Stop prompting. Start building a system.

Every few weeks someone shows me their prompt library. Hundreds of prompts, neatly tagged, lovingly organised. And almost none of them ever get used. Here is why, and what I do instead.

A prompt is a wish you type into a box. It works once, you get a result, and then it evaporates. Tomorrow you are starting from scratch, retyping a slightly worse version of the same wish. For two years I lived like that, mistaking a folder of clever prompts for actual leverage. It was not leverage. It was a museum.

The problem with prompts

Prompts fail quietly, and they fail for boring reasons. Once you see the pattern you cannot unsee it:

• They have no memory. Each one starts cold, with none of yesterday’s context.
• They do not compound. A hundred prompts is a hundred islands, not a system.
• They depend on you remembering they exist, in the right moment, which you never do.

The result is a strange kind of busywork: you feel productive collecting and tagging, while the actual work stays exactly where it was.

You do not need more discipline. You need a workspace that remembers for you.

What a system actually is

A system is the opposite of a prompt. It is the place the good prompts go to live, where yesterday becomes input for today. The difference is not the cleverness of any single instruction. It is whether the thing accumulates.

In practice, a system has three properties a prompt never will: it keeps state, it runs without you, and it gets a little better every time you use it. Miss any one of those and you are back to wishes.

The three properties, concretely

1. State. It writes down what it learned, so the next run is not the first run.
2. Autonomy. It can take a step on its own, on a schedule, without a human babysitting it.
3. Compounding. Every pass leaves the setup slightly sharper than before.

How I built mine

It started embarrassingly small. One folder, one file of notes, and a single command I could run from the terminal. The point was never to be impressive. It was to make tomorrow cheaper than today.

# one command, run every morning
$ rafa brief --since yesterday --write notes/
  reading 4 logs ... done
  drafting 3 ideas ... done
  saved to notes/2026-06-20.md

From there it grew the way systems do: one small, boring improvement at a time.

Key takeaway. Do not collect prompts. Build the smallest possible system that keeps state, runs on its own, and gets better each time. Then let it compound.

Where this leaves you

If you are sitting on a museum of prompts, you do not need more of them. You need to pick one repeated task and give it the three properties above. Start with a single file and a single command. Make tomorrow cheaper than today, and do it again. That is the whole trick, and it is the only one I trust.

I write these as I build them, mistakes left in. If that is your kind of thing, the next one lands soon.

Written in the open. If something here was useful, the best thank you is to subscribe and build along.