How to Hire Your First AI Employee
If your AI pilot went nowhere, the problem almost certainly was not the model. It was that nobody wrote down the job.
You are not implementing AI. You are making a hire. Everything you already know about bringing someone on applies: you write the job down before you fill it, somebody checks the work the first week, you look at the output on Friday, and if it is still wrong in a month you do something about it. Skip those four steps with a human and you get chaos. Skip them with an AI assistant and you get “we tried it, it didn’t really work.”
I spent last week at the AI Business Retreat 2026 in Spooner, Wisconsin, presenting four sessions to a room of business owners and operators. Rod Holum, CEO of Coulee Tech, invited me up to run the leadership and marketing tracks. Rod is also the author of the AI Business Maturity Model, the framework the whole retreat is built around, which sorts companies into five honest stages from AI Drifter up to Physical Integrator. Most businesses I talk to are stuck one level below where they think they are, and it is almost always for the reason below.
What follows is the core of the first session, written so you can run it yourself without a developer, an IT department, or a budget.
Why do most AI pilots fail?
Ask a room of owners how their AI pilot went and about a third will describe the same thing: enthusiasm, a few weeks of poking at a chat window, then silence. The post-mortem usually blames the tool.
It is almost never the tool. It is that the pilot was aimed at a department instead of a task. “Handles customer service” is a thing you want. It is not a thing you can hire on a Tuesday, because nothing starts it, nothing comes out of it, and on Friday nobody can tell you whether it happened.
What actually counts as a job?
A job has a trigger and an output. Compare these:
Not a job: answers the phone. Handles customer service. Does our follow-up.
A job: drafts the reply to the five emails a week asking whether we service Trane units. It starts when the email lands. It ends with a draft sitting in the office manager’s outbox. You can check it in under a minute.
That second one is small enough to fire on Monday and hire again on Tuesday. That is the bar.
The job is usually hiding in plain sight. It tends to be the thing your best-paid person does at seven in the evening because it never fit in the day. Drafting the same reply twenty times a week. Answering the question you get daily. Chasing quotes that went quiet. Meeting notes. The onboarding checklist nobody ever wrote down.
The six-line job description
Six lines. That is the whole thing.
- The job, in one sentence. One task, not a department.
- What starts it. An email lands, a form comes in, it is Monday morning.
- What it hands back. A draft, a list, a filled-in form. Never “an answer.”
- What good looks like, plainly enough that someone can check it in under a minute.
- What it may never do. Send, price, promise, sign, or talk to a customer alone.
- Whose name is on it. A person, not “the office.”
Line five is the one people skip, and it is the one that saves you money. Line six is the one that makes the whole thing real.
Why do you need two versions of it?
Here is the part that separates the businesses still running this in six months from the ones quietly fiddling with a prompt.
Part A is the contract. Six lines in plain English. It is what you hand to a person, and it is what you read on Friday.
Part B is the employee. Part A rewritten so an assistant can actually execute it, with the real facts filled in so it is not guessing. Prices, service area, names, brands, hours. Anything with a number in it.
Part A alone is a nice note that never runs. Part B alone is a prompt in somebody’s clipboard that nobody can check, hand off, or fix six weeks from now when it starts drifting.
When something is wrong on Friday, change Part A first, then rebuild Part B from it. Fix the job description, not the answer. Everyone who skips this ends up adjusting a prompt for six months and calling it a strategy.
If the prompt-writing part is where you stall, our guide on how to write AI prompts covers the Role, Task, Context, Format framework we used all day at the retreat.
What does a bad one look like?
Run your Part B against one real email from last week and read what comes back with a suspicious eye. Three failures show up constantly:
It invented a number. “Repairs like that typically run $400 to $900.” You gave it no pricing, so it filled the gap. Confident, specific, and completely made up.
It promised a date. “We can have someone out Thursday morning.” It has no access to your schedule.
It generalized past the list. “We service all major brands.” Now somebody with equipment you do not touch is expecting a truck.
Any of these going out over your letterhead is something a customer can hold you to. The fix is not a better model. The fix is a never list with real consequences on it, and a named human who reads every draft for the first two weeks.
The five-question done test
You have made the hire when you can answer all five:
- Could you read the job out loud to somebody in the hallway, in one sentence, and have them understand it?
- Can you name the exact thing that starts it and the exact thing that comes out the other end?
- Is there a real dollar figure, a real place, or a real name in there? If it is all generic, it will invent the specifics on Monday.
- Is there at least one thing on the never list that would actually cost you money if it happened?
- Is there a human being’s first and last name on it?
The most common failure is a job description with no facts in it. That produces confident, invented pricing on day one.
What happens on Friday?
Five minutes, once a week. How many did it do? How many went out the way it wrote them? What did your person change, and was it the same thing every time?
Same change every week means you left a line out of Part A. Add it and run Monday again. Nothing changed for two weeks means you can give it a second job or take a training wheel off. Still wrong after a month, let it go. It is a paragraph of text. Nobody has to be told and nobody gets replaced.
That last point matters more than it sounds. The reason this method works is that failure is cheap. You are not buying a platform or restructuring a department. You are writing six lines, watching them for two weeks, and keeping them or deleting them.
The rest of the series
The same shape held across the other three sessions. One messy input, a lot of finished output, and a person’s name on the result. Each one is written up separately:
- Turn one photo into a video in your own voice, and the honest math on renting the tooling versus owning it
- A month of social posts from one job site, because you do not have a content problem, you have a capture problem
- The daily business dashboard you check on your phone, built live on the projector with Tim Holum
There is also a full recap of the retreat, including who presented and the two systems we demonstrated.
Start with one job this week
Pick the smallest job you can stand. Write the six lines. Fill in the real numbers. Put a person’s name on it and let them read every draft for two weeks.
If it works, give it a second job. If it does not, you lost a paragraph of text and an afternoon.
We do this work with businesses across Houston and Dallas Fort Worth, usually starting with the one task that is eating someone’s evening. If you want a second set of eyes on which job to hire first, or you have gotten far enough to need it wired into your actual systems, that is the conversation worth having.