Français
Deutsch
The experiment that surprised me
In March 2025, a client asked me to help prototype an internal tool for tracking their team's certifications. Scope: a simple web application with database, dashboard, expiry alerts. Stated deadline: "ideally within the week, to validate the concept before investing."
I delivered a functional prototype in 31 hours. Here's exactly what happened, hour by hour.
Hours 0–3: Specification with Claude
I used Claude to transform the client's verbal requirements into structured specifications: user stories, data model, business rules, use cases. Claude identified 3 edge cases the client hadn't mentioned. Final specifications: 8 pages, complete.
Hours 3–12: Development with Cursor
I submitted the specifications to Cursor with the prompt: "You are a senior full-stack developer. Build this application according to the attached specifications. Stack: Next.js, PostgreSQL, Tailwind CSS. Priority: code readability and maintainability."
Cursor generated the complete architecture, database migrations, UI components, and APIs. I reviewed each component, fixed 4 functional errors, adjusted business rules on 2 points.
Hours 12–20: Testing and fixes
Functional testing across all use cases. 7 bugs found. 6 fixed by Cursor in 2 hours from natural language descriptions. 1 required an architecture decision I arbitrated with the client.
Hours 20–31: Deployment and documentation
Deployed on Vercel + Supabase. User documentation generated by Claude from the code and specifications.
Real limits
"This prototype isn't production-ready. It's a sophisticated proof of concept — not a system maintainable by a team over 5 years."
What this approach doesn't handle well: in-depth security, performance under load, WCAG accessibility, long-term maintainability, automated regression testing. For a true production system, a senior developer remains essential for architecture review. But for validating a concept, convincing a sponsor, or accelerating time-to-feedback — it's transformational.