PREP vs AGENTS.md, when to use each

If you work with AI coding tools, you have probably seen AGENTS.md: a file you drop in a code repository to tell an agent how to behave there: which commands to run, how to test, what conventions to follow. It is genuinely useful, and it is spreading fast.

People sometimes ask whether PREP competes with it. It doesn't. They solve different problems, and they work well side by side. Here is the clean line between them.

AGENTS.md: how an agent should work inside your code

AGENTS.md lives in a repository and speaks to a coding agent. It answers operational questions about that codebase:

Run the tests with npm test.
Use tabs, not spaces.
Don't touch the generated files in dist/.
The deploy command is vercel --prod.

It is instructions for an agent doing work in a repo. Its scope is that repo, and its audience is a tool writing or running code.

PREP: everything about your project, for any AI

PREP is broader in one direction and narrower in another. It is not about code specifically, and it is not a set of behavior rules. It is a small folder (PREP.md, LOG.md, and a memory/ folder) that holds the context of a project so that any AI can pick it up and continue:

What this project is and what we're trying to do.
Where everything lives (the MAP).
What's been decided, and what's happening now (the STATUS and the LOG).
The rules for this project specifically.

The project can be anything: a menu redesign, a family budget, a novel, a research thread, or, yes, a piece of software. And the point is portability: you can start with ChatGPT, continue with Claude, and hand it to Gemini tomorrow, because the memory lives in the folder, not in any one AI.

The one-line difference

AGENTS.md tells a coding agent how to work inside your repo. PREP tells any AI everything about your project.

One is behavior in a codebase. The other is portable context for a project of any kind.

They compose

The best part is that you don't choose. A software project can have both: an AGENTS.md at the repo root telling the coding agent how to build and test, and a PREP folder holding the project's story, decisions, and history that any AI can read, not just the coding agent.

In fact, AGENTS.md is exactly the kind of file a PREP MAP would point to:

## MAP
- code/ — the repository. See its AGENTS.md for how to build and test.
- decisions/ — why we chose this architecture.
- LOG.md — what happened, in order.

When to reach for which

Reach for AGENTS.md when you want a coding agent to behave correctly inside a specific repository.
Reach for PREP when you want the full context of a project to survive across sessions and across different AIs, code or not.

Use both when your project is code and you also want its story to be portable. They were built for different jobs, and neither replaces the other.

Want the full picture of how PREP works? Read the specification, or save your first project with PREP Save and try it in a minute.