The brief format that gets the design you wanted
Five fields that turn fuzzy asks into concrete renders.
Why briefs fail
Most “the AI didn’t do what I wanted” complaints aren’t agent failures. They’re brief failures. The operator says “build me a landing page” and the agent picks reasonable defaults. The defaults aren’t what was in the operator’s head. Twenty rounds of revisions later, nobody’s happy.
The fix isn’t a smarter agent. The fix is a tighter brief. Five fields, completed before any render starts, prevents most of the back-and-forth.
This is the cheapest quality gate in the stack. It costs ninety seconds. It saves an hour per artifact. And once you internalize the format, you start writing better briefs to humans too — the discipline transfers.
The five fields
Every brief, every time. If a field is genuinely unknown, mark it FUZZY and have the agent propose options before rendering.
| Field | Question it answers | Common failure |
|---|---|---|
| Who | Who is the audience for this artifact? | “Everyone” — produces a generic result |
| What | What is the single takeaway they leave with? | Listing 5 takeaways — produces visually flat work |
| Why | Why does that takeaway matter to the business? | “Because we need it” — no real direction |
| Where | Where does this artifact sit in the larger surface? | Skipping placement — agent over- or under-scopes |
| How to test | How will we know it worked when we look at the result? | Skipping it entirely — the most common miss |
The most-skipped field is “how to test.” Operators write the first four, then ship. The agent renders, the operator looks, says “hmm,” and the loop has no anchor. The test sentence is the anchor.
Three ways briefs fail
The patterns that produce the “23 versions of paragraph 2” loop. Hover any card to see the diagnosis.
The vibes brief
The kitchen-sink brief
The no-test brief
The fix in all three: concrete reference, single priority, observable test. Vibes, kitchen-sinks, and untestable goals are the three failure modes. The five-field brief makes all three visible before render.
The pre-render check
The diagnostic is one prompt. Paste it before any design ask. The agent restates your brief in five fields and flags any that are fuzzy.
Before you render anything, restate the brief back to me
in this exact shape:
Who: <audience, in one phrase>
What: <single takeaway, in one sentence>
Why: <what that unlocks, in one phrase>
Where: <placement, in one phrase>
How to test: <one observable sentence>
If any field is fuzzy, write "FUZZY:" and propose 2-3
specific options I can pick from. Do not start drafting
until every field is concrete or I've explicitly chosen
one of your proposals.Forces the brief to be complete and concrete before any render begins.
When you start using this, expect 2-3 fields to come back FUZZY on every brief. That’s normal. Pick from the agent’s proposed options and the brief tightens. Three rounds of FUZZY-resolution is still cheaper than three rounds of render-revision.
Annotated transcript
A real conversation: operator asks for an About section, agent forces the brief, agent flags an internal contradiction before drafting. Click any pinned annotation to expand the “why this turn matters” commentary.
Three operator turns, three agent turns. Result: a clean draft on first pass + the brief saved as a reusable template.
- operator
I want a section on the about page that explains who I am without sounding like a LinkedIn bio.
- agent
Before I draft, give me the brief in five fields:
1. Who — who reads this? (founder peer? potential customer? recruit?)
2. What — the single takeaway you want them leaving with
3. Why — what does that takeaway unlock for the business
4. Where — where does this section sit (top of page? deep in About?)
5. How to test — one sentence that says how we'll know it landedIf any field is fuzzy, say "don't know yet" and I'll propose options. Cleaner to surface the gaps now than to render and re-render.
- operator
Who: operators running their own ventures. What: that I build with discipline, not vibes. Why: filters out tire-kickers, attracts the right collaborators. Where: top of the About page, after the lede. How to test: a reader can summarize my approach in one sentence after reading.
- agent
Tight brief. Two callouts before I draft:
- "Discipline, not vibes" is the takeaway, but it's also a cliché. I'll only land it if the section demonstrates discipline (specific numbers, concrete artifacts) rather than claiming it. Agreed?
