Practical insights from building a production-ready application using AI assistance - from concept to deployment in an afternoon.
We built a todo app using Claude. Here's what actually worked.
Not just another todo app. This is a testbed:
Built with every bell and whistle:
All of it. In an afternoon. With AI doing the heavy lifting.
Claude Desktop + MCP tools. That's it. No IDE, no traditional development environment needed. Just focused conversations with an AI that can read, write, and modify code.
Direct, iterative conversations:
When Claude went off track or got stuck in a loop:
A brutalist todo app. Not because it's trendy, but because it forced clear decisions:
Check out the live app at todoornottodo.org.
Tech stack stayed modern:
Security and performance weren't optional:
Real example of building the drag-and-drop feature:
When it worked: Direct requests, clear constraints, quick iteration. When it didn't: Vague goals, no pushback on complexity, letting patterns repeat.
The key: You're the architect, Claude's the implementer. Stay in your lane.
Things we learned:
You don't need to be technical to build with AI. But you need to be:
Claude can write complex code. Your job is to prevent it.
The todo app is just the beginning:
But the real value? The template it provides:
We're open-sourcing this. Not because the code is special, but because the process is.
This is part one of our AI development series. Coming up:
Want to build something similar? Get in touch. We'll show you how.
The tools used:
The skills needed:
Everything else? Claude handled it. (including this post)
Stay tuned. We're just getting started with AI-driven development.