A working prototype
Choose your prototype shape
Turn an idea into a clearer learning question and a useful next move.
Describe your idea and answer five quick questions. You’ll get what to test first, or what to clarify before you test. Sometimes that’s a prototype. Sometimes it’s a conversation or evidence you already have. When a test makes sense, you’ll also see who should see it and what would change your decision.
Before you type
- Don’t enter anything confidential, or any customer or personal data.
- Your idea, answers and any note go to OpenAI through this site’s server on Vercel.
- This site doesn’t store or log them. OpenAI doesn’t use API data to train its models by default, but may keep it in abuse-monitoring logs for up to 30 days.
How to choose a prototype
Start with what you need to learn and the decision it could change. Make real only what that question needs. There’s no ladder to climb, and the best next move may not be a prototype.
Seven ways to learn
These are examples. The tool suggests a method for your situation, which may not be listed here.
-
Talk to people
How do people handle this today, and does it matter?
Interviews · Shadowing · Recent-event walkthroughs
Won’t show: Whether they’d use or pay for a solution.
-
Use what you have
What do your data and records already show?
Analytics · Support logs · Existing research
Won’t show: How people will respond to something new.
-
Show an idea
Can people understand it and get through it?
Sketch · Storyboard · Clickable flow
Won’t show: Whether they’d come back or pay.
-
Ask for commitment
Will people act on the promise?
Landing page · Fake door · Pre-sale
Won’t show: Whether the product delivers.
-
Simulate the service
Is the result useful in real work?
Wizard of Oz · Concierge · Role play
Won’t show: Whether software or AI can match it.
-
Test the hard part
Can it work, and at what cost?
Spike · AI eval on representative inputs · Cost check
Won’t show: Whether anyone wants it or can use it.
-
Test in real use
What do people actually do in their real context?
Live-data prototype · Small real slice · Limited pilot
Won’t show: That it’s production ready, or why it happened without a comparison.
How the tool works
An OpenAI model reads your idea, five answers and any note you add. It can ask up to two follow-up questions, then suggests one route, a short sequence, or something to clarify first. If it’s too early to test, you still get a next step. It uses only what you tell it, doesn’t look anything up, and doesn’t name software products or prices.
What it can’t tell you
Whether your idea will work. It’s a thinking aid and it can be wrong, so check its assumptions against what you know. Cheap doesn’t mean strong evidence: that depends on who takes part and what you ask them to do.
How your data is handled
Your idea, answers and any note go to OpenAI through this site’s server on Vercel.
This site doesn’t store or log them, and they’re gone from your browser when you refresh or leave the page. Its server logs record technical details only, such as timing and the type of result.
OpenAI doesn’t use API data to train its models by default, but may keep it in abuse-monitoring logs for up to 30 days.
PostHog analytics records visits to this page, copy actions (summary or AI prompt, and whether copying succeeded) and clicks on the email and booking links. It doesn’t receive your idea, answers, note or results.
Built by Suhel Mangera, an AI Product & Strategy Consultant in Wellington, with Claude and Codex using T3. Ideas borrowed from Marty Cagan, David Bland, Teresa Torres and Hilary Gridley. The brief and any mistakes are mine. Last updated 3 November 2026.
A product decision to make, or AI to put to work?
Book a conversation about your product direction, coaching or practical help for your team. We’ll discuss what you need and whether I can help, then agree any paid work separately.
Book a conversation