I've been tinkering with Claude Code for a couple months now. A lot of throwaway stuff, little experiments, nothing I'd show anyone. "Can it make a to-do app?" Yes. But this weekend I had an "oh shit" moment. Most customer-facing software is flexible across multiple use cases and wants to be used by everyone. In the era of Claude Code I can start deleting apps from my phone and purpose-building my own stuff. I can code something that's pre-configured to me, that just works for my use case and nobody else's.
What I Built
I live near 4th & King in SF, right by the Caltrain station and a couple Muni stops. Every morning I'd pull up the Muni app to check when the next train was coming, and the experience was terrible. Too many taps, too much visual clutter, arrival times for lines I never take. So I built Muni Quick Tracker. A single-page web app that shows me exactly the arrivals I care about. Only the stops near me. No configuration screen, no settings. And I made it look like 90s Nickelodeon because I could. The whole thing came together in a few hours. Go backend that pulls from the 511.org API with server-side caching (they rate-limit at 60 requests per hour, so the app caches smartly and staggers its calls). You can toggle between "minutes until arrival" and actual clock times. It detects data quality issues and warns you when the API is returning sparse or stale data. It's running in a Docker container on my home server. I open it on my phone every morning before I leave.
The Part That Surprised Me
This thing would have taken me weeks to figure out on my own. What I brought to the table was knowing exactly what I wanted. The spec was tight because that's what I do. I knew which stops, which lines, which display format, which edge cases mattered. Claude Code turned that spec into a working application. I uploaded it to GitHub so anyone can clone it and point it at their own stops. The config is just a YAML file. Swap in your stop IDs and you're done. GitHub: github.com/bdkoeh/muni-quick-tracker I'm going to build something bigger next.