The base station / small cell can have antennas with a large number of patches (32 to 256), so very narrow beams and high gain. This is used to compensate in part the very high attenuation at those frequencies. The device will have less antennas, so less gain.
Ideally, both device and cell will point their beams at each others, for maximum gain. There's a beam searching process as part of the 5G/mmWaves synchronization, and then beam tracking. All this is much easier when the device is static, for fiwed wireless access (FWA). When handling a mobile it gets trickier.
It's best to have a line of sight between the close small cell and the device, but it's also possible to have an indirect line with a bounce for example. There will be an impact on signal level, so if there's such a bounce you'll want to be closer.
For mobility, tracking a device is hard. Just turning a street corner or moving behind any object in the beam path can drop the mmWaves link. But today mmWaves for smartphones are deployed in NSA mode (non-stand alone), where it's always used in parallel with LTE in real time (dual connectivity or DC). Actually, all those device antennas consume so much that it's common to only use 5G/mmWaves in the downlink (cell to device) and stick to LTE in the uplink (device to cell). Reception is easier on the device as there's no power amplification as in the uplink. So when you drop the 5G connectivity, the LTE link is already there.
All those additional antenna modules explain why 5G phones are on the big side.