Spotify's engineers haven't written code since December. AI is doing it for them.
The example Söderström gave on the Q4 earnings call this week: an engineer opens Slack on their commute, tells Claude to fix a bug in the iOS app, and walks into the office to a waiting build. Not a demo. That's the actual workflow. The internal system is called "Honk," it runs through Claude Code, and senior developers, he said, haven't personally written a line since December.
Whether that's exciting or unsettling probably depends on where you sit. The output is real, Spotify shipped over 50 features in the past year, including AI Prompted Playlists, audiobook Page Match and About This Song. Söderström credited the pace to this workflow directly.
Another interesting piece of the earnings call was what he said about data. Spotify's listening history captures something public datasets don't: where you're from shapes what you want to hear, and that specificity is baked into billions of sessions in ways that don't exist anywhere else. What counts as workout music in the U.S. looks different than it does in Scandinavia or Western Europe.That regional signal is what Spotify is training on.
"This is a dataset that we are building right now that no one else is really building," Söderström said. "It does not exist at this scale."
On AI music: labels can now tag how much AI was involved in making a track, and Spotify is still actively removing what Söderström called "slop tracks."
The company is using AI heavily on the engineering side while trying to manage what it means for musicians on the supply side. Those two things will keep pulling against each other.
The engineering shift has already happened. The actual question now is what the job is. If you're approving code you didn't write, engineer might not be the right word anymore. Editor, maybe. That's where I land.
