SageMaker is more suitable for stages that have entered scaled training/inference and require managed endpoints and model version management; introducing it at the MVP stage brings unnecessary complexity.