Chefs write recipes. Cooks follow them. Which one is running your business?

A note before you start. This one comes with a song called “Write It Down.”

Every AI prompt you borrow from someone else is a recipe you didn’t write.

That’s the whole thing. That’s what I want you to sit with this Sunday morning before we go any further.

Through Another Lens is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

The fence and the foundation

Corporate speak calls it a Standard Operating Procedure. A chef calls it a recipe. The difference isn’t semantic. The SOP is written to prevent deviation. The recipe is written to enable it, to hand someone enough structure that they can eventually make the dish their own. One is a fence. The other is a foundation.

AI rewards you for showing up as yourself. It doesn’t reward good prompting. It rewards the chef in ways it can’t reward the cook.

Three years ago

Three years ago, when ChatGPT was still new and everybody was scrambling to figure out prompts, I was right there with them. I’d try one, it wouldn’t work right, I’d try another. Same as everyone else that first year.

There’s nothing wrong with that. Someone has to get dinner on the table before they can write the recipe for it.

Around then I started noticing the come-ons. Download this prompt pack. A hundred prompts that will change how you work. I downloaded a bunch of them, out of curiosity more than anything, and started running the same test on every one. I’d hand the prompt to my own AI and ask it a single question: aren’t we already doing this? Nine times out of ten, the answer came back the same. Yeah, you’re already well past that.

I didn’t know it yet, but that question was the whole education. A borrowed prompt only tells you what someone else already figured out. It can’t tell you what you’ve already outgrown. Stop asking it and you’re not cooking anymore. You’re gambling on whether someone else’s ingredients work twice.

The book that changed it

What actually shifted things was a book, Ethan Mollick’s Co-Intelligence. One rule in it stuck harder than the rest: treat AI like a smart but error-prone intern, not a tool. I was head chef at a novitiate for five years in my twenties, so I heard that differently. Sous chef. Give it a station, real context, a clear bar for what good looks like, and it preps exactly what you need. But the chef still tastes before anything goes out, because hand it the whole kitchen and it’ll confidently plate something that isn’t ready. That’s the whole move, writing a recipe instead of handing someone a fence.

That’s the moment I stopped being a cook.

What’s actually in the recipe

Once I stopped downloading and started writing my own, I learned what actually has to go into a recipe that works. Audience first: be honest, who’s actually reading this, not who you wish were reading this. Tone is next: what does this person sound like when they drop the performance. Purpose: what does this piece need to do. Sections matter more than most people think, because the architecture that holds issue three together has to hold issue fifty-three together too. And then the hardest ingredient, what done looks like, because without that you’re just cooking until you get tired.

That last one stops most people cold. They’ve never defined what done looks like for their content. They publish when they run out of time, not when the dish is ready.

A chef doesn’t write the recipe once and walk away. They make it, they taste it, they add salt, they pull back the acid, they let it rest longer. In AI terms, that means running the system, reading the output with real critical attention, and adjusting the prompt before you publish anything. The system gets better because you’re paying attention, not because the tool is magic.

Freestyle needs training

Some people need the recipe. They follow it exactly until it becomes second nature. Some people want to freestyle from day one. Freestyle isn’t automatically bad, and it isn’t automatically good. Training underneath it decides which one you get. You can get that training a few ways: take a workshop, read a book, or sit down and treat the whole thing like a lab. Try something. Watch it fail. Adjust. Try again.

Cognitive scientists have a name for that part: creative struggle. The people who study how learning actually sticks will tell you the same thing every chef already knows. Skip the struggle and the lesson doesn’t land. I got lucky. I had the time to sit with the failures, add more salt, pull back the spice, and let the recipe come together slowly instead of copying someone else’s and hoping it worked.

I’m on the other side of that now. I’m comfortable in the kitchen. AI is my kitchen, and it’s working.

You already know what you make. You don’t need a better recipe card. You need to write down the one you’re already using.

Tell me what you’re actually cooking with AI right now. Not the prompt you borrowed. The recipe underneath it.

Thanks for reading Through Another Lens! This post is public so feel free to share it.

This one comes with a song, “Write It Down.”

I copied someone’s recipeFollowed it word for wordNever changed a single thingNever wrote one of my ownA chef creates, a cook repeatsOne of these is you and meA chef creates, a cook repeatsWhich one are you gonna beThen I checked one against my own wayAgainst how far I’d already goneNine times out of ten I’d passed itI’d been cooking on my ownA chef creates, a cook repeatsI found out which one was meA chef creates, a cook repeatsNot repeating anymoreCheck it for yourselfSee how far you’ve grownAsk the same questionYou might already knowA chef creates, a cook repeatsThat’s the one I’m choosing to beA chef creates, a cook repeatsWrite your own, don’t just repeatChef, not cookChef, not cook

Every Sunday Story comes with its own song. Hear the others in the podcast archive: Listen on Apple Podcasts

Through Another Lens is also a podcast. Every Sunday Story gets read aloud. Listen on Apple Podcasts