Data teams need to shift towards context engineering, which integrates data governance, engineering, and science to create reliable sources of truth for company knowledge. The article compares current internal AI agents to outdated BI tools that directly connected to production databases, resulting in unreliable outputs. Just as data stacks were developed to ensure accurate reporting, a context stack is necessary now to manage AI agents effectively. Context engineering aims to optimize factors such as response accuracy, speed, and costs, while addressing the trade-offs between too little and too much context. To implement context engineering, teams must select appropriate sources of information, clarify definitions, and format data for optimal parsing by AI models. This mirrors traditional data engineering principles, emphasizing measurement and iteration. The need for governance parallels past data governance challenges, where varying interpretations of metrics led to confusion. Companies face similar issues with knowledge sources, where outdated or conflicting information can mislead AI responses. Establishing a context layer will provide a single, governed source of truth, essential for maintaining accuracy in AI-driven interactions. The article highlights the current lack of tools to build a context stack, noting that monitoring and evaluation frameworks for AI performance are still developing. Some data teams have begun creating scripts to manage context, but this approach can be cumbersome. The author suggests initial steps for teams include demonstrating context engineering within their analytics agents and exploring tools like file-system AI agents, which allow for more manageable context control and evaluation. The goal is to refine and improve AI agent performance through structured context management, paving the way for more reliable implementations in the future.