This article documents the multidisciplinary design and evaluation of a robot to assist carpentry workers on a building site in the United States. Combining ethnographic and technical design research methods, it contributes a path towards the human-centered development of construction robotics technologies. Since the 1980s, researchers have envisioned robots performing skilled construction tasks such as bricklaying, painting, spray-coating, or site monitoring. In this article, we envision instead robots performing simpler, ancillary activities supporting workers while they collaboratively carry out building tasks. We draw from an extended ethnographic engagement with construction workers to inform the design of a prototype rover able to accompany and deliver tools to carpentry workers installing formwork panels. Following a review of the state of the art in construction robotics and ethnography in technology design, we show how insights drawn from our ethnographic study informed the robot’s design as well as its innovative deep reinforcement learning (DRL) architecture for social navigation. Evaluating the robot in simulations, lab settings, and on a construction floor we document its benefits, including apt social navigation and user comfort in construction floors, and reflect on its limitations. Proposing “robot in the loop” as a design pattern combining ethnographic and technical design research, the article shows how the world of construction might be brought closer to the world of technology design, centering workers’ contexts and experiences in the design of new technologies aimed at supporting them.