APOC in general is there to fill some gaps in functionality in Cypher, but some features in Cypher have caught up over the years.
One prime example is apoc.periodic.iterate() for batching changes from streamed input, which could be from a Cypher query of the graph itself. Cypher didn't have a good answer for that until 4.4.x, when we introduced subqueries for transaction batching:
APOC path expansion procedures (such as apoc.path.subgraphNodes()) can be useful for some cases that aren't yet well supported in Cypher, such as expansion to distinct nodes in a graph. The path expanders have a way to prune paths when you reach a node that has already been visited, that can keep down the number of distinct found paths needing further processing, and cutting out that work can improve performance. The path expanders can also handle repeating sequences during expansion, something that is not yet possible (but actually not too far off) in Cypher.
APOC helper functions (for coll, map, convert, agg, and others) are awesome, providing some great utility at a very fine level. The converting to/from JSON strings is a strong highlight here. These alone can justify using APOC, and eventually some of these make their way into Cypher proper.
Connection procedures, for example loading via JDBC or other connectors, can also be very useful when you're pulling from different data sources.
