As a long-time Netflix user, I have always been fascinated not only by the stories on screen but also by the company’s constant reinvention behind the scenes. Netflix’s tech blog often reveals that evolution—how data, infrastructure, and machine learning shape the viewing experience for millions worldwide. A recent piece, “From Facts & Metrics to Media & Machine Learning: Evolving the Data Engineering Function at Netflix” captures a pivotal moment: the recognition that data engineering must transform when the raw material is not transactions or events, but media itself.
The essay describes a shift away from pipelines centered on structured metrics toward a model where audio, video, images, and text are treated as first-class data assets. To support this, Netflix has built a Media Data Lake and created a new specialization—Media ML Data Engineering—dedicated to storing, versioning, enriching, and serving media for machine learning. It is a redefinition of scope: data engineering not as the custodian of numbers, but as the enabler of media intelligence.
What makes this transition remarkable is how it reframes the boundaries of the discipline. In the metrics-driven world, the job was to ensure accuracy, lineage, and timely access. In a media-driven world, the responsibilities expand: embeddings must be generated, multimodal queries must be supported, and pipelines must integrate seamlessly with creative and research workflows. The data engineer is no longer reporting on what has already happened, but preparing content to fuel models that predict, enhance, and even create.
The implications ripple outward. Data engineering now sits closer to machine learning research, to experimentation, and even to creative production. Teams must balance the rigor of standards with the flexibility required by artists and scientists alike. They must also contend with scale: high-resolution video and complex audio dwarf traditional datasets in size and complexity. The role becomes less about maintaining dashboards and more about making previously inaccessible forms of data usable at speed and scale.
This evolution, however, is not without friction. Building a media-centric platform is costly, and its value must be demonstrated to justify the investment. Different disciplines—engineering, ML, and creative—do not always share vocabulary or priorities, which can make collaboration difficult. And for companies without Netflix’s resources, the same vision may remain aspirational for now. These tensions underline the broader point: as the nature of data changes, so too must the practice of engineering around it.
What the article ultimately conveys is that data engineering cannot be defined by yesterday’s artifacts. Metrics and dashboards will remain important, but they are no longer enough. The future lies in treating media as data, with all the technical and cultural shifts that entails. For Netflix, that means building infrastructure for media-aware machine learning; for the wider field, it signals that the scope of data engineering is expanding.
And perhaps that is why the essay resonates so strongly. Reading it, I found myself seeing the shift from three vantage points at once: as a viewer, reassured that the service I use daily is powered by forward-looking technology; as a reader, drawn in by the clarity of the narrative; and as a data professional, reminded that our field is not fixed but constantly reinventing itself. Netflix’s story may be unique, but the questions it raises about the future of data engineering belong to all of us.