Engineers Shouldn’t Write ETL: A Guide to Building a High Functioning Data Science Department

The article argues that the chronic friction between data scientists and engineers comes from a legacy Thinker-Doer split, where engineers are reduced to writing and maintaining ETL pipelines. That model creates mediocrity, misaligned incentives, and endless frustration for both sides.\n\nThe root cause is a BI-born structure that was simply rebranded for the Big Data era. Companies copy a three-group hierarchy-data scientists, data engineers, infrastructure engineers-without questioning whether the "data engineer" role, essentially an ETL specialist, still adds value when the volume of data does not require massive scaling. The result is a team of bored engineers and data scientists stuck in an assembly-line handoff.\n\nThe proposed solution flips the ownership model: data scientists keep end-to-end responsibility for their pipelines and production code, while engineers focus on building horizontal platforms, services, and abstractions that any scientist can reuse. Engineers become the providers of Lego-like building blocks-frameworks, scheduling tools, and scalable services-allowing scientists to iterate quickly without waiting for a dedicated ETL handoff.\n\nThis shift creates autonomy for scientists, reduces the need for a dedicated ETL role, and lets engineers apply their skills to broader, reusable problems. The result is faster delivery, fewer bottlenecks, and a healthier culture where both groups are aligned around shared outcomes rather than opposing responsibilities.\n\nStitch Fix applied this blueprint, restructuring its data platform to give scientists ownership of their code and focusing engineering effort on platform services. The experience shows that breaking the Thinker-Doer cycle yields tangible productivity gains and a more innovative data organization.