We went back and forth on this for an app my team was building and ended up going with Pinecone for a few reasons. One is that vector data is derived data, I don't want to manage it with the same care that I do my operational data. Second, for us we were using a RAG pattern to perform similarity searches on vectorized content and stuffing it into a prompt that we sent to the LLM. The developer experience of doing this with Pinecone was drop dead simple; tailored specifically for doing kANN searches. Our production database is Mongo which had very unproven vector capabilities and which did not perform well in our tests at the time. The obvious downside here is that now we have another database to manage and we take on the additional complexity in our architecture along with additional costs. Initially my thinking was in line with the SingleStore CEO, but the more I have thought about it, the more comfortable I am with the idea that vector data is specialized and has different management requirements and is better suited to live in a specialized data store. Just my personal opinion...