Two simultaneous meetups in Berlin and San Francisco independently converged on the same architectural conclusion: agentic AI systems need context graphs as a structured memory layer. A context graph combines a knowledge graph with decision traces, provenance, and temporal validity, enabling agents to reason across workflows in an auditable way. Speakers emphasized that building context graphs is fundamentally a knowledge management problem, not just an engineering one, requiring knowledge elicitation and formal ontologies before any persistence layer. Key production principles included resolving data contradictions at build time and designing governance with full delegation chains from the start. The emerging consensus is that context graphs will become standard infrastructure in AI-capable software, analogous to unit tests.
Sort: