Migrate SageMaker Pipelines to ZenML

This skill translates Amazon SageMaker Pipelines and related workflow code into idiomatic ZenML pipelines. It handles the full migration workflow: analyzing the SageMaker pipeline, classifying each construct, translating what maps cleanly, flagging what needs redesign, and producing a working ZenML project.

How migration works at a high level

SageMaker Pipelines and ZenML can run on the same AWS substrate, but they ask you to think about pipelines differently. Native SageMaker Pipelines is a step-type DSL: you choose ProcessingStep, TrainingStep, TuningStep, TransformStep, ConditionStep, and other managed-service wrappers, then wire them with pipeline parameters, step .properties, PropertyFile, and JsonGet.

ZenML is Python-step-first: you write @step functions that consume and produce typed artifacts, and the active stack decides where they run. When you use ZenML's SageMaker orchestrator, you are usually not leaving SageMaker. You are keeping SageMaker as the execution backend while changing the authoring model to ZenML pipelines and stack settings.

That means migration is not a rename-the-primitives exercise. Some patterns translate directly, some are approximate, and some require deliberate redesign.

The three mapping types

Every SageMaker concept falls into one of these categories: