As an early Machine Learning practitioner working with a small number of models it was easy to think that wrapping a model in an API was all that was needed. Since then, having worked with customers that deploy thousands of models at scale, I’ve realised quite how naive that worldview was. The road to MLOps best practice is long, hard and poorly defined… In this talk I’ll cover the mistakes I’ve made, what I’ve learned along the way, and how DevOps principles are bleeding over into the field of ML.