MongoDB's New Retrieval Features Highlight Why SQL Still Beats Vector Search for Structured Data

MongoDB this week launched three AI retrieval upgrades — context-aware embeddings, hybrid search, and a native reranker — aimed at improving accuracy in agent-based applications. These tools address known weaknesses of vector search, such as approximate relevance matching and loss of context during document chunking, but are designed primarily for unstructured text data. For structured data like CSVs, Parquet files, or SQL exports, such retrieval techniques are unnecessary, since a standard SQL WHERE clause already delivers exact, deterministic results with perfect recall and precision. Applying semantic search strategies to data that already has a schema wastes tokens and introduces retrieval errors that compound through agent retries. The core argument is that retrieval accuracy is not a model problem for structured data — it is a solved problem, and SQL has always been the answer.