Chapter 04 · The 30-Day Plan

Four weeks.
One thing working.

No experience needed. One small step at a time. By the end of the month, one annoying task in your week runs itself, and you understand exactly how.

New to the words? Every underlined term has a plain definition. Or skim the basics first. The Basics Open the Glossary

The journey

Each week has one job.

Do the steps in order. Skipping ahead is the number one reason this fails. One to three hours total per week is plenty.

Understand it, then pick your target

No building yet. Just clarity.

Read The Basics once. It takes six minutes. You only need the three big ideas, not every detail.
Pick one boring task you repeat every week (a follow-up email, a weekly summary, sorting inbound messages). Small is good.
Write that task down step by step, like a recipe, so a stranger could do it without asking you anything. Do it with a colleague who keeps asking "why."

By Friday: a one-page recipe of a task you do all the time, and a clear head about what an agent actually is.

Make your first reusable helper

Still nothing connected. Safe to experiment.

Build your first skill (a saved set of instructions the AI reuses every time, so you stop re-explaining yourself). Start with your writing style.
Use the built-in helper that builds skills for you. It asks questions; you answer in plain words.
Test it by hand: paste a real past example into the AI and check it does the job the way you would. Fix the wording. Repeat until it is right.

By Friday: one working skill that makes the AI sound like you, every time, with no copy-pasting instructions.

Let it touch one safe tool

Real work now, but nothing a customer can see.

Connect the AI to one low-risk tool you own (your notes app feeding your to-do list, for example). This connection is called MCP; you do not need to understand the plumbing.
Let it do the task on real inputs for a full week. You still press the button each time. You are watching, not trusting yet.
Each time it is slightly off, tighten the recipe and the skill from Week 2. This tightening loop is the whole game.

By Friday: a one-job agent doing real work in a place where a mistake cannot hurt anyone.

Give it a little rope, carefully

Earned trust, not blind trust.

Look back at the week. Which part did it get right every single time? Which part still needs you?
Let it do only the part it nailed without asking you first. Keep reviewing everything else.
Add a once-a-day glance to make sure it still behaves. Then pick your next boring task and start again at Week 1.

By Friday: one task partly off your plate for good, a habit for adding more, and proof this works.

The four rules under all of it

If you remember nothing else.

RULE 1

Write it down first

If you cannot explain the task on one page, the AI cannot do it. The page comes before any tool.

RULE 2

Start where it cannot hurt

First runs happen somewhere only you would notice a mistake. Customers come later, once it has earned it.

RULE 3

Watch before you trust

You check the output every time until it is right many times in a row. Then loosen up slowly, one piece at a time.

RULE 4

Smallest version that works

One task, one tool, one helper. You can always add more next month. Most people try too much and quit.

So you can dodge them

The five common trips.

The trip	What it costs you	Do this instead
Skipping the write-it-down step	Weeks in, nobody can tell why it broke	One-page recipe before any tool
Giving it access to everything	One small slip touches your whole business	Only what this one task needs
Trusting it too soon	A quiet mistake piles up for weeks	Watch every output until it earns slack
Starting too big	It gets messy, you give up	One task, one tool, one helper
Installing random tools you found online	Hidden instructions you did not write	Stick to trusted, official sources

Before it runs without you watching

Five plain yes / no questions.

Could a stranger run this task from my written page alone?

Did I test it on real past examples before connecting anything?

Can it only touch the one tool this task needs, nothing else?

Is there a moment I review before anything I cannot undo?

Can I look back later and see exactly what it did and why?

All five yes?

You are in a safe place to let it run. Any no? Stay one step back until it is a yes.

Keep going

When you finish week four.

These practices are common across the tools most people already use. Nothing here is locked to one product. Loop back through whenever you add the next task.

REFRESH