A year ago, most ML engineers I talked to had never heard of vector databases.

Now it's in every other job description.

Here's the quick version of why they matter: modern AI represents everything as high-dimensional vectors (embeddings). To find similar items — whether it's documents for RAG, products for recommendations, or images for search — you need a database optimized for similarity search on those vectors.

Traditional databases are built for exact matches. Vector databases are built for "find me the closest thing to this."

I've used most of them at this point. Pinecone if you want managed and production-ready. Weaviate for feature-rich open source. Chroma for quick prototyping. Qdrant if you care about performance. pgvector if you're already married to PostgreSQL.

My honest recommendation for most people starting out: Chroma for learning, then Qdrant or Weaviate for production.

Understanding embeddings and vector search isn't optional anymore for AI engineers. It's the new SQL — foundational knowledge you're expected to have.