AI Product Management at Spotify: A View From Three Angles
Henrik (Henok) Lindström, Divita Vohra and Humberto Corona · 9 November 2026
Read the originalThree Spotify product managers show that managing AI products is three different jobs: owning research, owning a shared AI model, or owning a feature listeners use.
Henrik Lindström, Divita Vohra and Humberto Corona are principal product managers in personalisation at Spotify, the music streaming service. They work on the same technology and serve the same listeners, yet their daily work and their measures of success differ. Each describes what product management looks like at one layer: research, a shared recommendation AI model, and a feature listeners use.
- Lindström, in research, turns a promising model design into decisions the company can take, starting with a concrete problem and agreeing what evidence would justify the next step.
- Vohra owns a recommendation AI model used by other teams. Its behaviour, Vohra writes, cannot be specified, only shaped through training data and checked with tests of output quality.
- Vohra's best result this year came with a team that agreed how success would be measured before it agreed the architecture.
- Corona, who owns a feature listeners use, builds tests from real usage and AI-based judges tuned to the feature's own interactions, and runs them continuously after launch.
A product organisation hiring for AI products may need to say which layer a product manager will own, because each layer is judged by different evidence. This is an inference and is not the authors' claim.
Test an AI feature with examples taken from real use, have an AI model tuned to that feature grade its answers, and keep running these tests after launch.
Agree with a partner team how success will be measured before agreeing on the architecture of a shared AI model.
Derived by Working Surface from the article; more in the Playbook. Source line: Evaluation is not a phase that happens before launch; it is a continuous product in itself.
10 November 2026: as AI agents do more of the building, teams are naming the one point where a person's judgement must sit: approving a merge or release, vouching for a summary or defining good output.