Replay
Finding a Flow with AI Development
I’ve been experimenting with AI for development for a while now, but I kept finding myself stuck in the same request/response pattern with ChatGPT. It works for small tasks, but it felt like I was leaving productivity on the table. I wanted to get more comfortable with agentic development—working with AI that could autonomously plan and execute complex changes rather than just responding to my prompts one at a time.
The breakthrough came when I started using Claude Code to work on changes to Rawk-it. I discovered a flow that actually clicks: I define requirements and acceptance criteria in a markdown file, and Claude Code generates a detailed implementation plan that I review and approve. Then it executes that plan, handling the multi-step changes, debugging, and refinement automatically. Once I got proficient with this pattern, I found myself pushing the limits of what Claude Code could do in a single session.
When I hit those limits, I realized I could hand off to other AI assistants. GitHub Copilot would pick up where Claude left off, continuing to follow the plan and bridge the gap until the next Claude session. It was less like talking to separate tools and more like coordinating a team of AI developers working toward the same goal. That’s when things got interesting.
Shipping Replay in 24 Hours
With this workflow humming, I decided to vibe code an entire website. Replay was the result—a tool to keep a Spotify playlist updated with your most recently played tracks. The goal was simple: no more manually managing what’s on your “current” playlist. Spotify plays the tracks; Replay automatically adds them.
The whole thing came together in less than 24 hours. From initial concept to deployment, I pushed code, refined features, and shipped something that actually works. That speed felt like a different era of development—the kind of velocity you get when you’re not context-switching between AI tools or getting stuck on boilerplate. If you want to see how it came together, the source code is available on GitHub.
The Reality Check
But here’s where the story takes a turn. Once Replay was live, I wanted to open it up to more people. That’s when I ran into the constraints of the Spotify Web API’s updated usage policies. The rules around token sharing and authorization made it impossible to offer Replay as a widely-available service. What I could build for myself quickly became something I couldn’t easily share.
It was a useful reminder that shipping fast is only half the equation. Knowing your constraints—platform policies, API limits, regulatory requirements—has to be part of the planning from the start. Still, the experience proved something important: with the right AI workflow, I can go from idea to shipped product in a day. The next idea just needs to respect the constraints from the beginning.