Back to all of the ideas

Same prompt, different context

Watch one request turn from a hollow email into a useful one, then start a context file of your own.

I asked an AI to write a follow-up email to my friend Blair after we met about using Zapier more at Patreon. This is the whole prompt, and it never changes.

Write a follow-up email to Blair after our meeting about using Zapier more at Patreon.

Without context

Hi Blair, I hope this email finds you well. Thank you for taking the time to meet with me today. I really enjoyed our discussion about Zapier and Patreon. Please let me know if you have any questions or if there is anything else I can help with. Looking forward to connecting again soon. Best regards, Eric

With a context layer

Blair! Great catching up today. Loved hearing where ‘Treon is headed. I owe you the governance one-pager I mentioned, so it’s attached. I’m still getting replies to the newsletter issue that featured your build, and I’m forwarding my favorite. Eric

Drag to compare the two emails.

The first email is technically correct and completely hollow, because the AI knew nothing about me, nothing about Blair, and nothing about what we talked about.

The prompt stayed the same and so did the model. The second time, the AI had simply read a syllabus about me first.

The four questions a syllabus answers

Blair and I both come from education, and we kept noticing that people treat AI like a magic box to be commanded when it responds far better to being taught like a student. A good syllabus answers four questions, and so does a good context layer.

  1. Who is this course about?

    Who you are, how you work, and how you like to be talked to. Five sentences about yourself is a working version.

  2. What is the reading list?

    Where your real knowledge lives. Point at the source wherever you can, because a pasted copy goes stale the day you paste it.

  3. What are the assignments?

    The tasks you repeat, written down once so that you can ask for them by name. Your first one can be a single task you do every week.

  4. What does good work look like?

    An example you loved, one you did not, and a single tiebreaker source that wins when your documents disagree.

Start your own syllabus

Answer in plain sentences and skip anything you are not ready for. Nothing you type leaves this page, and you can copy or download the result as a markdown file that any AI tool can read.

Your context file

my-context.md

Your answers will appear here as a markdown file.

If you would rather talk than type

Paste this into the AI tool you already use and let it interview you instead.

Help me build a personal context layer: a small set of files that teaches AI who I am and how I work. Interview me one question at a time to fill in four sections: (1) who this is about and how I like to work, (2) where my real knowledge lives, (3) the tasks I repeat and how I want them done, (4) what good work looks like, with an example and a tiebreaker source. Then write it all as one markdown file I can save and reuse in any AI tool.

A syllabus like this is never finished, and I think that’s a feature. I have rewritten mine eight times, and each rewrite took minutes because editing is so much easier than authoring.

Want to run this with your team?

I teach all of these as live sessions, where people work on their own real examples.

Bring this to your team