I’ve been listening to guided meditations a lot recently, mostly on the Waking Up app. The daily meditations there have a nifty feature where you can extend them from the default ten minutes to any length, so if you want to sit for 30 minutes or an hour, you can. I’m not sure exactly how it works, but it seems to proportionally increase the pauses between instruction. So, instead of just a bunch of dead space at the end, you get a natural spread of the talking.

This only works with Sam Harris’s “daily” meditations, though, and not the wealth of the other meditation content on the app. I find myself also wanting to make those longer, and recently I started attending a local Waking Up group where we start with a 30-minute meditation, and we want to use non-Sam 10- or 15-minute sessions.

So I vibe coded “spacer”, a web-thingy that takes an audio link or file and proportionally expands the gaps between speech to meet a target duration. Feel free to try it out!

Spacer is the first useful thing I have entirely vibe-coded, that is, I didn’t write a bit of the code or even really look at it, I just prompted Claude Fable 5 (via Claude Code) to do it. I have been using Claude Code extensively over the last year to develop Ripls, but always with me looking at the code and taking a strong hand in the design.

I decided to post the source code of the audio spacer with a permissive (MIT) license. This felt a bit odd. One way of looking at LLM-generated code is as one big pillaging of the intellectual property and effort of those that created the training data, which has been indiscriminately scraped into the maw of these frontier models. What does it even mean, then, to copyright (or not) the output of an LLM? Another interesting question is, what is the value of this source code that was produced in less than an hour by GenAI, and which presumably will be even easier and quicker to create in the future? I decided to post it anyway, but I also included the full transcript of me prompting Claude–this is where all my effort was put, so it feels most like the “source code” that I would set free if open-sourcing a traditional app.