Semantic layers, which have long been integral to business intelligence tools, are becoming essential for deploying agentic AI systems effectively, as discussed in an episode of the Agents of Data Podcast featuring Frank Weigel, Julian Wiffen, and Sam Perrin from Matillion. A semantic layer serves as a bridge between natural language and technical systems, translating company-specific terminology and connecting business concepts to data structures. This becomes crucial for AI agents, like Matillion's Maia, to accurately perform tasks such as building sales reports by identifying relevant data, applying business rules, and navigating complex data ecosystems. The podcast emphasizes the importance of semantic layers in overcoming the hidden knowledge problem, where fragmented institutional knowledge is often trapped in the minds of employees, outdated documentation, or system metadata. The adoption of graph databases is explored as a flexible solution for building semantic layers, allowing for dynamic relationships and context to be captured. As AI systems take on more data management roles, data professionals are shifting from routine data tasks to curating semantic layers and validating AI outputs. The ownership and maintenance of these semantic layers raise questions about centralization versus departmental control, but investing in semantic layer development is seen as a strategic move for organizations looking to enhance AI reliability and reduce human intervention.