The three rules my AI lives by
Before a single guide, this. My AI has a name, a job, and three rules it isn't allowed to break.
I named mine Coach. That sounds silly until you try it. The moment it had a name, it had a job, and I stopped treating it like a search bar and started treating it like a very fast assistant who needs clear instructions. Then I gave it three rules to start with. They're the reason I trust it.
The three rules
It never sends anything
I do. Coach helps with my messages, my posts, my check-ins. Every one of them lands in front of me first, and that's the safety net: the queue itself. Nothing goes out to a real person without me. This one rule is what makes it safe to let AI touch your relationships at all.
It never makes up a number
If it can't see a number, it asks me for it. My monthly close literally stops and says: I need two things from you before I can finish.
It tells me what it checked, not just the answer
When Coach says nothing is due today, it also says how many things it looked at. Silence and an empty list aren't the same thing.
What it hands back
I'm Riley, your assistant.
A couple of things before I can help with these:
1. Text to your sister: I can draft this for you to send, but I don't have the ability to send texts myself, and I don't know what time you expect to arrive or your sister's name/contact info. Could you give me:
- Your sister's name
- What time you expect to be late until (or how late)
- Any reason you want included, if any
Once I have that, I'll draft the message here for you to copy and send yourself.
2. Grocery spending last month: I don't have access to your bank accounts, credit card statements, or any budgeting app - so I have no way to know this number. I'm not able to look it up or estimate it without inventing a figure, which I won't do per your rules.
To get you that number, you'd need to either:
- Check your bank/credit card statement and tell me the total, or
- Give me access to a specific record (like a spreadsheet or exported statement) you want me to read through
Let me know the sister details and I'll get that text drafted right away.
Set it up once
Open the Claude app, phone or computer, free plan is fine
Start a new conversation.
Paste this at the top, then carry on with whatever you needed:
You are my assistant. Your name is [pick a name]. Three rules. One: you draft, I send. Never tell me something was sent or done unless I told you I did it. Two: never invent a number, a date, or a fact about my life. If you do not know, ask me. Three: when you tell me something is fine or nothing is due, tell me what you looked at to decide that.
When you make your first Project (the next guide shows you how), paste the same three rules into its instructions, so they hold for everything that lives there.
Where this goes wrong
You set the rules once and never check them. Mine printed all clear every morning for two weeks while it was quietly skipping the exact emails I needed, and I found out from a missed deadline, not from the AI. A rule you never test is a wish. Once a month, ask it to show you the last five things it decided were fine and what it looked at to decide that.
The real win
This is the foundation. Every guide on this site sits on top of it, and every one of them is safe to try because of rule one: nothing reaches a real person until you've read it.
Numbers on this page come from my own system, computed 2026-09-10: 43 distinct mornings the 6 AM job ran, 2026-07-28 to 2026-09-10 · 1423 ticked checkboxes across the live queue and the yearly archive.
The rules need a job to do. The list is the first one.