We gave our agent the exact metric definition. It still wrote the wrong SQL (clarilayer.com)

Anthropic and OpenAI recently revealed insights into the challenges of improving data agents, concluding that context management is the primary hurdle, rather than SQL generation itself. While both companies highlighted the importance of structured procedural knowledge and context layers, they assumed access to well-organized datasets and data teams, overlooking individual analysts who often lack such support. This gap inspired the development of ClariLayer, a tool designed for individual analysts working with disorganized data, enabling them to efficiently use AI-powered agents like Claude Code, Cursor, and Codex to generate accurate SQL queries without a dedicated data team. Key technical observations include Anthropic's findings that accuracy drastically improves with structured procedural guidance, demonstrating that simply increasing context leads to erroneous outputs. OpenAI emphasized the architectural complexity required for meaningful context, structuring it into effective layers, yet both faced issues with trust and conflicting information from multiple sources. ClariLayer remedies these issues by transforming metric definitions into structured data and incorporating proactive recall protocols. This product aims to empower analysts to manage context effectively, reducing errors and ultimately enhancing data-driven decision-making in environments where traditional data governance is absent.