PRD
One page that says what you are building, written before anything gets built. An AI builds exactly what you wrote, not what you meant.
“An hour on the brief saves a day of building the wrong thing.” The PRD is where you make the cheap mistakes.
This is step four of the seven step recipe, on its own page. The other six are How to create anything →
Problem
Say what hurts before you say what to build
One paragraph: who has the problem, what the pain is, and why it matters.
No features, no screens, no solution yet.
- Who is this for, and what do they do today?
- What does it cost them, in time or in money?
- Can you say it without describing the fix?
No evidence yet? Write that down. An assumption stays labelled an assumption. Invented proof reads exactly like the real thing, and then gets built on.
Done
Write down how you will check it
Swap fast, better and intuitive for something you can actually check.
This is the line the work has to cross before anyone says finished.
- What can you point at and call it working?
- Could someone else test that without asking you?
- What has to keep working while this is added?
The bar, in one question: could an AI build this without asking you a single thing? If the answer is no, the brief is not finished.
Scope
Fence it, or it grows while you watch
Must-have is the smallest set that solves the problem.
Won’t-do is the tempting list you are leaving out on purpose.
- What is the smallest version that still helps?
- What are you not doing in this one?
- What should it never touch?
An AI will do anything you do not fence off. Split won’t-do into three: always do, ask first, never.
Talk it out, let Claude write it up
Nobody writes a good brief from a blank page. You ramble, it asks, you read it back.
- 1
Say it out loud, messily. The goal, who it is for, what good looks like, what you already tried.
- 2
Ask it to interview you first. The questions a builder would otherwise have to guess at.
- 3
Ask for the brief, not the build. Problem, done, scope, then the detail.
- 4
Read it back as the builder. Paste it into a fresh chat, ask that one to build the thing, and every question it fires back is a gap.
The free prd skill → runs all four for you. In Claude Code a skill is a SKILL.md file in its own folder, and Claude uses it when it is relevant.
Three briefs that are not ready
A feature list
no problem named- Opens with what it should let you do.
- Never says what it is for.
No way to check
done is undefined- Better, faster, more intuitive.
- Nothing to test at the end.
No won’t-do
the fence is missing- Scope grows every session.
- The AI keeps helpfully adding.
Too big for one brief? Phase it. Spec the smallest slice that ships, build that, then write the next one.
- Problem first: one paragraph, with no solution in it.
- Done: something you can check, not a feeling.
- Scope: must-have, won’t-do, and what it may never touch.
- The test: could a builder who has never met you build it?
This week: write the brief for the thing you keep meaning to build. Do not build it yet.
Start the project properly: How to create anything → Getting the words out is its own skill: How to Prompt →
Do it with someone. Everything here is free and you can run it alone. If you would rather work through it with me, one to one, book a free 30-minute chat. No pitch.