The Rulebook
You tell your AI not to do something. Next week it does it again, because what you typed lived in one chat. A standing rule is different: it's written down with the date and the story, a check enforces it, and it binds every agent, every new hire and every new model. Type a mistake below and watch it become one.
What went wrong this week?
Type it the way you'd tell a friend, or tap one. You get the rule, the check, who it binds and who decides, on a card you can save.
Runs in your browser. Nothing you type is uploaded.
Is it a rule, or a wish?
Think of one thing you've told your AI or your team more than once. Check only what's honestly true about it.
We run on ours
These are counts from our own rulebook, read on November 10, 2026. Every Monday it is checked again for any rule that lost its check.
The rules not yet behind a check are on a list to get one. We publish the number because a rulebook that hides its gaps is a wish list.
What a rulebook gives your business
- Your judgment, written onceHow you'd decide, so a new situation gets your answer instead of a guess.
- Rules with a date and a storyEvery rule traces back to a real mistake. No story, no rule.
- What needs no permissionBoundaries you approve one time. Inside them the work just happens, and you get a receipt instead of a question.
- Your decisions, on fileIn your words, with the date. Nobody asks you the same thing twice.
- Proof it happenedFinished work carries its evidence. "Done" is something you can check.
- Every mistake feeds backA failure becomes a new rule and a new check the same day. That is the whole loop.
Who decides: three tiers
| Tier | What it means | Examples |
|---|---|---|
| A · Do it | Inside a boundary you approved once. It happens and you get a receipt, not a question. | Fixing a broken page. Following up on schedule. Proving its own work. |
| B · Check first | It touches a rule or something you already decided. The rulebook is checked, then it acts. | Texting a customer. Quoting from the price list. Pausing a campaign. |
| C · Owner only | It stops and asks you one question. | Spending money. A new price or promise. Anything public under your name. Anything that can't be undone. |
How to turn a mistake into a standing rule
- Write down what went wrong. One sentence, the way you'd tell a friend. "It quoted the wrong price."
- Turn it into one rule. A plain sentence a new hire would understand on day one.
- Give it a check. Something that can be answered yes or no every time. If the check is "be careful", it isn't one.
- Say who it binds and who decides. Everyone who could repeat the mistake. And whether it can be done without asking, after a look at the rules, or only by you.
- File it with the date and the story. Outside the chat, where every session starts. The story keeps the reason alive after everyone forgets the day it happened.
Why instructions don't stick, and rules do
- An instruction lives in a chat. The next session starts without it. A rule lives where every session starts.
- "Be careful" can't be checked. A rule comes with a yes-or-no test that runs every time.
- One agent heard you. A rule binds all of them, and the hire you make next year.
- Nobody remembers why. A rule keeps the story, so it doesn't get quietly deleted as "old".
- You get asked again. A decision on file, in your words, is looked up instead.
FAQ
What is a standing rule?
A standing rule is a correction that outlives the conversation it was given in. It is written down outside the chat, it carries the date and the story of what went wrong, something checks it every time, and it applies to everyone who could repeat the mistake: every AI agent, every new hire and every new AI model.
Why doesn't my AI remember what I told it?
Because an instruction typed into a chat lives in that chat. The next session, the next agent and the next model start without it. A rule only sticks when it is kept somewhere every session reads, and when a check enforces it instead of relying on memory.
What is the difference between a prompt and a standing rule?
A prompt asks for good behavior once. A standing rule makes the behavior permanent: it is filed, dated, tied to the mistake that caused it, and backed by a check. You correct once and the correction is still there next month.
What makes a good check?
A good check can be answered yes or no every single time, without judgment. "Be more careful with prices" is not a check. "Any message with a dollar figure is matched against the price list before it sends" is.
What should an AI agent never decide on its own?
Spending money, new prices and promises, anything that can't be undone, legal and tax calls, and anything the public sees under your name. A rulebook writes down what may be done without asking, what needs a look at the rules first, and what only the owner decides.
Is what I type uploaded anywhere?
No. The rule maker runs entirely in your browser. Nothing you type is sent to us or saved. If you use the Claude or ChatGPT buttons, your text goes to that assistant, and only when you click.
Do you run your own business this way?
Yes. The counts on this page come from our own rulebook, read when the page was built: how many standing rules are on file, how many are enforced by a check that needs no memory, how many boundaries are approved in advance and how many owner decisions are on record.
Can you set this up for my business?
Yes. Text RULEBOOK to Rudy and he will send you the tool and tell you what a rulebook for your business would cover.
Stop repeating yourself to software.
Text RULEBOOK to Rudy and he'll send you the tool and what a rulebook for your business would cover.