# AI Product Management Roles at Spotify Three distinct product management perspectives on building AI-driven features at Spotify. - **Source:** [Spotify Engineering](https://engineering.atspotify.com) - **Published:** 2026-11-09 - **Topic:** [technology](https://brifing.si/c/technology) - **Original article:** - **On Brifing:** https://brifing.si/s/c0ae6965-f456-4218-8a80-540afe42a29f ## Defining AI Product Roles Product management in AI is not a singular job function but varies based on the specific layer of the technology stack. At Spotify, three distinct roles exist within research, foundation models, and user experience. Each role requires different metrics, daily workflows, and definitions of success to deliver value effectively. ## Tech Research Product Leadership The tech research lead focuses on technical leverage and scientific expertise to solve complex, cross-team problems. Success is measured by the ability to provide reusable methodologies and architectural insights that inform long-term investment. This role bridges the gap between experimental prototypes and reliable systems that serve millions of global users. ## Foundation Model Product Management Managing a foundation model involves treating internal teams as primary customers. The product lead must translate diverse requirements into specific model behaviors through data and evaluation criteria. Success depends on proving that a shared model can meet quality, latency, and cost requirements across multiple distinct user-facing Spotify product experiences. ## Experience-Layer Product Strategy The experience PM builds the final interface where users interact with AI capabilities. Unlike traditional software, the input here is open-ended and probabilistic, requiring new evaluation instruments. This role focuses on turning ambiguous user requests into delightful, trustworthy interactions while managing the system's response when confidence levels are low. ## Shared Company Priorities Despite different daily focuses, all three roles align under the same company priorities and serve the same external users. Collaboration is essential because an idea from research often becomes a model capability, which then enables a new user experience. This interconnectedness ensures that technical advancements translate into tangible listener value. ## The Challenge of Evaluation Evaluation is a continuous process rather than a pre-launch phase. For research, it involves assessing model behavior; for models, it means defining criteria for recommendations; for experience, it requires building custom feedback loops. Each layer must adapt its measurement tools to ensure improvements are meaningful for their specific problem space. ## Navigating Technical Constraints Building AI products requires balancing complex technical constraints with existing system commitments. Research leads must identify which questions matter across teams, while model PMs must ensure their systems integrate seamlessly. Experience PMs handle the reality of latency and reliability, ensuring that the final product remains functional for hundreds of millions. ## Influence and Strategic Context Research impact often takes months to manifest in roadmaps, making influence a critical skill. The research lead provides strategic context without prescribing answers, connecting work to decision-making windows. By involving relevant teams early, they ensure that research findings are actionable before the opportunity to implement them has already passed. ## Data-Driven Model Training In foundation model management, behavior is learned through training rather than specified in code. The PM decides what the model learns, which data is used, and how to verify the results. These decisions shape the roadmap for each training cycle, directly impacting how well the model serves various product surfaces. ## Handling Open-Ended Inputs AI experiences differ from traditional products because user inputs are often vague or complex. While traditional software relies on predictable button taps, AI agents must interpret intent. This shift requires PMs to design for ambiguity, ensuring the system remains useful even when requests do not map to specific catalog entities. ## Continuous Feedback Loops Experience teams build custom datasets from real usage patterns to identify where the system falls short. By using LLM-based judges and custom evaluators, they measure whether changes represent genuine improvements. This direct feedback from user interactions is vital for refining the product and closing the gap between general and specific. ## Compounding Organizational Impact Success in AI product management is achieved when these three layers work in harmony. When research provides leverage, models offer reusable capabilities, and experiences deliver value, the impact compounds. This specialized approach allows Spotify to navigate the fast-paced evolution of AI while maintaining a focus on listener-centric product development. ## Sources covering this story - https://engineering.atspotify.com/2026/11/ai-product-management-at-spotify-a-view-from-three-angles/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._