I wouldn't be surprised if most developers using Cassandra are coding direct to the metal via the native Thrift API, or via the Hector or Astanyax APIs (and the non-Java equivalents), rather than using the newer, but far more limiting, CQL.
CQL simplifies some common use cases (think of it as SQL where the raw datatypes include collections as well as scalars). CQL 2 and CQL 3 don't sit well together, and in my testing they couldn't see data the other had inserted due to how they interpreted the database structure. IMO, avoid unless it handles your use case perfectly.
However people are interested in the replication and redundancy features of Cassandra despite having SQL-style problems, and maybe this solution will meet their requirements. It's got to be better than MySQL master-slave replication :-)