Rivers can look normal on the surface and still be running dry underneath. Streamflow drought happens when rivers and streams drop below normal levels for weeks or months, shaped by snowpack, groundwater, and reservoir operations as much as by rainfall, which is why it can persist even after the rain returns. Until now, water managers had little advance notice of it. River DroughtCast, a new machine learning tool developed by the US Geological Survey, changes that: it forecasts streamflow drought one to thirteen weeks ahead at thousands of sites across the United States, each forecast paired with an explicit measure of uncertainty. It is the first operational tool of its kind to combine multiple drought intensities and forecast horizons at national scale. We spoke with John Hammond, the research hydrologist who led its development, about how the model was built, what it can and cannot yet do, and where it is headed next.