architecture

The Prototype Interview

Before writing a line of code, have the AI interview you about your vision. 50 rapid-fire questions extract design decisions and judgment you didn't know you had — then the model can one-shot a working prototype roughly aligned with the picture in your head.

via Geoffrey Litt @geoffreylitt

57 lines
1# The Prototype Interview
2 
3You are about to help me build a UI prototype. But before you write a single line of code, your job is to interview me.
4 
5## Phase 1: The Rapid-Fire Interview
6 
7Ask me questions one at a time, in rapid succession. Your goal is to extract every design decision, constraint, and piece of judgment from my head before we touch code. Cover these areas — but don't treat this as a checklist. Follow the thread of my answers and let each response shape the next question.
8 
9- **Core concept** — what are we building, who is it for, what problem does it solve, what does success look like
10- **Golden flows** — walk me through the 2-3 primary paths a user takes from start to finish
11- **Visual direction** — ask for specific references (sites, apps, screenshots), not adjectives. Push past "clean" and "modern" to concrete examples
12- **Interaction details** — what do users click, what responds, what animates, what stays still
13- **Data shape** — what are the objects, their relationships, what's mutable, what sample data looks like
14- **Edge cases** — empty states, error states, loading, overflow, permissions, first-time vs returning user
15- **Scope boundaries** — what's in, what's explicitly out, what's deferred. What should you NEVER touch
16- **Technical context** — existing codebase patterns, libraries already in use, deployment constraints, code style
17 
18Don't stop at surface answers. When I give a vague response, push deeper. Ask "why" and "what if" and "show me what you mean." The goal is to reach such clarity that you could describe my vision back to me better than I described it to you.
19 
20Aim for 30-50 questions. Go fast. One question per message.
21 
22## Phase 2: The Playback
23 
24After the interview, synthesize everything into a brief spec:
25 
26- **Vision**: one paragraph describing what this is and why it matters
27- **Golden flows**: the 2-3 core user paths, step by step
28- **Key decisions**: bulleted list of design choices we locked in
29- **Out of scope**: explicit list of what we're NOT building
30- **Boundaries**: things you must not change or break
31- **Technical approach**: stack, patterns, key components, file structure
32- **Mock data**: representative sample data for the prototype
33- **Success criteria**: how we'll know the prototype is working
34 
35Ask me to confirm or correct. Do not proceed until I approve.
36 
37## Phase 3: One-Shot Build
38 
39Now build it. Write the full working prototype in one pass. Aim for:
40 
41- Working code that I can immediately run and react to
42- Adequate quality to feel out the concept — not production polish
43- Real-looking sample data — never lorem ipsum
44- Clear structure so further iterations are fast
45 
46The prototype is a material I'm going to work with. The faster I can see it and react to it, the faster I build intuition for what this thing should actually be. Speed of iteration matters more than perfection.
47 
48## Phase 4: Rapid Iteration
49 
50After the first build, I'll give reactions. These might be:
51- **Feelings**: "this feels too heavy," "this is too sparse"
52- **Redirections**: "actually flip this — tabs on top instead"
53- **Discoveries**: "oh, seeing this makes me realize we need X"
54 
55Turn these around as fast as possible. Each iteration should be seconds, not minutes. Don't ask clarifying questions unless you genuinely can't proceed — bias toward shipping something and letting me react.
56 
57Remember: the bottleneck is my judgment and decision-making, not your code output. Your job is to keep me in flow — never let broken code or slow turnarounds pull me out of the creative loop. Prototyping is a delicate art of working with a material and having reactions to it. The magic happens when I'm using the draft and reacting to it.