Prototyping using natural language to create click tracks with Subdivide; both typed and with voice. Feels so much smoother than all the clunky individual text inputs that would otherwise be needed for each component.
Not sure who else remembers Compiz on Linux, but I started recreating it in the browser with 5.6 Sol. Wild how far a single prompt got me, and how much nostalgia hit when I ran it.
3D desktop cube. Wobbly windows. 🤘
I've been building a virtual office for my startup. I wrote some longer thoughts on what it does and why, along with some things I learned on the way. https://t.co/7siAa0kCpA
Work should also be fun sometimes, so why not have a virtual office? Persistent AI teammates with shared knowledge, memory, specific roles, and scheduled tasks. All of this works with a simple bridge that connects to Claude Code and Codex CLIs so no new tools are needed. WIP.
@AndersDrage@valiocon I really make the coolest friends at @valiocon every year. Thanks for sharing Anders.
Looking forward to you launching your metronome app too! It’s gonna have that Manly Steel Dragon touch.
Having some fun experimenting with view transitions + motion gestures while working on a portfolio update. I’m using a shared element for the videos so playback is continuous across views.
Big launch for https://t.co/AMuyjkUqGk: You can now scan sheet music to create click tracks. We’ve wanted to do this since day one, and can finally do it accurately and reliably.
Having some fun experimenting with view transitions + motion gestures while working on a portfolio update. I’m using a shared element for the videos so playback is continuous across views.
My friend @alecrichter_ wrote a great piece on one of the harder lessons in building with agents: they can't always be trusted to follow written directions. He's a smart dude building cool things.
https://t.co/IOTfdlN2gE
Snap a picture of your sheet music, get a fully built click track to practice with.
I tried to add this feature to Subdivide about a year ago, but it didn’t really work; AI models got confused about bar lines vs. note stems and music recognition software was too slow and made too many mistakes.
Now here we are a year later. GPT-5.6 can read and interpret music structure accurately and in much less time than it would take to manually create the track.
Coming soon. 🎶