- The test is good. "Summarize in one sentence" is a real gate. I'll write toward that — every paragraph earns its place against that summary.Drafting now. I'll show you one version. If it doesn't pass your test, we revise the brief, not the prose.
- operator
Agreed on both. Draft.
- agent
[draft delivered — three short paragraphs, two with concrete numbers, one with a named artifact link]
My own pre-test read: a stranger could plausibly summarize this as "runs three ventures with explicit playbooks; shows the playbooks; not a coach." That's close to your test. If you read it cold and the summary is different, the brief was wrong — tell me which field to revise.
Brief template — copy and paste
Save this in a snippets file. Paste before any design ask. Fill all five fields. Mark FUZZY for any that are genuinely unknown.
Brief: <one-line title of the artifact>
Who: <the audience, in one phrase — be specific>
What: <the single takeaway they leave with, in one sentence>
Why: <what that takeaway unlocks for the business, one phrase>
Where: <where this sits in the larger surface, one phrase>
How to test: <one observable sentence describing how we'll know
it worked when we look at the result>
References (optional):
- <link, screenshot, or "looks like X but not Y">
Anti-references (optional but powerful):
- <one thing you don't want it to look like>
- <another>
Anti-references do disproportionate work. Saying “not corporate-SaaS, not landing-page-template-#7” narrows the agent’s search faster than positive references alone. Vibes are exclusionary.
Prompt kit
Three prompts for briefing, refining, and grading design work. Save in your CLAUDE.md or a personal snippets file.
Before you render anything, restate the brief back to me
in this exact shape:
Who: <audience, in one phrase>
What: <single takeaway, in one sentence>
Why: <what that unlocks, in one phrase>
Where: <placement, in one phrase>
How to test: <one observable sentence>
If any field is fuzzy, write "FUZZY:" and propose 2-3
specific options I can pick from. Do not start drafting
until every field is concrete or I've explicitly chosen
one of your proposals.I'm about to ask you to design something. First, help me
write the brief.
My fuzzy ask: <paste your one-liner>
Walk me through the five fields:
1. Who reads / sees this
2. What single takeaway they leave with
3. Why that takeaway matters to the business
4. Where this lives in the larger surface
5. How we'll test that it worked
For each field, ask me ONE clarifying question if my answer
is too vague. Stop after five questions max — fuzzy briefs
are better than over-specified ones.Here's the brief: <paste 5-field brief>
Here's the result: <paste output / link / screenshot>
Grade the result on each field:
- Who: does the result speak to that audience? (1-5)
- What: does the takeaway land in the first read? (1-5)
- Why: does the result advance the business reason? (1-5)
- Where: does it fit its placement? (1-5)
- How to test: does it pass the test sentence? (yes/no)
For any field below 4, propose the smallest revision that
would lift it. Don't redesign — surgical revisions only.Apply this — your next design ask
15-minute exercise. Take one fuzzy ask you’ve been sitting on. Run it through the brief. Render once. Grade against the brief, not your gut.
Your first five-field brief
Each step takes 2-5 minutes. Progress saves automatically.
- 01Pick a fuzzy ask you've been sitting on. One sentence, no detail."Redesign the home page." "Write the about section." "Make the dashboard cleaner." Any of these.
- 02Run the pre-render check. Fill in all five fields before any render.If a field is genuinely unknown, mark it FUZZY and have the agent propose options. Don't paper over the gap.
- 03Have the agent draft once. Then grade the result against the brief — not against vibes.Use the third prompt in the kit. Numerical scores keep the conversation honest.
- 04If the result fails, revise the BRIEF, not the prose. Re-render from the new brief.This is the move that breaks the "23 versions" loop. The brief is the source; the render is downstream.
- 05Save your three best briefs as a snippets file. They become your brief vocabulary.After ~10 briefs you start recognizing the field that's usually fuzzy for you (most operators undercook "how to test"). Knowing your weak field shortens future briefs.