From the early 1920s to the end of 1930s, US meteorologists implemented a program of weather observation flights by airplanes as a chief method of collecting data from the upper-air. The seemingly well-defined task of obtaining more weather data as quickly as possible posed serious technical and managerial challenges to the meteorologists, who struggled to fix both the transient air and the pilots' risky weather flights. The relationship between meteorologists and pilots mediated through the recording instrument - meteorograph - reveals that the practices of collecting and representing scientific data were intimately entangled with various questions of professional identity, control in scientific practice, and the contested meaning of the 'automatic.' Putting the instrument and the project in this context enables us to understand why the new instrument 'radiosonde' that finally replaced the airplane method was called the 'robot observer.'.