Market Moves Toward Unified SQL and RAG Tooling
The enterprise search market is seeing a push for unified tooling that combines traditional SQL with RAG systems for unstructured data. This trend is exemplified by companies like Mode, which recently released a new SQL editor designed to serve both data analysts and agentic workflows, blurring the line between structured and unstructured data analysis.
- Mode's parent company, ThoughtSpot, acquired it in a $200 million cash and equity transaction to combine Mode's code-first approach with ThoughtSpot's natural language, self-service analytics. This deal is expected to increase ThoughtSpot's annual recurring revenue to over $150 million. - The unification of SQL and RAG is driven by the need to analyze the 80-90% of enterprise data that is unstructured and growing faster than structured data. Traditional SQL is insufficient for formats like text, images, and audio, which require techniques like natural language processing. - A key technical challenge in this unification is creating a common master data model that can apply consistent business terminology across both structured tables and unstructured content. Without this, it's difficult to accurately join insights from both data types. - Competitors are also addressing this trend. Glean, for instance, uses a knowledge graph to map relationships between content, people, and activities, combining vector and lexical search in its RAG systems to enhance relevance. - The evolution of RAG is moving towards "Agentic RAG," where AI agents can autonomously decide whether to query a SQL database for structured data or a vector database for unstructured information. This allows the system to handle more complex, multi-step queries that require reasoning. - This agent-based approach often involves a multi-tool architecture. For example, one agent might transform a user's natural language question into a SQL query, another validates it, and a third retrieves unstructured context from a vector database before synthesizing a final answer.
