The article explores the challenges of traditional semantic layers in business intelligence (BI) tools. Each tool—like Looker, Tableau, and Power BI—has its own way of defining metrics, relationships, and data models, which often leads to duplication of effort and a lack of consistency across platforms. These definitions are confined within their respective tools, creating a chaotic environment where data definitions drift apart and complicate workflows. The so-called "gold layer" in data warehouses typically ends up being a mix of quick fixes and denormalized tables, shaped to work around the limitations of these tools. To address these issues, a trend emerged between 2020 and 2022 where companies began to extract semantic layers from BI tools into standalone products like Cube and dbt Semantic Layer. However, implementing these solutions often involves cumbersome YAML files, manual annotations, and a steep learning curve, making it a tedious process. Enter the Boring Semantic Layer (BSL), championed by Julien Hurault. This approach prioritizes simplicity and predictability over complexity. BSL offers deterministic definitions, eliminating the need for overly complicated semantic reasoning that often fails at scale. It integrates seamlessly with Python, allowing developers to work within familiar environments without the overhead of additional infrastructure. The article also highlights the integration of BSL with dlt and LLMs (Large Language Models) using the Model Context Protocol (MCP). DLT automates schema discovery, capturing foreign keys and metadata without manual input. This data is then fed into an LLM, which generates a structured semantic model. The output can be consumed by various interfaces, like Streamlit for visualization or a FastAPI server for RESTful queries. This setup allows users to access the same data model through different channels, streamlining the process and reducing redundancy. By shifting the LLM's function from generating raw SQL to querying a validated semantic layer, the risk of errors is minimized, making it easier for organizations to use natural language for data interactions.