Building the Context Flywheel for AI Data Agents

Building a context flywheel for AI agents requires moving beyond traditional metadata catalogs toward a comprehensive context layer that integrates institutional, semantic, and procedural knowledge. Prukalpa Sankar, founder of Atlan, emphasizes that while AI models have achieved exponential intelligence growth, their enterprise utility remains limited by a lack of shared organizational context. By leveraging AI-native infrastructure—such as a context lakehouse—organizations can automate the bootstrapping of metadata, column descriptions, and metrics, effectively teaching data the language of business. This shift enables agents to operate with higher autonomy, moving from human-in-the-loop to human-on-the-loop governance. Ultimately, successful AI-driven organizations treat context as a living, compounding asset, using decision traces and simulation environments to refine agentic accuracy and break down traditional data silos, thereby transforming data practitioners into context engineers.