Spotify has over 70,000 datasets containing petabytes of data, processing 1.4 trillion data points daily[reference:49]. Data problems used to follow a specific pattern: you'd look for the relevant dashboard, and if it didn't exist, you'd spend hours writing queries[reference:50]. The company built an AI data assistant called Vedder to solve this problem[reference:51].
The Context Layer Approach
Rather than relying on raw schemas or query history, Spotify's solution uses a "cluster model" where domain experts curate a context layer[reference:52][reference:53]. This context layer includes:
- Curated datasets with clear descriptions
- Vetted question-SQL pairs that work
- Business documentation that explains what the data means
The key insight is that a good data agent needs more than a capable model, it needs the right context[reference:54]. Raw database schemas are too noisy. Query history is too messy. A curated context layer encodes the domain expertise that human data analysts would normally provide.
Production Usage
Vedder has been actively utilized since August 2025[reference:55]. Over 2,100 Spotifiers have used it in more than 13,000 conversations[reference:56]. Users ask questions in simple English and get reliable data within seconds[reference:57]. The assistant democratizes access to insights across thousands of fast-moving teams[reference:58].
The Human in the Loop
The system keeps the human in the loop[reference:59]. Domain experts curate the context layer, ensuring that the assistant provides accurate and trustworthy answers. Models will keep getting better, but the quality of the context layer remains the limiting factor for reliability[reference:60].
Lessons for Context Engineering
Spotify's experience offers lessons for building reliable AI assistants. Raw data is too messy for AI to interpret correctly. Domain experts are essential for curating the context that makes AI useful. The context layer approach works at scale: it's how you encode domain expertise so that an AI assistant can provide consistent, reliable answers.
The context layer behind Vedder shows that the future of enterprise AI isn't just about better models, it's about better context[reference:61]. The models will keep improving, but the quality of the context you provide will determine whether your AI assistant is useful or just another chatbot.