Some casual experimentation have shown a 2-3x difference in the download size for large requests between CSV and JSON. Previously we jettisoned the CSV format in transit because of some strangeness around readr::read_csv's parsing. But if JSON parsing leads to more OOMs than necessary, maybe we should reconsider.
It is unlikely that the interface will need to change at all. We will however need to make sure that the data type behavior stays the same as it is.