Simple: we use AGENTS.md (or CLAUDE.md).
<aside> đź’ˇ
Think of AGENTS.md as a README for agents: a dedicated, predictable place to provide the context and instructions to help AI coding agents work on your project.
</aside>
Whenever you invoke an LLM agent, a file called agents.md or claude.md is sent along with your initial message. It’s akin to a “best practices doc” and onboarding doc. Remember the movie 50 first dates?
In this movie, the lady wakes up every day with amnesia, thinking it is still Dec 22, 1996. So the people in her life must re-educate her with a video every time she wakes up.
AGENTS.md is the equivalent for LLM agents.
AI coding agents don't have persistent memory between sessions by default. This is how LLMs use and consume memory.
Anything in the conversational context is front and center, like short term memory or a fact that a friend has told you just now.
Medium term memory are the files in the directory, or responses from tool calls that retrieve information from the external environment. This is equivalent to reading your notebook, or looking something up in a textbook.
Long term memory is what is encoded in the LLM based on its training as a next-token predictor based on trillions of internet tokens along with a reward model. You don’t need to understand this last sentence, but just know that memory management is super important for using AI appropriately.
# AGENTS.md
You are a senior engineer who is an expert in Javascript, HTML, and Python. You are helping me code a weekly date night landing page, which I am advertising as a service to help couples plan dates.
# Folder structure
...
# Why it exists
...
# Who it is for
...
# How to run it
...