Spotify's Top Developers Stop Writing Code
Spotify's top developers have not written code since December, as the company has shifted to an AI-driven workflow where generative agents handle code generation, QA, and deployment. Co-CEO Gustav Söderström acknowledged that while AI writes the code, it doesn't yet "write the guarantees," highlighting new challenges in automated validation and monitoring for AI-orchestrated engineering pipelines.
- The new workflow is powered by an internal system called "Honk," which integrates with Anthropic's Claude Code. - Engineers prompt "Honk" through Slack, even from mobile devices, to generate code for bug fixes or new features, which can then be reviewed and deployed. - This shift has reportedly led to a significant increase in development velocity, with Spotify shipping over 50 new features in 2025. - The role of senior engineers has evolved from writing code to reviewing and supervising the AI-generated output, focusing more on system architecture and product decisions. - To handle large-scale code changes and maintenance, Spotify had previously developed a "Fleet Management" system for automating code transformations across repositories, which laid the groundwork for this AI-driven approach. - While the AI generates code, the human engineer is still responsible for curating the intent, monitoring the impact, and approving the final release, keeping accountability in human hands. - This AI-driven approach extends to Spotify's data science, where they are building unique datasets based on nuanced user listening habits and preferences to train their recommendation and personalization models. - Concerns within the broader engineering community suggest that over-reliance on AI for code generation could lead to a "whack-a-mole" bug cycle, where fixing one issue introduces others due to the AI's lack of holistic codebase context.
