Agentic AI Experiment: Building dodolanan.com with Claude Code

A while back, in my spare time, I did some casual research into agentic AI — the goal was one thing: find the most efficient workflow for integrating an AI agent into my everyday development process. Instead of just reading theory, I decided to practice directly: build a real project from scratch using Claude Code.

The result: dodolanan.com, a collection of mini games I managed to build and ship.

What was most interesting about this experiment wasn’t “AI can write code” — that’s old news by now. What was new to me was seeing an AI agent go beyond being a smart autocomplete and actually automate tasks end to end: from project scaffolding, writing game logic, to helping with the deployment process.

A few insights I picked up:

• Agents work best when instructions are clear and scoped — not “build me a game” but specific, feature by feature
• Fast iteration is key: give it small tasks, review the results, move to the next task, rather than asking for everything at once
• AI agents are great at speeding up boilerplate and repetitive work, but the big architectural decisions still need to stay in my hands

For a solo developer like me, this workflow was a game-changer — projects that would normally take weeks got significantly faster without sacrificing code quality.