Why four files and not one?
In this post you discover why splitting project documentation into four simple markdown files gives AI the exact context it needs, avoids repeated explanations, and makes every session start where the last one left off.
AI has no memory. Without structure, every session starts from zero. Here is how to fix that.
Tags: Vibecoding · Project structure · AI workflow · Productivity
Most vibe-coded projects do not fail because the idea is bad. They fail because there is no structure. You open a new AI session, explain the project from scratch, get halfway through a feature, close the tab, and next time you open it, the AI has no idea what you were building or why.
This is not a flaw in the AI. It just has no memory between sessions. So the fix is simple: you give it the memory yourself.
I use four markdown files for every project. They live in the root of the repo, they get committed like any other code, and they are the first thing I paste into any new AI session.
You do not fill them manually. You use a prompt. The AI interviews you, asks the right questions one at a time, and builds the file from your answers.
Why four files and not one?
You could dump everything into one big file. But keeping them separate means you only paste what is relevant. Starting a new session to work on a feature? Paste `plan.md` and `progress.md`. Revisiting a technical decision? `research.md` has it. Each file has one job. That makes them easy to update and easy to use.
