundred past clients who recommend it has the first two before it has thought about the third.

Where to find him: gregisenberg.com.

Where it shows up: starter recipes.

Andrej Karpathy#

Two ideas, and they sit under everything else. The first is that what the model has in front of it matters more than the exact words you typed: the work is in assembling the right context, not in polishing the prompt. The second is the LLM wiki: a body of plain markdown pages that the model reads and writes, organised so it can find things, that gets better the more you use it. That is what the brain is. Eight layers of markdown in a repository you own, read at the start of every session, grown by correction and capture.

Where to find him: karpathy.ai.

Where it shows up: how your assistant thinks, grow the brain, who edits what.

What is mine#

Less than it looks, and I would rather list it than let you assume.

The eight layers the brain is organised into, from what you are for down to how you work and write. The guides I wrote for the people I coached, which these pages are a rewrite of: a skill is a saved recipe, the rule of two (done it twice, about to do it a third time, that is a skill), the meta-prompt that asks the assistant to audit your own library, and the idea of building in waves rather than all at once. Earn before you act: read, then draft, then act, and the lowest rung that solves it. The four house rules on what it will not do, which are mine only in the sense that I learned each one the hard way. And the voice loop, where the edits you make to a draft are the lesson.

Where it shows up: write your own skill, starter recipes, teach it your voice, grow the brain, a day with your assistant, the glossary.

If you take one thing from this page, take the originals. Woods for the method, Herk for the practice, Isenberg for the order to build in, Karpathy for why the brain is files. This section is the version I could fit on a shelf beside a working week.