Give AI instructions you don’t have to repeat

I keep written instructions for how AI assistants should work with me: use plain language, preserve unrelated work, and distinguish what was checked from what was guessed. This is a small version you can adapt.

Get the template

Download the personal-instructions template. No signup required.

Try it on one task

1. Open the template and change the preferences to match your needs. Remove anything you do not want.
2. Paste the instructions at the start of a new AI chat, followed by one task you can check yourself.
3. Try: “Turn these notes into a short email: I can attend Saturday. Ask what time to arrive. I can bring folding chairs, but I haven’t decided how many.”
4. Read the draft. It should ask about arrival time without inventing a time or promising a number of chairs.

These are instructions to try, not a guarantee that the assistant will follow every rule. Keep a saved copy and paste it into another chat when needed; this exercise does not depend on automatic memory.

For coding projects: AGENTS.md

Download the coding-project template. In a Codex project, save it as AGENTS.md in the repository root. If that file already exists, merge the relevant lines instead of replacing it. Start a new session in that project and ask the assistant to summarize the instructions before giving it a small task.

Codex reads project instructions from AGENTS.md; other coding tools may use different conventions. See OpenAI’s instructions documentation for discovery and override behavior.

Check what changed

Did the draft follow your preferences? Did it add a fact you never supplied? Change one instruction and repeat the task. A short rule about a recurring problem is more useful than a long list you never inspect.

Want help adapting this?

If the result is confusing or you want to use this with your own workflow, contact me or book an introductory AI session. You can use the free template without booking anything.

See all things you can try.