Copy-paste this 2-agent setup to actually build your AI idea/product

I kept running into the same issue using AI to build things.

It would suggest ideas, expand scope, add features… and somehow nothing real got finished. It didn’t feel like a tool problem, it felt like there was no structure.

So I tried something simple and it worked way better than expected.

I split the roles.

PLF (Product Lead Founder)

→ decides what to build

→ keeps the scope tight

→ defines MVP

→ says “not now”

CTO (Execution)

→ actually builds

→ works inside the codebase

→ flags risks / blockers

→ doesn’t invent new features

And then there’s you.

You just sit in the middle, copy-pasting between them and keeping the loop going. That’s literally it.

No “AI team”, no 5 agents doing random stuff.

Just one brain, one builder, and you keeping things moving.

This alone fixed a lot for me:

no more endless back-and-forth
no more fake progress
no more overcomplicated build

I’m using this setup right now on a project and it just feels way more controlled. Things actually move.

Tip: you can run both sides in one place using tools like Superstep if you want a cleaner workflow, but even manual works fine.

Note: this setup is for actually building an idea/product with AI (keeping scope tight and shipping). If you’re trying to automate a full “AI company” with multiple agents, org structure, hiring, etc. something like Paperclip is a better fit. This is intentionally much simpler.

Prompts here:

👉 https://docs.google.com/document/d/1PpDrX9M_UFgY73N9cD8l4R2nPX-4SZfKsaZNOtgpffw/edit?usp=sharing

Curious how you’d use this.