The article explores the intricacies of vector search, emphasizing the necessity for specialized solutions over general-purpose databases to handle high-dimensional data efficiently. It highlights how vectors, or embeddings, are heavy transformations of data that require significant storage and specialized indexing techniques like the Hierarchical Navigable Small World (HNSW) to maintain performance during search operations. The comparison between ACID and BASE models illustrates the need for BASE-oriented architectures in vector search to prioritize availability and scalability. The discussion delves into the challenges of filtering within vector search—addressing the limitations of pre- and post-filtering techniques—and introduces the concept of filterable HNSW as a solution. The article argues for the superiority of dedicated vector databases, such as Qdrant, which are designed to incorporate advanced features like GPU-accelerated indexing and multivector support, ensuring they stay at the forefront of innovations in AI and data retrieval. These databases are presented as crucial for applications requiring high-speed, large-scale vector search capabilities, such as recommendation systems, big data analysis, and dynamic data exploration, underscoring that traditional databases with added vector capabilities cannot match the efficiency of purpose-built systems.