Grounding AI Agents in Business Meaning: Semantic Layers, Ontologies, and MCP Elicitation in Practice

Sreeni Ramadorai argues that AI agent failures often stem from a lack of shared business context rather than model intelligence. To bridge this gap, the author proposes using semantic models and ontologies to formalize business terminology and relationships. By implementing a 'semantic layer'—a governed glossary of terms—agents can look up definitions instead of guessing. The article demonstrates a practical approach using the Model Context Protocol (MCP), where agents are restricted to approved data paths and forced to halt when encountering undefined terms. Furthermore, the author highlights the use of 'elicitation' to handle ambiguity and require human approval for high-stakes actions. By treating agents like new employees who require clear onboarding and defined boundaries, developers can transform tribal knowledge into structured, machine-readable rules, ensuring that AI agents perform tasks exactly as intended within a specific business environment.