In a new essay, Eugene Yan makes the case that data scientists should stretch beyond building models in isolation and take on more of the overall lifecycle. He argues that the common split between data science and engineering teams creates unnecessary bottlenecks, and that a more end-to-end approach lets data scientists see how their work actually performs in production. That visibility, he suggests, leads to faster iteration and better outcomes.
Yan draws on experience at Stitch Fix and Netflix, where data scientists were expected to handle parts of the pipeline that many teams treat as separate responsibilities. Both companies found value in having data scientists participate in deployment and monitoring, rather than handing models off to an engineering team and waiting for feedback.
He is careful, however, to note the trade-offs. End-to-end work can pull data scientists away from deep technical focus and demands a different skill set. Yan acknowledges that the model isn't a universal fix: it works best in environments where engineering support is scarce, or where product and data work are tightly coupled. The piece is less a prescription and more a weighing of costs and benefits, making it a useful reference for teams deciding how far their data scientists should go.