← back to blog
May 2026

A songwriting skill — from a hidden duck to a few hundred songs

I want to be upfront about something: the power of AI song creation has me a little terrified. What started for me as a throwaway proof of concept has turned into a real passion, and I haven't fully reconciled those two facts. I've made a few hundred songs at this point. I've gifted plenty of them, made people cry, and choked up myself more times than I'd admit — because the act of writing a song, even with AI in the loop, drops you into the poetry of it. You hunt for the line that lands, and when it does, it hits hard.

It started with a goodbye. I was leaving a company I felt deeply connected to, and I wanted to leave a mark. I'd become the guy who hid little resin miniatures around the office as a joke — people would find a random resin duck tucked somewhere and it became a whole game between me and my coworkers. So on my way out, I bought about 200 of them and hid them everywhere as I cleared my desk. To announce the ducks, I used Suno to write a song featuring my coworkers by name, and posted it in the chat on my last day. You can hear it below. That was the moment the whole thing clicked for me.

Tool-wise, I split the work. I've used both ChatGPT and Claude heavily, and they're good at different things. ChatGPT tends to win in the purely creative lane — it just feels more powerful when I'm chasing a vibe or an image. Claude is where I go to lock down structure and get the more rigorous, shaped version of what I want. So I built a songwriting skill to formalize that, and built it to work the way real writing works: in passes, not in one shot.

Most AI lyric tools hand you a finished block of words. If it's almost right, there's nowhere to go — no collaboration, no revision, take it or leave it. My skill starts as a facilitator instead of a vending machine. It helps me find the concept first, then drafts structure, verses, chorus, bridge. The piece I obsessed over is a dedicated refinement phase that treats the song as a draft to improve. And at the end it can output a production prompt — the style description you feed an AI music tool to actually hear it out loud.

The design lesson is one I keep relearning: the value isn't in generating text, it's in the workflow wrapped around it. A good chorus is a few revisions away from a mediocre one. Check out some of the songs I spun up over on my Suno profile.

A few hundred songs deep, and counting.

I'm still uneasy about where all this goes for music and the people who make it. But I also can't pretend it hasn't given me something real. And honestly, this one's allowed to just be fun — not everything I build has to be for work.