Understand before you operate
What the tool actually is and does, before it touches your workflow. Ground truth first: read the live system, not the sales page.
The practice · owner to owner
Training & Curriculum.
Some owners want their own people able to do this, so I teach it. The course is built into your engagement, around your company and the tools we made together, and it runs as coursework over nine weeks. By the end, your people have a baseline understanding of how these tools are put together, and enough to start building on their own. Not prompts. Not another subscription.
Understand the tool before you use the tool
Most shops meet AI the same way: buy the subscription, watch the demos, get a week of wow and a month of drift. That’s not a tool problem. The models are the easy part — you have the same ones I do. Building with AI runs on what you believe before what you buy: what the tool actually is, where it’s strong, where it lies, and how your team holds it accountable. Get that structure right and every hour spent with AI leaves something the next hour builds on. Get it wrong and you’re renting magic tricks.
What I teach
Six commitments. Every engagement runs on them — and every one of them was built in the field before it was ever taught.
Infrastructure that compounds
Working sessions that open and close like job sites, records that carry forward, protocols your whole team runs. Value that stacks instead of evaporating.
Concepts, not commands
Your team learns to think in the concepts, not copy prompts. Discovery creates owners; prescription creates dependents. I teach the first kind.
Methodologies and protocols, not just prompts
How work opens, how it closes, how a claim gets verified before anyone trusts it. The prompt is the last ten percent.
What → How → Who → Why
Settle what is true, then how it works, then who owns it — never in reverse. And keep asking why you're building at all. First principles, not fashion.
Questions of force, not opinion
Structured questions that pull out what your organization already knows but has never said out loud. Discovery, not brainstorming. The answers were already yours.
I got here by building. The field agrees.
I built my practice first and read the field second. It turns out the published literature has names for most of what I was already running — and where it has none, I kept my own. That’s the credential worth checking: not that I read about these ideas, but that I arrived at them independently, in production, and can show the receipts.
Ground truth first
the fieldobservability-driven
Check the live system before trusting any document about it. The document lies politely; the system doesn't.
One is an event, two is a pattern, three is a practice
the fieldevidence-graded promotion
Nothing becomes a rule in my shop until it has proven itself three times. Enthusiasm is not evidence.
Clean hands
the fieldseparation of duties
The mind that decides is never the hand that executes. Some workers get zero judgment on purpose.
Designed absence
the fieldsecure by design
The safest way to keep a capability from being misused is to not build it. Absence is a feature you can ship.
Asserted vs. enforced
the fieldnegative testing
A control described in a document is not a control. It counts once you've run the attack and watched it fail.
The close writes the next open
the fieldcontext persistence
No session ends without recording what shipped, what's owed, and where the next one starts. Momentum is a system, not a mood.
The receipts
Rounded down, never up, from the repositories themselves, measured September 2026. Ask and I’ll walk you through any of it.
12+production repositories
4,000+commits
700K+lines of hand-written TypeScript
50+custom AI agents built
Ask to see a measurement and I’ll show you the command, not a slide. That’s rather the point.
Ask anyone you hire about their failures
If you get a shrug, they haven’t operated anything. Here’s one of mine, told straight.
This spring I built a quality gate whose entire job was catching fabricated claims. Its own documentation carried one — it described a guard that had never been built, and the test suite stayed green while I attacked the exact thing it claimed to protect. The fix wasn’t a patch. It was a standing rule that now governs every build in this house: a control doesn’t exist until you’ve run the attack and watched it fail. I keep a named and dated library of failures like that one, and it’s the most valuable asset I own. You trust the tools in your shop that earned their handles by breaking. Same principle.
How it works
no deckThe first conversationOwner to owner: where AI already sits in your building, where the confusion is, what compounding would look like in your operation. If I'm not the right fit, I'll say so and point you at something honest.
your company, your toolsNine weeks of courseworkBuilt around your company and the tools we made together. Your real workflows on the table, your team in the room. They leave owning the ideas, not renting mine.
the baselineBuilding on your ownThe working structure stays with you: the protocols, the records, the verification habits, in your own accounts. Your people start building from there.
- You own everything. Everything lands in your accounts and your documents. If we stop working together, all of it keeps working.
- Concepts transfer; dependence doesn't. The engagement is designed to make me less necessary every month, not more.
- No inflated claims. Every number I show you is measured and dated, and I'll show the measurement. Expect the same standard pointed at your vendors.
Talk first. Everything else follows.
There’s no deck and no proposal template, which tells you most of what you need to know.