This resonates with me. I think that a product ML team can only be successful in the long run if its members can pivot between modeling work and engineering work as the business needs change. There is an entire class of communication and prioritization problems that evaporate when the same people who build the models are pushing them to production.