Spec-Driven AI... Actually Works?

I built Terminal Whisper—a voice-to-text companion that turns your speech into clean, polished transcripts you can use anywhere. It started as a fun Vibe Coding experiment… until it became an unmaintainable monster.

In this video, I walk through how I rebuilt it from scratch using spec-driven development, powered by AI tools like Kiro, Claude Code, and GitHub Spec-Kit. Along the way, I share what went wrong with my first build, what I learned about context engineering, and how I shifted from “just shipping features” to building something that actually lasts.

You’ll see:
A quick demo of the new Vissper app (macOS native, built with Swift)
How spec files, contracts, and to-do plans keep AI coding clean
Why “garbage in, garbage out” still rules—especially in AI-assisted development
Lessons from taking a messy NodeJS prototype and turning it into something elegant

If you’re into AI-assisted coding, tool-building, or just curious about what it’s like to code with AI instead of for it—this one’s for you.

👉 Have a look at https://vissper.com

💬 Leave a comment if you’d like to contribute or test Vissper.