In our conclusion, we instead distill practical guidance for working in human-centered ML. We argue that the field is irreducibly interdisciplinary, requiring the kind of "Renaissance teams" Bill Buxton describes: groups where members bring deep expertise in their own domain and sufficient literacy in others to collaborate effectively. We stress that human-centeredness must be built into the ML pipeline from the start rather than added at the end, particularly in data collection and labeling, where shortcuts tend to carry costs that surface later. We take the opportunity to call for attention to the full range of stakeholders involved: not just direct users, but crowd workers, impacted communities, and people across different social scales, economic circumstances, and geographies. We then close by arguing that human-centered ML is an ethical practice, and that human-centered approaches, despite the name, must ultimately extend to humanity and the planet as a whole. We make a final argument that human-centered ML is not a mere specialization within the field but as a necessary condition for machine learning to have positive effects on people and society.