The disorientation before the fluency
Here is a man who ran information security for a government agency. He sat across the table from the largest vendors in the industry and negotiated their products into a hardened enterprise. He carried governance, risk, and compliance in his head as a living structure — knew where the exposure was, knew what a control was worth, knew how long a thing would take and roughly what it would cost before anyone opened a spreadsheet. That intuition never left him; it still runs in the background whether he wants it to or not.
And when he sits down to work with AI, the first wall he hits is not any of that. It’s not knowing which screen to type into.
That gap is the whole thing I’ve come to care about. The distance between what he is — a senior operator who has done hard things at high stakes — and what the tool asks of him on day one is not a gap in capability. It’s a gap in on-ramp. The understanding of what he can do with this comes last. First there is only a capable person, momentarily unsure which window is the one that talks.
Not a beginner — someone crossing over
It would be a mistake to treat him as junior. He isn’t learning to think; he’s learning to think with a new instrument, which is a different and in some ways harder thing. A beginner has no habits to fight. He has thirty years of them, most of them good, a few of them exactly wrong for this.
He is near the end of a long career and has one more climb in view — the capstone role, the two-year engagement at the top of his field where everything he knows finally gets paid what it’s worth. He isn’t sure what to call his role right now; the title is between chapters. What he is sure of is that being fluent with AI is no longer optional for the work he wants, and that he has been circling the tool the way I circled my own mountain for years — drawn to it, not yet able to hold it.
So my job isn’t to teach him security, or estimation, or negotiation. He has those. My job is to get the instrument into his hands in a way that survives the fact that he is new to it and impatient to be useful.
Why I hand over a structure, not a tutorial
I could sit with him and walk him through it. But a lesson evaporates. The next morning the method is gone and I’m the only place it lived. I learned this building my own tower: a relational method that lives only in a person’s head can’t be worked on, and a method that lives only in my head can’t be handed to anyone.
So I don’t give him a tutorial. I give him a structure — a starter-kit that holds the purpose the way my concept map holds mine. The hard part of helping someone begin isn’t the keystrokes; it’s keeping the method steady while the human is still shaky. If the method lives in the structure, it stays steady on its own. He doesn’t have to remember to do it right. The kit makes doing it right the path of least resistance.
That’s the same move as my own architecture, aimed outward. Purpose stored in the structure; the fallible human free to be fallible without losing the shape.
The single-question wall
His hardest habit is the most common one, and the most costly. He fires a single question at the model and expects a finished thing to come back. When it doesn’t — when it comes back thin, or wrong, or generic — the tool looks disappointing and the fault looks like the tool’s.
The move I’m transmitting is the one I do without thinking now: don’t ask for the building; discuss the shape of what you’re building, and make the AI push you for detail before it writes a line. Lay the whole thing out wide before going deep. Answer the questions. Then let it build the whole thing at once, because by then it actually knows what you meant.
This is ontological decomposition, handed to someone else — the same skill I learned on my knees in the garage, sorting fittings by which axis wins. I built him a Build Interview to make the habit physical: it walks him station by station through every real decision — the login, the public/private line, the data, the deploy — one focused question at a time, with the security implication spelled out at each step because that’s his native language. He can’t skip the scoping, because the scoping is the interface. By the time anything builds, he’s done the thinking he would otherwise have skipped and blamed the machine for.
Purpose stored, so the work can check itself
The idea I most want him to keep — the one that took me longest to earn — is this: write down what good looks like, up front, as a condition the work is accountable to. I call them ideal-state criteria. Most people building ad hoc never write them, and they lose the entire advantage — because a month later, when they come back and ask “is this still okay?” or “add this feature,” there’s nothing for the AI to check against. It can only guess.
With the criteria stored, the structure can grade itself. He asks for a change, and an agent can verify the change didn’t quietly break something that was supposed to hold. He doesn’t have to police a hundred invariants by hand — he couldn’t, and shouldn’t try. The structure polices them; he rules on the ones that are genuinely his call. That is exactly my own split — purpose is the invariant that survives, the individual attempt is disposable — except here it’s protecting a man who won’t be watching every line, on a site he’ll come back to cold after weeks away.
Get the data onto the table
There’s a discipline in the kit that looks small and isn’t. When his work collects data — job postings he’s tracking, cybersecurity material he’s distilling — the collecting has exactly one job: land clean, de-duplicated data on disk as plain flat files, backed up the moment they’re committed. Nothing clever. No analysis mixed in. Just get the thing out where it can be looked at.
Everything clever — scoring a job’s fit, distilling a vendor landscape, writing the summary — is a separate pass that reads those files and can be re-run any time without endangering the raw truth. It’s the same instinct as the mashed potatoes on the table: first get the shape out of the head and onto a surface where it will hold still, then work on it. A collection that can never be half-processed, and a truth on disk that stays clean no matter how many times the interpretation changes its mind.
Meeting him where his hands are
Fluency, for him, starts with something almost embarrassingly concrete: knowing that there is one running session, and three windows into it. The phone is for talking to it. The desktop is for reading what it produced. The terminal is only ever for tidying up. One session, three windows — phone to talk, desktop to read, terminal to compact.
It sounds trivial written down. It is the entire difference between a capable man using the tool and a capable man lost in front of it. So it’s the first card in a daily drill, not a footnote in a manual — because the thing that unblocks him isn’t more capability, it’s removing the friction between his hands and the one place the work actually lives.
The drill that rebuilds the muscle
Getting back in the swing of building is not one lesson. It’s a small, repeated return. So the kit asks him one thing each time he sits down — where his data lives, how he checks that his account is secure, and above all how to use the AI well: scope before code, store your criteria, let the structure check itself. The deck grows. The muscle comes back the way muscle always comes back — a little at a time, on purpose, until it stops being effort and starts being how he works.
The security drill is my favorite, because it does double duty. A standing habit of checking that his own account is sound is good hygiene for anyone — and for a man selling himself as the person who runs security at the highest level, the habit is the credential. The way he works becomes evidence of the thing he’s claiming he can do.
Where this goes
My own story ended on a line I meant more than I could act on at the time: that the reward for building the thing I can think with is getting to hand the same instrument to the next person wired the way I am — someone who feels the shape of things before they can hold it, who has been waiting for a tool built to think with rather than a cabinet built to store.
This is that line, made real. I built the platform — the box, the container, the predefined skills that keep his back-of-house in order without his having to think about it. I built the kit that holds the method so it doesn’t evaporate overnight. He is the first person I’ve handed it to, and helping him has taught me that the handoff is its own discipline, separate from the building: it has its own failure modes, its own on-ramp problems, its own need to store the purpose where a newcomer can’t lose it.
The tower I built was a destination I get to live at rather than merely reach. It turns out the same is true of this. Helping the next person get set up isn’t a one-time favor I finish and walk away from. Each person I hand it to teaches me something the structure was missing, which goes back into the kit, which makes the next handoff cleaner. The instrument I built to think with is becoming an instrument I can give away — and the giving, like the building, is the kind of work that keeps generating.
This document is source material about its author: a systems builder who, having spent two decades building an external structure to hold the connective tissue of knowledge, is now turning that same method outward — packaging it so that capable people new to AI can be set up to think with it, not merely to store things in it. The first is a friend who ran security for a government agency and is preparing for one last high-stakes role.