Context
Who, what, why
The situation you are building for.
WhoWhatWhy & painCurrent reality
The method
Context. Goals. Constraints.
The framing method I use to turn a vague idea into something real and testable, fast. It is design thinking, tightened for building with AI: get the frame right and you can prototype, test and learn in days.
Where it comes from
It grew out of design thinking
Twenty+ years of UX work taught me tools can change but the foundation stays the same: understand the people and the problem, define what a good outcome looks like, build something, put it in front of real users, learn, repeat. Every project I run still follows it. Frame-First is that same discipline, tightened for a world where the prototype takes an afternoon, not a quarter.
Empathise · Define
Context + Goals
Who it is for, what has to improve, and what a first build would need to prove.
Ideate · Prototype
Constraints, then build
Name the real limits, then let AI explore fast inside them.
Test · Iterate
Validate with real users
Watch what people actually do, change one piece, and run it again.
The frame
Three things to get clear first
Get these right and the build has somewhere to go. The cheatsheet carries the full set of prompts for each.
00 · Start in conversation
Before you touch any tool or code, work through all three in dialogue with AI. Bring what you have: research, insights, competitor examples, and ask AI what is missing. Refine them together, in conversation.
Goals
What should improve
The outcome you want the user to reach, and how it should feel to use.
User goalsSuccessDesign styleInteraction
Constraints
What must stay true
Where real limits become design boundaries.
Security & privacyTechnical limitsFidelityScope
The hard part
Turn blockers into constraints
Most stalls are really "the ask is too big" or "the input is too sensitive." Name the rule underneath and the build keeps moving.
Blockers turned to constraints
Every blocker hides a rule. Drop the "so we can't", and rewrite what is left as "It must…" or "It cannot…".
"AI can't use our internal data."
It must use synthetic or redacted data.
"We don't have backend access yet."
It must run on the front-end with mocked responses.
"It's too risky to build fully."
It cannot go past a clickable prototype this round.
04 · After the frame
Build, validate, iterate
The frame is step one, not the whole job. The real work is the loop that follows: build small, put it in front of real people, and let what they do drive the next move.
- Build the first version fast
- Put it in front of real users
- Act on what they actually do
- Change one piece, or rethink the whole idea
Then run it again Small loops keep you fast and honest: you find out what is not landing while it is still cheap to change.
The stack, honestly
What I build with
People ask, so here it is. The method does the work. The tools just run it, and they move fast.
Research and framing
Perplexity
Working out the brief, stress-testing what I think I know, and spotting what I have missed.
Building
A local IDE, Antigravity today, with Claude and Gemini
Two models within reach at once. I move between them depending on how hard the problem is.
Ask me next month and the IDE might have a different name. That part changes at the speed of AI. What has not moved in two years: Perplexity for the thinking, then a real IDE with the strongest models for the making.
Want this run through on your own project? A vibe coding session walks the whole workflow with you.
Try it yourself
Start sharper than a blank page
One prompt that walks you through Context, Goals and Constraints in Claude, ChatGPT or Perplexity. It gets you a clearer starting point, not a finished brief and no substitute for a session. No sign-up, nothing to install.
Take it with you
The Frame-First AI Cheatsheet
Context, Goals and Constraints on one sheet, with the full prompts and how to turn a blocker into a constraint. Prints on A4 or US Letter, in English or Dutch.
One page, made to share
Download the PDF, or open the full view in your browser.
Download PDF: English · Nederlands
Bring me a vague idea
A Framing Session runs this method against your real problem. You leave with the brief and a clear first build.