Emil Eifrem uses a simplified bank-account-opening agent to explain the costs of manually connecting every agent to enterprise data. Repeated discovery, duplicated records, changing schemas and isolated learning make that approach difficult to maintain.
Emil Eifrem proposes three shared components: a business-facing ontologyAn ontology is an explicit representation of the concepts, categories and relationships used to describe a domain., a technical ontology describing data sources and a mapping between themInteroperability is the ability of independently built systems to exchange information or work together correctly., plus execution tracesAn execution trace records the sequence of operations and results produced while a system carries out a task. showing what worked. The example connects identity-verification intent with appropriate data services and uses outcomes to inform later choices.
Emil Eifrem argues that this shared semantic layerContext engineering designs how relevant instructions, evidence, state and tool information are assembled for a model's task. can centralize governanceGovernance defines decision-making authority, rules and accountability within a system or organization., reduce repeated wiring and support learning across agents. The talk presents a blueprint and reported enterprise experience, not a controlled demonstration that every organization will obtain those benefits.
Watch on YouTube




