🎯 The product idea

Locked in before the experiment starts: an AI-powered prompt library SaaS — a simple tool where users can save, organise, tag, and share reusable ChatGPT/Claude prompts. Target user: content creators and marketers who run the same types of prompts repeatedly and want a personal library they can search. Monetisation: free tier (50 prompts) + Pro tier (£5/month, unlimited).

Why this idea: it solves a problem I have personally. It's narrow enough that one developer (or one AI agent) can build it in 30 days. And the Stripe integration test will be real — if it makes £1, the experiment is honest.

"The idea is simple on purpose. The experiment is about whether AI can build it — not whether the idea is clever."

📅 The 4-week sprint plan

Week 1 · Days 1–7

Plan & scaffold

  • Generate full PRD with ChatGPT
  • AI designs database schema
  • Antigravity scaffolds the Next.js app
  • Auth (Supabase) working by Day 7
  • Design system chosen (Tailwind)
Week 2 · Days 8–14

Core features

  • Prompt CRUD (create, read, update, delete)
  • Tag system + search
  • Share link generation
  • Basic dashboard UI
  • Mobile responsive (AI QA pass)
Week 3 · Days 15–21

Monetisation

  • Stripe checkout integration
  • Free vs Pro gate logic
  • Upgrade flow UI
  • Email confirmation (Resend)
  • Billing management page
Week 4 · Days 22–30

Launch & sell

  • Landing page (AI copywritten)
  • Deploy to Vercel + custom domain
  • SEO meta + sitemap
  • Launch on Product Hunt + Reddit
  • Track first revenue

📊 What I'm tracking

Technical: Days to first working build, number of AI-generated errors that needed manual fixing, lines of code written by human vs AI, number of Antigravity agent runs. Business: Sign-ups, free-to-paid conversion, total MRR by Day 30. Time: Total hours spent — if this took 200 hours of prompt engineering it's not a viable side hustle regardless of revenue.

❓ Frequently Asked Questions

Will you publish the source code?

Yes — if the experiment succeeds, the full GitHub repo goes public after the 30-day window. If it fails, I'll still publish what was built so others can learn from the approach. Part of the experiment's value is the code audit: seeing exactly what percentage of lines an AI actually wrote versus what I had to fix manually.

Why £100 as the revenue target?

It's the minimum that makes this feel like a real business rather than a free tool. At £5/month Pro, that means 20 paying users in 30 days — achievable for a launched product in a relevant niche, but not trivial. The target is set before the experiment starts and won't move based on performance.

What if the AI can't build it?

That's a valid and interesting result. If Antigravity or Claude Code hits a wall — a Stripe integration bug it can't resolve, an authentication edge case it loops on — I'll document exactly where and why. A failed experiment with specific failure points is more useful than a vague success story. I won't bail out and switch to manual coding.

Why a prompt library? Doesn't that already exist?

Yes — PromptBase, FlowGPT, and others exist. This isn't about disrupting the market. It's about proving that AI can build a working product in a specific niche in 30 days. The competitive landscape is intentionally familiar so I can benchmark whether the AI-built version is competitive on core features. The experiment is about build speed and AI capability, not market differentiation.

How does this relate to your XR background?

My background is in XR/AR/VR development — not web SaaS. That's intentional. The point is that AI should be able to scaffold a SaaS product even for a developer whose expertise is elsewhere. If a senior XR developer with no SaaS experience can build and launch a web product in 30 days using AI tools, that's a meaningful signal about what AI-assisted development actually enables.