• MLOps is a relatively new term that began to gain traction at the end of 2019.
• The rapid growth of machine learning sparked the creation of MLOps as a new category.
• MLOps vendors can be split into several categories: deployment & serving of models, model quality & monitoring, model training, and feature stores.
• The deployment & serving of models and model quality & monitoring have parallels in software engineering and data engineering.
• Model training is a data pipeline and should be part of the data engineering discipline.
• Feature stores are similar to a typical data infrastructure architecture used by companies that require both streaming and batch processing capabilities.
• Vendors should focus on providing useful tooling to support MLOps operations rather than duplicating existing data infrastructure.
• The focus should be on building tooling and practices for data, ML, and product engineers to work together more effectively.