Why it is being added
Retrieval-augmented AI features need similarity search. Running that beside existing relational data avoids synchronising two systems.
When a dedicated store still wins
Very large vector sets and high query volumes, where a purpose-built index outperforms a general-purpose one by a wide margin.