Organizations ranging from startups to cloud vendors are releasing AutoML platforms designed to accelerate ML workflows, enable rapid prototyping, and democratize access to model development. However, most AutoML systems still require some manual work. To understand how these systems are used by data practitioners, including modes of human-machine collaboration, Tableau researchers interviewed 29 individuals at organizations of various sizes and firmographic profiles. Specifically, they examine how data practitioners use visualizations when interacting with AutoML systems. They find that most companies struggle to implement and glean value from AutoML technology (e.g. due to obstacles associated with data preparation, debugging deployed models, etc.) and encounter challenges with visualization tools designed to facilitate this work.