Twenty One Media
aiAugust 10, 2026

One File, Three Homes: Teaching Claude Who You Are

We just shipped module 4 of the Freedom Ag training portal, and the core concept is simple enough to explain in one sentence: write a business file once, put it somewhere Claude reads automatically, and stop re-explaining yourself.

Most people don't do this. They start every chat with a paragraph of context that says essentially the same thing: here's who we are, here's how we talk, please don't do the thing I hated last time. That context evaporates when the chat closes. They write it again tomorrow.

The business file ends that loop.

What the File Contains

The worked example we built for Freedom Ag runs about ten sections. We'll use it to show what matters.

The first three are factual: who you are, who is likely talking to Claude at any given moment, and what you actually sell. That last one sounds obvious until you've watched Claude confidently recommend a product line the client doesn't carry.

The role section is worth spending time on. Freedom Ag has at least four distinct users of their AI setup: management, operations, sales reps in the field, and office staff. Management wants the decision and the number behind it, short. A sales rep is in a truck or a field, probably on a phone, and needs something they can say out loud to a grower at a glance. Office staff want a finished document, formatted and ready to send. Those four people need different outputs from the same AI. Telling Claude who's asking, and what that means for how to respond, is not optional. It's the difference between useful and generic.

The voice section is where most businesses leave money on the table. Freedom Ag's file says: "plain-spoken, short sentences, say the thing." It lists words they use ("stand in the gap," "return per acre," "get you taken care of") and words they never use ("synergy," "solutions provider," "leverage as a verb"). It says to read any draft back in the voice of a farmer standing in a shop at 6am, and if it would sound like a brochure out loud, rewrite it.

That's not vibes. That's a spec.

The Rules Section Is the Most Important Part

Every business file needs a rules section. Hard lines Claude cannot cross, regardless of what the prompt says.

For an agricultural retailer, two rules are non-negotiable:

First: never state a price, availability, product rate, or delivery date that isn't in the source material I gave you. Put a [CONFIRM: ...] placeholder instead. A wrong rate in a customer's hands is a real problem in a way that a clumsy sentence never is.

Second: never invent an agronomic recommendation. Claude can organize, summarize, and draft around what you tell it. It cannot decide what goes on an acre.

Those rules belong in the file, not in a prompt. Putting them in a prompt means someone has to remember to include them every time. Putting them in the file means they're there on every chat, every project, every scheduled task that references the file.

Three Places It Can Live

The file needs to end up somewhere Claude reads automatically. There are three options:

A Project's Instructions panel is the right choice for most work. The file loads every time a chat opens inside that project. If most of your Claude use is organized around a project, put it there.

Cowork's Global Instructions (under Settings) means every Cowork session, across all projects, has access to the file. Use this if your work isn't cleanly project-organized or if multiple people on the team are doing different things and all of them need the same baseline context.

A CLAUDE.md file in a project folder is the developer path. Claude Code reads it automatically on every session in that directory. If you're building with code, that's the home.

Same file, three homes. Which one you use depends on where your work happens.

The Rule for Keeping It Alive

The file is only useful if it's current. We teach one rule for maintaining it: when Claude gets something wrong twice, that's a missing rule, not a bad prompt.

If you're correcting the same behavior in every chat, the correction belongs in the file. You paste the file in, say "add a rule that says never do X," and you get back the edited section. Once you update the file, you stop having that conversation.

Isaac has been running a Claude.md file for about a year and a half. He describes it as massive, thousands of individual instructions accumulated from corrections. Video scripts go in a Google Doc. Pricing never appears in customer-facing copy. If I ask for this, then do that. A year of that habit turns your AI from a capable but foreign tool into something that actually knows how you work.

That's the goal of module 4. Not a clever prompt. A file that gets smarter every time you use it.