Let it be known that I made my first foray into agentic AI coding, two days ago on July 6th, 2025. I worked with Claude Code to start prototyping a little tool to build m3u playlists. This will come in handy for making my local collection of music files exposable as part of my OwnTone project.
It wasn’t quite vibe coding, as I didn’t let Claude off the leash to make changes on its own. All of its requests were reviewed personally. At the same time, I haven’t written a line of code in the repo. In fact, I’ve barely looked at any of the source code. Meantime, my initial runs for building playlists look pretty good. To top it off, now a bunch of ideas for extending the tool are cooking in my head.
The experience hasn’t been mindblowingly life altering, but vaguely satisfying. This is a project I would have likely procrastinated on interminably. With an hour of effort, give or take, at least I’m started. Lots of boilerplate and mind numbing testing avoided. So there’s a path forward. And thoughts of many other deferred projects that could now be within reach. Credit to Andrew Ng, writing in The Batch, for the nudge to finally push out from the dock.
I’m completely sympathetic to all the folks who are apprehensive to pessimistic on where this is going and what it will ultimately cost. For me, guarded experimentation is the right path forward. It feels like there is some there there. Maybe (probably) not enough to match the huckster BS and snake oil, but possibly a useful normal technology. YMMV.