Some data practitioners have asserted that Kafka can replace a relational database. In this paradigm, users would append events to Kafka and read from downstream views representing the present state. However, to realize this vision, users would need to implement a DBMS with ACID guarantees in their application code. In this post, Arjun Narayan (CEO of Materialize) and George Fraser (CEO of Fivetran) describe some of the problems that DBMS address, like isolation anomalies, which often occur in event-sourced architectures. They recommend using Kafka alongside a database (with CDC for event generation) to better isolate transactions.