Neo4j makes the case for knowledge graphs as shared context for AI agents

Neo4j is pushing a simple argument: enterprise AI agents get much better when they share one governed layer of business knowledge instead of each carrying its own private version of the truth.

That point came from Jesús Barrasa, the company’s field chief technology officer of Gen AI, in a conversation with theCUBE Research’s John Furrier at GraphSummit. His worry is familiar to anyone who has watched enterprise software multiply: teams solve one problem for one agent, then repeat the same work for the next one. The result is inconsistent answers, duplicated effort and a growing mess of prompts, skills and logic embedded inside individual agents.

Barrasa’s answer is a knowledge layer built from the organization’s data assets, concepts, policies and processes. In his framing, that layer is not just about retrieval. It is also about explainability. If an agent gives an answer, people should be able to trace the source data and the pieces used to produce it. That matters more once agents stop chatting and start acting.

Neo4j’s pitch is that knowledge graphs can carry that context across multiple use cases rather than forcing a company to design a giant enterprise model up front. Start with one use case, then align the next one to it. Build the layer step by step. Barrasa also argued that large language models can speed up the construction of the ontology behind that layer.

He also wants companies to measure something beyond a single agent’s output. As more agents are added, the cost of building each new one should fall. At the same time, organizations should track the cost of drift: what happens when agents diverge, and what it takes to reconcile them. That is a less flashy metric than a demo, but probably a more honest one.