Free starter guide · Kevin Unk

The AI Operating System

Build the AI system that runs your business. No code required. A chatbot answers questions; a system does the work. This guide shows you how to build one, step by step.

5layers every AI-run business has

3agent patterns to copy

7day build calendar

3copy-paste prompts

Keys: J/K next / previous section · N next steps

00 · The shift

Why systems, not just chatbots

Most business owners who "try AI" do the same thing: open a chatbot, ask a few questions, get a decent answer… and nothing changes. The chatbot sits in a tab. Three weeks later they still spend every afternoon buried in the same busywork.

A chatbot is a tool. What changes your business is a system: tools wired together to do a job end to end, with a memory of who you are and an owner reviewing the output.

Chatbot

Answers
Waits for you to ask
Forgets you after the session

AI Operating System

Does the work
Runs on a schedule, watched or not
Remembers your offers, clients and decisions

An honest note: no "10× your revenue in a week" here, and no fake numbers. Building this takes a few focused days, a written description of your business, and the habit of reviewing output before it ships. What it gives back, on repeat, is hours every week.

01 · The picture

The 5 layers of every AI-run business

From a solo operator to a forty-person team, it's the same five layers. Tap each one.

One line: memory is what it knows, agents are what it does, tools are how it touches the world, Mission Control is how you oversee it, deployment is how it runs on its own.

02 · Step 1

Find your 5 highest-leverage tasks

The 3-day time audit. For three working days, list every recurring task you touch: what it was, how long it took, and whether a robot could do it with the right instructions. Then score each one below. The scorer ranks them for you, and your list stays saved in this browser.

TaskHow oftenMinutesSame every time?Score

Classic first agents

Content + follow-up

Turn raw ideas into on-brand posts on a schedule. Answer new leads fast and get them booked.

Also high-leverage

Onboarding + research

Welcome new clients and collect what you need. Turn a question into a sourced briefing.

Don't automate (yet)

High stakes + consent

Anything a client relies on for a personal, high-stakes answer, or messaging people who never opted in.

03 · Step 2

Build the memory layer

A system is only as smart as what it remembers. Give an agent a blank slate and you get generic answers. Give it your offers, your process and your voice and it sounds like a well-briefed assistant.

Your knowledge

SOPs, services, pricing, process. Not written down? One "how we do things" session is the highest-value hour in this guide.

Your business data

Offers, clients, common objections, content that worked.

Your decisions

When you change a price or process, note why. Later, the system uses it to avoid old mistakes.

The pitfall: memory rot. A price you changed three months ago that's still in the system produces confidently wrong answers. Update the source doc the same day you change anything customer-facing.

04 · Step 3

Give your system hands

A system that only writes the email still leaves the sending to you. Connecting agents to email, calendar, CRM and files (what technical people call integrations or MCPs) is how the worker reaches their desk.

Golden rule: one integration at a time, tested before it touches a paying customer.

First

Email + calendar

Where follow-up and booking agents live.

Second

Your CRM

New leads flow in and every touch is logged.

Third

Files + docs

So the system reads and produces the documents you use daily.

05 · Step 4

Deploy your first agents

An agent is a worker you've hired. It needs a clear job, inputs, tools, a human-review point, and an escalation path for when it's unsure. Build the review point in from day one.

Pattern 1

Content pipeline

Inputs: a raw-ideas doc
Tools: docs, images (not publishing at first)
Review: you approve each post
Unsure: asks you instead of inventing

Pattern 2

Lead follow-up

Inputs: new lead from CRM or form
Tools: email, calendar, booking link
Review: you approve the templates up front
Opted-in contacts only

Pattern 3

Research brief

Inputs: a question or topic
Tools: web search, notes
Review: skim before acting
Flags uncertainty, links sources

Want these built around your business instead of from scratch? That's what Kevin does 1:1. It starts with a one-hour discovery call.

Book the $97 call →

06 · Step 5

One screen: Mission Control

The more you automate, the more you need one place to see what happened. Your role becomes review gate, not do-the-work.

mission-control · today● all systems

▸ brief what happened overnight + what needs you today ▸ reviews output waiting for your sign-off ▸ health which automations ran, which failed ▸ revenue new leads · booked calls · closed deals

07 · Step 6

Measure, iterate, scope

One metric per system

Content: posts published per week
Follow-up: time from new lead to first touch
Onboarding: signup to fully onboarded
Research: briefs you actually used

Weekly 30-minute review

What did each system produce?
Which metric moved, and why?
What broke or got weird?
What one change improves next week?

Build vs. buy. Build what your business is built on. Hand off what matters but would eat weeks to learn. When a build starts eating your week, getting help is just good economics.

08 · Your plan

The 7-day build calendar

About an hour or two a day. Tap a day when it's done; your progress is saved.

09 · Ground rules

The no-slop checklist

What AI can't do alone

Replace your judgment
Work without a written business
Handle high-stakes contact without a review point
Invent facts: it must cite or flag uncertainty

Before anything goes live

Every source doc is current
Every agent has a job + review point
Every tool tested safely first
Messaging runs on consent
One metric, reviewed weekly
The system can say "I don't know"

10 · Use today

Copy-paste starter prompts

Paste into any capable AI assistant. Replace everything in [brackets].

01 · "How we work" doc (Day 2)

Interview me, one question at a time, to write a "How We Work" document for my business. Cover: what I sell and the exact prices, who my best customers are, the questions they ask before buying, my process from first contact to delivery, my tone of voice, and things I never say or promise. When you have enough, write the document in plain headings and bullet points. Mark anything I did not confirm as [UNCONFIRMED] instead of guessing.

02 · Score your task list (Day 1)

Here are the recurring tasks I logged over three days, with minutes per task: [paste list]. Score each one 1-5 for frequency, time cost and repeatability. Rank them, recommend my top five to automate first, and for each say what a human must still review. Flag anything that involves contacting people, money or legal consent as "review required".

03 · Agent job description (Day 4)

Write a one-page job description for an AI agent that does this job: [one sentence]. Include: inputs it works from, the tools it may use, its exact output format, the human-review point before anything is sent or published, and what it must do when unsure (stop and ask, never guess). Then give me three test cases, including one where the right answer is "I don't know".

11 · Next steps