Building Beauty Bar: Week 1 of Vibe Coding a Full App

This week I started building Beauty Bar from scratch...

The Idea

Beauty Bar is an AI-powered booking and recommendation engine for high-end salons. Clients describe what they want in natural language — "something low-maintenance for summer" — and the system recommends services, stylists, and time slots.

It's a real product I'm building for a real salon. And I'm building the entire thing in public.

Week 1: What Happened

Day 1-2: Architecture

I started with the stack: Next.js 16, Supabase for auth and database, Claude for the AI recommendation engine. Standard choices, but the interesting part is HOW I built it.

I used Claude Code for nearly everything. Not copy-paste prompting — actual pair programming where the AI writes code, I review it, and we iterate.

Day 3-4: The Booking Flow

The core booking flow took two days. This was the first major test of "vibe coding" — giving Claude Code the full context of what I wanted and letting it build the components.

What worked:

• Component generation was fast and accurate
• TypeScript types caught integration issues early
• The AI understood Supabase patterns without much guidance

What broke:

• Complex state management needed manual intervention
• Edge cases in the calendar picker required human thinking
• The AI kept over-engineering the error handling

Day 5: The AI Recommendation Engine

This was the highlight. Building an AI system to recommend salon services based on natural language descriptions. I used Claude's API directly with a carefully structured prompt that includes the salon's full service menu.

Lessons from Week 1

1. AI pair programming is real — but it's pair programming, not autopilot
2. Start with the hard part — the recommendation engine de-risked the whole project
3. Building in public creates accountability — I shipped more in a week than I would have in three

What's Next

Week 2: payment integration, stylist profiles, and the admin dashboard. Follow along on YouTube.