prompt-engineering
Where Documentation Goes to Die
The information was never the bottleneck. Making it usable meant building a small app, and that afternoon is now a paragraph of instruction and a paste.
A reference document is a filing cabinet: everything is in there, and nobody opens it. The gap between having the information and being able to use it is where most documentation quietly dies.
I ran a small experiment against that gap. I took one page, Anthropic's guide to prompting well, and instead of reading it a fourth time, handed it to a single prompt (one I keep in PUCO for exactly this) and asked for something I could actually use. What came back was a working little app: a searchable reference, a guided builder that assembles a prompt one part at a time, a path for diagnosing a result that went sideways, favourites, quick help on every screen. It installs on a phone. It works offline after the first load. I did not write a line of its code.
The prompt did. Not a clever one-liner, but a long, deliberate instruction that knows how to research a topic, group it into sensible categories, scaffold a proper progressive web app, add search and offline caching, mark anything it could not verify as "please check", and package the result. The work moved up a level. I did not build the tool; I built, or rather reused, the one prompt careful enough to build the tool for me, out of whatever I feed it next.
That is the part worth keeping. The source happened to be a prompting guide, so the artefact is now a prompt that teaches prompting, which is the kind of recursion I find funny at eleven at night and slightly less funny at nine the next morning. But swap the source and nothing else changes. Feed it a command-line reference and you get a searchable CLI companion. Feed it the phrasebook your reception staff keep losing and you get a guided assistant for the awkward phone call. Feed it your own onboarding notes and you get the thing you always meant to write and never did.
The material you already own, the docs nobody reads, the notes rotting in a folder, the knowledge that lives only in your head until you leave, was never the bottleneck. The bottleneck was that making it usable meant building a small app, and building a small app is exactly the kind of afternoon you keep postponing. That afternoon is now a paragraph of instruction and a paste.
Which leaves a more uncomfortable question than the technical one. If the tool is this cheap to make, the excuse for leaving good information unusable just got a good deal thinner.