30 Days of Vibe Coding - The Wrapup
30 projects in 30 days using AI-assisted coding. Here's what was easy, what was hard, what was unexpected, and where this is all going.
30 projects in 30 days using AI-assisted coding. Here's what was easy, what was hard, what was unexpected, and where this is all going.
It's done. Thirty projects. Thirty days.
A platformer. A Snake clone on a Nokia 3310. An RPG with isometric graphics. Tetris with procedural music. Breakout. A Pomodoro timer in Go. A git dashboard. A notes app with an MCP server. A kanban board an AI can manage. A Miro clone. A Trello clone. Wordle. A portfolio generator. A weather dashboard with ASCII art. An auto-battler. A tic-tac-toe game where the AI learns from losing. A hacking game about an AI escaping containment. A real-time polling app. A reaction wall for events. A collaborative mood board. Anonymous chat rooms. A live Q&A tool. Windows 95 in a browser. A blog redesign. An ambient sound mixer with zero audio files. Collaborative pixel art. A terminal emulator in Rust. A code editor in Go. A Notion clone. And an operating system that contains all of the above.
This was possible because of coding agents. Mainly Claude, sometimes Codex, but also Watchfire to orchestrate complex work by turning rough ideas into detailed specs for my agents to execute against.
The Numbers
- 30 projects shipped and deployed
- ~326,000 lines of code across all projects
- ~1,200 commits total
- ~450 Watchfire tasks (from 4 tasks for tic-tac-toe to 43 for the code editor)
- 8 tech stacks: TypeScript/React, Go/Bubble Tea, Python/Textual, Rust/Tauri, Go/Wails, vanilla JS, Astro, Next.js
- 6 Firebase projects with real-time sync
- 5 games, 5 TUI apps, 3 native desktop apps, 17 web apps
- 3 platforms (macOS, Linux, Windows) for the native apps
- 7 languages per blog post (English, Portuguese, Spanish, German, French, Italian, Japanese, Chinese)
What's Easy
The things that worked almost every time, with minimal iteration.
Games and self-contained apps. Anything with clear rules and a defined scope. Tetris, Wordle, Breakout, Snake. The AI knows what these games should feel like because they're well-documented patterns. You describe the game, the AI builds the game. The results were playable on the first build almost every time.
Frontend UI. React components, Tailwind styling, layout, animations. The AI generates clean UI code fast. Card layouts, modals, sidebars, dashboards, form flows. If you can describe what it should look like, you get something close on the first pass.
...
---
Read the full article →

Add a comment: