The challenge

A founder came to us on a Thursday afternoon with a validated idea, a clear user story, and a hard deadline: a conference demo on Saturday morning. 36 hours.

Here's exactly how we did it.

Hour 0–2: Requirements & architecture

We spent the first two hours in a video call mapping every screen, every database table, every API endpoint. No code yet. Just decisions.

Output: a complete requirements doc, a Supabase schema, and a component list. This is the most important two hours of any project.

Hour 2–8: Foundation

• Next.js project scaffolded with our base template (15 min)

• Supabase project created, schema applied (30 min)

• Auth flow: sign up, login, magic link (45 min)

• Base layout: navigation, sidebar, theme (1 hour)

• All three with Cursor + Claude doing the heavy lifting

Hour 8–16: Core features

This is where the AI tools really shine. The core user flow — create, read, update, delete — for the main entity took about 3 hours. Including:

• Database queries with full type safety

• Optimistic UI updates

• Error handling and loading states

• Empty states and onboarding

We slept at hour 16. Yes, we slept. You make worse decisions when exhausted, and bugs you'd catch rested take an hour tired.

Hour 16–24: Billing, polish, edge cases

• Stripe Checkout integration (45 min)

• Webhook handler (30 min)

• Plan gating with RLS (20 min)

• Mobile responsiveness (1.5 hours)

• Edge cases, error messages, validation (2 hours)

Hour 24–32: Landing page + deploy

• Landing page in Framer (3 hours — yes, Framer is that fast)

• Custom domain, DNS propagation

• Vercel production deployment

• End-to-end testing

Hour 32–36: Buffer and demo prep

We always build in buffer. Things go wrong. DNS takes longer than expected. A Stripe webhook needs rewriting. Buffer time is not wasted time — it's insurance.

The product demoed at the conference. The founder got 12 sign-ups on the spot.

What made it possible

It wasn't heroics or luck. It was: a pre-built template, a standardized stack, AI tools for code generation, and a founder who made decisions fast. Remove any one of those — the timeline doubles.