Why I keep a prompt journal

The cheapest AI tool I own is a plain text file. Every time a prompt works unusually well — or fails in an instructive way — it goes in, with two lines of context: what I was trying to do, and what happened.

Six months in, the file has changed how I work more than any new tool has.

What the journal taught me

My failures are repetitive. I fail the same way: asking for too much in one shot, not specifying output shape, forgetting to state the failure behaviour. Seeing your own patterns in writing fixes them faster than any course.

Good prompts have structure, not magic. Almost every winning prompt in the journal has the same anatomy: role and goal, input, output shape, constraints, worked example, failure instruction. Once you see the skeleton, prompting stops feeling like sorcery.

Prompts are reusable code. The journal became a library. New task? Search for the closest old one, adapt, go. What used to be twenty minutes of fiddling is two minutes of lookup.

The format, if you want to steal it

Date. Task. The prompt itself. What happened. What I'd change. That's the whole schema. A file beats an app for this — it's greppable, portable, and it never gets sunset.