As a community, we’re in the process of crossing that river right now. A few years ago it was an accomplishment to get a text based CRDT working at all. Now implementations are starting to compete on features and performance, and they’re starting to see some use in real world applications. But there’s still some edge cases to iron out and understand in terms of memory size and pruning and things like that.

In a few years the rough edges will be ironed out and well understood, and there will be a good set of CRDT implementations you could use without worrying about this stuff. I think Yjs might already be there.