You seem to think that this is just a service that reads and writes data like any CRUD app. The hard problems stem from how the system deals with failures and

You seem to think that this is just a service that reads and writes data like any CRUD app. The hard problems stem from how the system deals with failures and how the system propagates writes across the replicas while meeting latency and consistency SLAs. On top of that the system needs to be built in a way that it can be maintained by many developers each working on a small piece of the system without knowing the ins and outs of the system as a whole. In addition, when the system fails debugging and mitigation needs to be able to be parallelized across many developers so that availability SLAs can be maintained. You can read about this in “Designing Data-Intensive Applications” by Martin Kleppman where he discusses the complexity involved in building distributed systems.