Coding Lesson of the Day: Give Your AI a Project State File

Series: Coding Lesson of the Day

Learning outcome: the reader can create a tiny project-memory file that helps an AI agent restart without re-explaining everything.

Time: 5 minutes

TLDR

If you are new to coding with AI, do not start by asking the agent to build the whole app. Start by giving it one clear memory file.

Create a file named:

PROJECT_STATE.md

Then write three lines:

# Project State

**What I am building:** A simple personal website.
**Who it is for:** Friends, clients, and people who want to understand my work.
**Next useful result:** A homepage with my name, one paragraph, and a contact link.

That is enough to make the agent less forgetful.

Why This Matters

AI agents are fast, but they are not automatically organized. A chat can grow long, restart, or lose the thread. A project-state file gives the agent a small source of truth it can inspect before acting.

This is the beginner version of local-first project memory: your project knowledge lives in a file you own, not only in a chat window.

Tiny Example

Bad starting prompt:

Build me a website.

Better starting prompt:

Inspect PROJECT_STATE.md first. Then tell me the smallest useful next step. Do not build yet.

The better prompt gives the agent:

• context
• a boundary
• a smaller task
• a chance to explain before editing

Try This Now

1. Make a new folder for one small idea.
2. Create PROJECT_STATE.md.
3. Add:

- what you are building - who it is for - the next useful result

4. Ask your AI agent to inspect the file.
5. Ask for one next task, not the whole project.

Agent Prompt

Read PROJECT_STATE.md and summarize what you understand about this project.

Then answer:
1. What is the smallest useful next task?
2. What should wait?
3. What would you need me to decide before building?

Do not edit files yet.

Human Check

After the agent answers, ask yourself:

• Did it understand the project without me re-explaining everything?
• Did it suggest one task instead of a giant plan?
• Did it tell me what I still need to decide?

If yes, you just created your first project memory loop.

Trinity Check

• Outcome: you have a working project-memory file.
• Agency: you still own the goal and decisions.
• AI Hygiene: the agent has bounded context and is not editing before inspection.

Learn Next

• Read the free manual: ../Agentic Coding Starter Manual.html
• Try the next file: TODO.md
• Then learn basic Git so you can checkpoint your project before risky edits.