It completely changes how to think about service based architectures. In fact, with a properly designed architecture sitting around Kafka APIs and a plethora of queues magically disappear, because instead of function and request-response driven the system becomes reactively data driven, sitting around a single bus.
This makes designing actual services a lot easier because instead of thinking about a few hundred function names all you need to think about are actual data flows (data comes in at a central point -> gets transformed -> gets moved out at a central point).
As a added bonus it scales like crazy, to the point that hammering through a terabyte of data looks like nothing.
A great talk from Ben Stopford (Confluent employee, so mostly Kafka centric) about this subject can be found here:
https://www.slideshare.net/benstopford/devoxx-london-2017-re...