I Think I Found Something

A non-dev founder's accidental discovery about .md files, AI, and why we've been delegating work wrong this whole time.

---

I'm not a developer. I've never written a line of code before this year. I run multiple businesses — a digital agency, an influencer marketing, startup and even a restaurant.

I use AI every day. Not to write emails — to think, build, and make decisions.

And a few weeks ago, I accidentally stumbled onto something I haven't been able to stop thinking about.

---

It started with a workshop.

I was running a live demo for my team — showing them how to use AI in their actual work. Not theory. Hands-on, real tasks, real output.

During the session, I noticed I was pre-prompting things. Setting up context chains so the output would be more predictable. I kept optimizing as we went — making the prompts tighter, the structure cleaner.

Then I had a thought: if AI already knows the full context of this workshop, why can't I just pack all of this into a single .md file and let it act as a teaching assistant?

So I did. And it worked.

---

Then it kept working.

I tested it with friends. Different roles, different industries. Same result every time.

But the real moment came when someone on my team asked: "can we send work to each other like this?"

That's when it clicked.

A .md file isn't just a prompt. It's a handoff. One file that carries the full context — the thinking, the scope, what "done" looks like — so whoever receives it doesn't have to guess.

No thread to scroll through. No meeting to re-explain. No assumptions.

---

Turns out, this isn't a new idea. GitHub just released Spec Kit — 81K stars in weeks. A community project called BMAD has 42K. Both saying the same thing: stop winging it, start writing structured specs.

But they're both aimed at developers.

This works for everyone.

---

You're already doing this. Just in the wrong format.

Every email you write to delegate work is a prompt. Every Slack message explaining a task is a prompt. Every meeting brief is a prompt.

The problem isn't that people are bad at communicating. It's that the format keeps losing information along the way.

A .md file fixes the container. Both humans and AI can read it, execute on it, and pack the results back in the same file.

I tested it across Claude, ChatGPT, and Gemini. Same file. Same result.

---

A friend who's a photographer asked how to use AI for pre-production. I didn't know anything about photography workflows.

I gave him the file anyway.

The AI guided him through how he works now, what could change, how to structure it. By the end, he had a working system.

I was basically a conductor who doesn't know the notes. I just set up the structure — and let it run.

---

So here's the deal.

I put the whole thing on GitHub. Files for Claude, ChatGPT, and Gemini. No gate, no email, no catch.

.md is just a text file. There's nothing to build. Nothing to gate. Just a better way to use what everyone already has.

If you've read this far — just try it.

👉 ai-minion-wizard on GitHub

https://github.com/peuarchukiati-rgb/ai-minion-wizard

👉 or grab a coffee chat → cal.com/peak-euarchukiati-3sjqpb/15min

---

This is the first post on KLAI. No schedule, no content calendar — just discoveries as they happen.

— Peak*