Spotify told investors that its most senior engineers “have not written a single line of code since December.” The comment came during the company fourth quarter earnings call and it highlights a new internal workflow that relies on generative AI to create and deploy production code.
Gustav Söderström described an internal system named Honk that uses Claude Code to generate code on command. He gave a concrete example of an engineer asking an AI agent on Slack to fix a bug or add a feature during a commute. The engineer then receives a packaged build and can merge the change before reaching the office. The company said this flow has sped up development and deployment.
Honk Internal System
Honk is the name Spotify uses for the automation and orchestration layer that sits on top of generative coding models. The system integrates with developer tools and with messaging. It can produce code edits, run tests and push builds back into a review flow. Spotify credits Honk with helping it ship more than fifty features and app changes over the past year.
Spotify said recent launches include Prompted Playlists Page Match for audiobooks and a new About This Song feature. The firm used these examples to show how rapid iteration produced real product outcomes for listeners and creators. The company also said the approach is evolving rather than finished.
How it Works
Developers now act more like supervisors than typists. They give natural language instructions to the AI. The model returns code changes and test results. Engineers review the output and then approve merges. Spotify said many senior engineers now focus on higher level design and validation. The company highlighted the role of Claude Code as the underlying generative model in their pipeline.
Spotify also said this change came after a key transition in December when a version of Claude Code became capable of reliably producing and testing code at scale. That milestone pushed the firm to invest in a production system that could harness the model safely and repeatedly. The company described the December moment as the day the workflow crossed into practical use.
Code Generation Meets Caution
The move changes how product teams scale. More rapid iteration may let Spotify test ideas faster. The company said it expects better product velocity and more frequent releases. Spotify also said it will guard unique data that it deems valuable for model training. The firm believes its music and audio data cannot be easily copied by general models and that this data gives it an advantage as it builds internal models.
There are practical and ethical questions. Code generation raises concerns about correctness security and maintainability. Quality gates and human review remain essential. Spotify indicated that engineers still supervise the output and that production approvals are part of the flow. The company also noted work to prevent spam and to label synthetic content for artists and listeners.
