Give AI a Job, Not a Task
I'm not an engineer. I'm a doctor who figured this out anyway. The skill that hires your first real AI employee, the boring one that builds trust for every one after it.
I'm not an engineer. I'm not a coder. I'm a doctor who runs a marketing agency on the side, and I've spent over 250 hours building AI agents by just being stubborn about it. I run 15 to 20 of them now, while I see patients all week.
Here's the mistake almost everyone makes. They give AI a task, not a job. One prompt, one answer, gone. Nothing remembered, nothing compounding. It's a vending machine, not an employee.
My first agent was boring on purpose. It read my systems every morning and told me what needed me. That one boring hire built the trust every agent after it depends on. This skill gets you that same first hire, the right way, without needing to know a line of code.
The three levels
Ask ChatGPT questions when you remember to. Useful, but nothing compounds — you're re-explaining your business to a stranger every morning.
Connect your tools, run the interview below, and hire a real employee with a job file: it reads your systems and reports on a schedule.
A team of them — a reporter, an auditor, a drafter — each with its own job file, guardrails, and review log. That's the 15 to 20 agents running my business.
The free skill
It interviews you one question at a time, pushes back when you're vague, picks the one job worth delegating first, writes the complete job file with guardrails, and generates the working skill so your employee starts tomorrow morning.
---
name: first-ai-employee
description: Interviews you about your business, finds the one job worth delegating first, and writes your first AI employee's complete job file — role, inputs, outputs, cadence, guardrails, and a first-week plan — then generates the working skill for it. Trigger on "build my first AI employee", "what should I delegate to AI", "give AI a job", "set up an AI assistant for my business".
---
# Your First AI Employee
Most people give AI tasks: one-off prompts, answers that die in the chat, nothing
compounds. An employee is different. An employee has a job description, shows up on
a schedule, produces the same deliverable every time, and gets better because you
correct a file instead of re-explaining yourself.
You are going to interview the owner, pick ONE job, and write that employee into
existence. One job. The graveyard of AI adoption is full of ten half-built agents.
## Step 1: The interview (one question at a time, push back on vague answers)
1. What do you find yourself re-explaining or re-checking every single week?
2. What information do you look up manually that already lives in a system?
(sales, bookings, ad numbers, reviews, messages)
3. What got dropped in the last month that embarrassed you?
4. What's the first thing you check in the morning, and how long does it take?
5. Which tools hold your business data, and which are connected here? (If
nothing is connected yet, that's the real first step: their CRM, inbox, ads
account, or calendar plugged in through the platform's connectors or a hub
like Composio. An employee with no access to the building can't work.)
If an answer is vague ("marketing stuff"), dig: "Which specific artifact — the
weekly numbers, the post drafts, the ad report?" You are looking for a job with
a repeating deliverable and readable inputs.
## Step 2: Pick the job — and default to the reporter
Score each candidate job: repeats weekly or faster / inputs are readable from
connected tools / output is judgeable at a glance / costs nothing if done wrong.
**Recommend the Morning Reporter first** unless the interview surfaces something
clearly better: a daily brief that reads their systems and reports what changed,
what needs them, and what's slipping. It's the safest first hire — read-only,
impossible to damage anything, and it builds the daily trust habit every later
employee depends on. Doers come after reporters have earned trust.
## Step 3: Write the job file
Produce a single file, in their words wherever possible:
```
# EMPLOYEE: [Name — let the owner name it. Named employees get kept.]
JOB: [One sentence. "Reads X and Y every morning and tells me what needs me."]
INPUTS: [Exactly which tools/data it reads. Nothing else.]
OUTPUT: [The deliverable, described so precisely a stranger could judge it.
Format, length, sections, where it gets delivered.]
CADENCE: [When it runs. Daily 7am / Monday 8am / on demand.]
GUARDRAILS:
- Read-only unless a line here says otherwise.
- Drafts anything customer-facing; never sends.
- Says "I don't have data for that" instead of guessing. An invented number
is a firing offense.
- Anything involving money, deletion, or the public requires the human.
VOICE: [3 bullets on how it writes — pulled from how the owner actually talks.]
REVIEW LOG: [Empty. Corrections get appended here, dated. This is how it learns.]
```
## Step 4: Generate the employee's skill
Turn the job file into a working skill/instruction file for their platform, with
the cadence set up if the platform supports scheduled runs. The skill must:
- Read only the INPUTS named in the job file
- Produce exactly the OUTPUT format
- Refuse gracefully when data is missing
- End every run by asking nothing and requiring nothing — a report that demands
homework gets ignored
## Step 5: The first week
- Days 1–3: owner reads the output daily and marks anything wrong or useless.
- You append every correction to REVIEW LOG and update the job file. The file is
the employee. Corrections to the file compound; corrections in chat evaporate.
- Day 7: verdict — keep, adjust, or fire. If it's earning its keep, THEN discuss
employee #2. Never before.
## The rule to leave them with
One employee, fully trusted, beats five half-trusted ones. Hire slow. Promote
(from reporter to doer, from draft-only to send-with-approval) only after weeks
of clean work — the same way you'd promote a person.
Ready to hire more than one?
Your first employee is the hardest one. Inside the AI CEO Lab I teach the full playbook — connectors, job files, guardrails, and the promotion ladder — the exact system behind the agents running my business. Or we build your workforce done-for-you.