Workshop

Designing with AI

We'll design an authorization dashboard from scratch — together, live, using AI to explore more ideas than any one person could alone.

Define→ Generate→ Rate→ Evolve

Mindset

The Bicycle Shop Advantage

Langley spent 5 years building one design and tested it twice.
The Wrights tested hundreds of wing shapes in weeks.

Wright Brothers

Bicycle mechanics
Homemade wind tunnel
Hundreds of wing shapes tested
Cheap materials, fast cycles
Flew first

Samuel Langley

$50K government grant
Smithsonian scientist
One design, over-engineered
Years of planning
Crashed into the Potomac. Twice.

AI gives you a wind tunnel. Test dozens of ideas before lunch.

Context

Levels of Iteration

Not all AI workflows are equal. We're aiming for Level 3.

Level 1

Chat Window

Copy-paste with ChatGPT. "Make it darker." "Add a sidebar." One change at a time, no structure, no memory.

Slow · unstructured · manual

Level 2

AI Products

Google Stitch, Figma Make, v0. Beautiful outputs, but their flow, not yours. Limited export, no custom feedback loop.

Fast · polished · opinionated

Level 3 ← us

Custom Tooling

Your templates, your rating system, structured feedback export. Design the flow itself — then AI fills it.

Your flow · your tools · full control

Level 2 tools are great for exploring. Level 3 is how you ship what's in your mind.

Today

Authorization Dashboard

A unified management UI that brings 3 AuthZ models together — so Ops can configure user access to features in one place.

authz-admin.app

3

AuthZ Models

248

Active Roles

1.2K

Policies

86

Features

Plan

50 Minutes, Four Phases

5 min

Define

Brain dump into the shared doc. Who uses this? What data matters?

15 min

Generate

AI reads your brief and creates a diverse set of rough designs.

15 min

Rate

Browse, compare, rate each design. Leave notes on what works.

15 min

Evolve

Export feedback to AI. New round of designs. Repeat until we nail it.

Phase 1

Define

Before AI touches anything, answer four questions.

Product

What does the app do?
Who's the user?
One key metric?
What can they act on?

Aesthetic

Dark, light, or both?
Data-dense or spacious?
Playful or serious?
Brand colors?

        // example brief

        Name: "AuthZ Admin"

        Users: "Ops team configuring user access"

        Data: "Roles, policies, feature flags, audit logs"

        Vibe: "Clear, structured, trustworthy"

Phase 2

Generate

AI reads your requirements and creates a rough set of designs — then you go broad and deep.

Macro — Design Gallery

Explore the variety

Browse all designs full-screen. Rate each one, leave notes, export structured feedback for AI.

Micro — Design Lab

Tune the details

Pick a design you like and fine-tune — themes, spacing, fonts, colors, layout — all adjusted live.

Phase 3

Rate & Annotate

Every design gets a score and your notes on what works and what doesn't.

Rate & leave notes

⭐ Star rating — 1 to 5 per design
🏷️ Quick tags — Favorite, Ship It, Needs Work, Skip
📝 Notes — "Love the spacing, hate the header"
📋 Export — structured feedback, ready for AI

Five axes

Utility — does it work?
Delight — does it feel good?
Speed — does it feel fast?
Discovery — can users explore?
Innovation — is it fresh?

Phase 4

Evolve

Your ratings become the next prompt.

        // auto-generated from your ratings

        LOVED (more like these):

          A1 (5★) "Keep the warm serif type + cream bg"

          C2 (4★) "Clean and airy. Needs denser data tables."

        FIX (iterate):

          B2 (3★) "Too dark. Lighten the greens."

        KILL (avoid):

          B3 (1★) "Too busy. Buried the key metrics."

Gen 1 → Feedback → Gen 2 → Ship

Let's go

Let's give it a whirl

Brain dump into the shared doc.
We'll generate the first round together.

Make the wind tunnel. See what works. Iterate fast.