ClickHouse has evolved its internal data warehouse (DWH) from a traditional BI-first approach to an AI-first model, now catering to over 300 users and managing 2.1 petabytes of compressed data. This shift aims to streamline the analytics process, allowing users to extract insights without writing complex SQL queries. The previous system created significant bottlenecks; teams like Engineering, Sales, Product, Finance, and Cost Optimization faced delays from having to write intricate SQL to answer even simple business questions. What should take minutes often turned into hours of work, leading to a reliance on a small analyst team and a frustrating wait for insights. Recent advancements in AI technology, particularly with the release of Anthropic's Claude 3.5 Sonnet and later versions, have transformed this landscape. These models can now write complex SQL queries, correct errors based on database feedback, and consider extensive business context when generating insights. The introduction of the Model Context Protocol (MCP) allowed seamless integration between LLMs and various data sources, eliminating vendor lock-in and enabling ClickHouse to develop its own MCP server. This integration is key to making AI a practical tool for data analytics. Building a robust AI-first analytics platform involves more than just connecting databases. ClickHouse has established a comprehensive business glossary that provides necessary context for the data, ensuring LLMs can generate accurate analyses. They’ve also implemented enhanced data quality processes to address potential issues that LLMs might misinterpret, which could lead to incorrect conclusions. By focusing on these foundational aspects, ClickHouse aims to empower users to leverage AI for data insights effectively, reducing the friction that previously hindered their analytics capabilities.