# How Spotify Selected and Scaled Its AI Observability Platform

Spotify’s presentation explains how its Hendrix AI/ML platform team evaluated observability and evaluation platforms for the company’s growing set of agentic features and why it chose a shared approach instead of leaving each team to work independently. The speakers define observability as seeing what an agent is doing and why, and evaluation as continuously measuring quality so agents do not degrade over time or create unsafe, low-quality experiences. They compare building an in-house platform, which offers control over Spotify’s bespoke use cases, with buying from a vendor, which reduces maintenance burden and benefits from a broader customer-driven roadmap. Spotify then describes a three-stage vendor selection process using representative internal agents and scoring vendors across 11 dimensions, with scalability, API extensibility, and vendor team strength emerging as the most important differentiators. The resulting shared platform enables consistent, apples-to-apples evaluation across teams, while Kong’s AI gateway automatically captures traces and supports multi-model evaluation at Spotify’s scale.