Intracortical brain-computer interfaces hold the potential to improve the quality of life for patients living with motor control disorders. However, a critical barrier to the successful clinical translation of these devices is recording instability, which, if unmitigated, can quickly cause control to deteriorate. Recent findings have indicated that high-dimensional neural population activity resides in a low-dimensional “neural manifold”. Here I will introduce the concept of neural manifolds and briefly recap recent findings showing that neural manifolds constrain the types of brain-computer interface mappings that can be easily learned. Finally, I will show how these neural manifolds can be leveraged to mitigate the effects of neural recording instability, enabling stable control in the presence of even severe recording instabilities.