There are things you do in your business that live in your head. This is a simple, repeatable way to teach an AI to do one of them — well enough that you stop checking its work. One rule runs through all of it: the AI does the work, and you stay in control.
There are things you know how to do that nobody can do quite like you. Writing a certain kind of email. Reviewing a deal. Pricing a job. Handling a client the right way. Each one of those is a capability — a skill that lives in your head.
AI is good enough now to do a lot of these for you. The hard part was never whether it can. The hard part is handing it over — getting the job out of your head and into the AI so cleanly that it does it right every time, and you trust it enough to stop hovering.
That is what this method is for. It works the same way every time, for any capability.
You give it instructions — everything it needs to do the job the way you would. There are six parts: what the job is, why it matters, how it’s done, what it needs (its information and tools), what a good one looks like, and how to check it. Those six parts are the AI’s full instructions for that one skill.
Once you see that, the whole method fits in three sentences:
You write the instructions. The AI follows them. You watch and correct until you trust it.
The rest of this page walks through each part — what the instructions look like, what you do, what the AI does, and how trust gets built.
The instructions live in a folder — just like one on your computer. One skill, one folder, six files inside. Each file is one part of the instructions, written in plain English. Click any file to see what’s inside. This is a real example: finding ten therapists to recruit in Arizona.
Add 10 people to our recruitment spreadsheet, and prepare a draft message for each one.
Recruit a new team member for our behavioral health unit in Arizona. The sooner we find the right people, the sooner we can serve more patients.
| Name | Location | Draft message | |
|---|---|---|---|
| Maria Lopez | Phoenix, AZ | linkedin.com/in/marialopez | “Hi Maria, your work in behavioral health around Phoenix caught my eye. We’re growing our Arizona team and I’d love to share what we’re building.” |
| James Carter | Scottsdale, AZ | linkedin.com/in/jamescarter | “Hi James, your background in adolescent therapy stood out. We’re hiring for our behavioral health unit and I think you’d be a strong fit.” |
You don’t write these from a blank page. A guided tool walks you through each part and writes the instructions for you.
Handing over a skill has two sides. You pass it over; the AI performs it. The two sides mirror each other, because what you write down is what the AI reads.
The first six steps on each side meet at the six files. You write them. The AI reads them.
| You write it by… | The file | The AI uses it to… |
|---|---|---|
| Defining the job | The Job | Know what to do |
| Giving the reason | The Intent | Know what it’s for |
| Showing the steps | The Process | Do the work |
| Equipping it | The Context | Gather what it needs |
| Modeling one done right | The Benchmark Example | Compare its work |
| Setting the bar | The Checklist | Check itself |
The seventh step on each side isn’t a file. Your Trust happens after the folder exists. The AI’s Start and Hand off wrap around the work. That’s the next part.
Trust is not a switch you flip. It’s earned by watching the AI not mess up. Think of a new self-driving car. At first your hands hover near the wheel and you watch every move. Once it proves itself, you relax and let it drive. Same thing here. This is the loop that gets you there.
Capture is the AI getting better — like the car getting a software update. Review and Release is you getting comfortable — you stop hovering over the wheel. The work gets better and your trust grows, together, run by run.
The AI moves up a level each time it goes several runs with nothing to correct. It moves back down the moment the mistakes return. Trust goes up and down with proof, not with hope.
Say the AI gives you a result and it’s off. You don’t have to dig around. Every kind of “wrong” points to one of the six files. You ask one question — what kind of wrong is this? — and the answer tells you which file to open.
| What’s wrong with the result | What it’s really telling you | Which file you fix |
|---|---|---|
| It did the wrong thing entirely | It didn’t understand the job | The Job |
| It did the task but missed the point | It didn’t know why it mattered | The Intent |
| It used wrong, old, or made-up information | It didn’t have the right answers | The Context |
| It skipped steps or did them out of order | The how is off | The Process |
| It’s the wrong shape, format, or tone | It had no good one to copy | The Benchmark Example |
| It let a mistake through it should have caught | The bar was too low | The Checklist |
Wrong thing? Job or Intent. Wrong information? Context. Wrong steps? Process. Wrong look? Benchmark Example. Should’ve been caught? Checklist.
Whatever you fix, also add a line to the Checklist — so that exact mistake can never slip through again.
Fix the file so it does better next time. Add the check so it catches itself next time. That’s how the skill keeps getting sharper instead of repeating the same mistakes.
You fill the six files. The AI runs them. The Trust Loop builds your confidence, run by run. The Fix Map tells you exactly where to go when something’s off. And your hands move from on the wheel, to hovering, to off — as the AI earns it.