AI-Builds
A shared home for prototypes built with AI
- Role
- Product Designer — hub architecture & workflow
- Company
- Campminder
- Team
- 4 product designers, the whole design team
- Timeline
- Ongoing since March 2026
My role focused on
At a glance
The problem
Each designer prototyped alone in their own repo, on their own deploy pipeline, so nothing was shared and reviewers needed a live walkthrough to see anything.
The insight
Built one shared hub where any designer can describe a flow to Claude Code and get a real, interactive React app, deployed automatically.
The impact
The whole design team adopted it, not just one champion. 30+ live prototypes exist today, each one a shareable link instead of a scheduled walkthrough.
Designers building with AI
Every product designer, not just one champion
Live interactive prototypes
Real React apps, not static Figma flows
Time to a shareable link
Auto-deployed, no setup for the reviewer
01, Context
Four designers, four separate pipelines
Figma prototypes hit a consistent wall: complex flows, edge cases, and conditional logic are hard to express without custom code. Each designer who wanted more had built their own one-off repo and deploy setup, which meant redundant work, dead links, and nothing anyone else on the team could find.
Handoff to engineering was lossy too, static screens rarely captured error states, loading behavior, or branching logic. And getting a PM a shareable link often meant a designer had to be online to walk them through it.
Why it matters
4
designers, 4 separate pipelines
Each one managing their own deploy setup before this existed
0
shared place to see anyone else's work
Prototypes lived in personal repos, dead links piled up
Lossy
handoff to engineering
Static screens couldn't capture error states or branching logic
Async
PM review, but painfully so
A shareable link often meant waiting for a live walkthrough
02, The approach
Designers direct, Claude builds
We built a shared, AI-assisted prototype hub where designers author interactive, code-based prototypes using Claude Code. Each prototype is a real React app, fully interactive, shareable via URL, deployed automatically on every push.
The key insight: designers don't need to know how to code. Claude Code handles the implementation. Designers focus on the flows, the copy, the logic, and the experience, Claude writes the code.
CAMPMINDER
AI-Builds
Select a campsite
RANGER’S CAMPSITE
9 PROTOTYPES
SCOUT’S CAMPSITE
12 PROTOTYPES
EMBER’S CAMPSITE
4 PROTOTYPES
TRAILBLAZER’S CAMPSITE
6 PROTOTYPES
Where ideas come to camp
One shared hub
A single repo where every designer has their own folder. A landing page surfaces everyone's prototypes with a unique URL, no setup required to view.
Describe it, Claude builds it
Open Claude Code, describe the flow, reference Figma screens or the design system, Claude writes the interactive prototype.
Iterate conversationally
“Add an error state when email isn't found,” “make the brand toggle show the camp logo,” changes ship as fast as you can describe them.
Isolated by design
Each designer's folder is isolated. You can't accidentally push someone else's changes.
Figma-aware
With the Figma MCP integration, Claude reads component specs, tokens, and layout directly from the file.
Auto-deployed
Every push to main deploys automatically. A shareable link is ready in about 30 seconds.
03, Key decisions
What made it stick with the whole team
A tool only one person uses isn't a system. Every decision here optimized for the other three designers adopting it without a training session.
One shared repo, not one per designer
Separate repos didn't scale, each designer managing their own pipeline created redundancy, inconsistency, and dead links. One hub, one deploy story.
Designers don't need to know how to code
Claude Code handles the implementation. Designers focus on the flows, the copy, the logic, the AI writes the code.
Code over Figma for complex logic
Figma prototypes break on conditional logic, edge cases, and branching flows. A real React app expresses any behavior a static prototype can't.
Self-serve review over live walkthroughs
A shareable URL that works with zero setup turns PM review from a scheduled walkthrough into something async, on their own time.
Three views for three audiences
The same prototype serves the design team's full exploration, a clean flow for user testing, and a scoped current-vs-new comparison for engineering, one build, three lenses.
04, Proof at scale
17+ flows, one migration
The Auth0 readiness prototype is what this approach enables at scale: 17+ interactive flows covering every caregiver scenario, from new-parent onboarding to expired-link recovery to multi-camp account linking, plus a one-click brand toggle between camp-branded and Campminder-branded views.
It replaced a mix of Figma prototypes, static screens in Notion, and verbal walkthroughs, and it covered edge cases Figma couldn't express, with zero setup required for a PM to open and review it themselves.
17+
interactive flows built
Covering every caregiver scenario: onboarding, recovery, guest accounts, role upgrades
3
views from one prototype
Exploration for design, a clean flow for testing, a scoped kickoff view for engineering
1 click
to toggle brand
Every flow switchable between camp-branded and Campminder-branded, no duplicate files
05, Results
From presentation artifacts to living tools
This turned prototypes from something you present into something PMs explore on their own, engineers reference directly, and designers keep current without a separate production pipeline.
Designers who adopted it
Not just one AI-curious champion
Live prototypes today
Across onboarding, staffing, hiring, and camp-branding work
Setup to view a prototype
Open the URL, no local install or walkthrough needed
Deploy time per push
From `/push-live` to a live, shareable link
Reflection
What I took from it
The hardest part wasn't the tooling, it was designing a workflow the rest of the team would actually adopt without a training session. A tool only one person uses isn't a system, it's a habit.
Watching Figma-only workflows give way to real interactive prototypes changed what I expect from my own process now, I reach for code earlier, not as a developer, but as a designer who wants the edge cases to be honest from day one.
Next project
Design System Agent