Dash initially functioned like typical enterprise search systems, focusing on retrieving information through a traditional RAG pipeline. As users began to rely on it for more complex tasks—like interpreting and summarizing data—the limitations of straightforward retrieval became evident. To adapt, Dash needed to transform from a search tool into an agentic AI capable of reasoning and taking action. This shift introduced challenges around context engineering, defined as the process of providing the model with only the necessary information at the right time to enhance decision-making. The introduction of new tools in Dash's workflow initially led to slower, less accurate outcomes. Each added capability expanded the model's decision space, creating confusion rather than clarity. The Model Context Protocol (MCP) aimed to define and describe the tools available to the model, but adding more tools often consumed valuable context space without improving accuracy. This prompted a reevaluation of context management. Dash's approach now emphasizes three strategies: limiting tool definitions, filtering for relevance, and employing specialized agents for complex tasks. By consolidating retrieval options into a single tool backed by the Dash universal search index, the model's reasoning became more efficient. The creation of a unified index and a knowledge graph helped rank results based on relevance, ensuring that only meaningful context reached the model. For complex tasks like query construction, the model was separated from these details through the introduction of specialized agents. This division allowed the main planning agent to focus on execution while the search agent handled specific retrieval tasks, improving overall efficiency and performance.