Enterprise AI knowledge graphs move into production focus, Microsoft VP says
Microsoft’s Jeevan Pathuri said knowledge graphs can cut AI agent build times by giving enterprise models reusable business context.
By Dominic Okoye · Staff Writer
· 3 min read
Enterprise AI knowledge graphs are becoming a practical requirement for companies trying to move AI agents out of pilots and into production, according to Jeevan Pathuri, Microsoft Corp.’s vice president of software engineering. In an interview with SiliconANGLE Media’s theCUBE, Pathuri said connected data models can give language models the business context they lack when they operate only on generic training and isolated queries.
The discussion reflects a broader enterprise AI shift: buyers are spending less time talking about model access and more time asking whether their internal data is clean, connected and governed enough to support agents that make useful decisions. The funding market has followed that theme. SiliconANGLE reported in June that context-graph startup Jedify Inc. raised $24 million to help enterprise AI agents use company-specific business knowledge. Investors in that round were not detailed in the interview.
Why do enterprise AI knowledge graphs matter?
A knowledge graph represents data as entities and relationships, rather than only as rows in a table or documents in a repository. For AI systems, that structure can make it easier to reason across products, suppliers, customers, policies and workflows without rebuilding the same context for each use case.
Pathuri argued that the main value comes from the links among enterprise data, not from the data objects in isolation. He said modeling information as nodes, edges and relationships helps language models answer more complex business questions than a conventional SQL query can handle on its own.
Microsoft’s supply chain work was the concrete example Pathuri gave. He said that representing a bill of materials as a graph helped Microsoft build 10 to 15 production agents within a few weeks. Without that shared layer of connected context, he said, the company would have had to create agents one at a time while repeatedly reconstructing relationships and metadata.
Reusable context is the production bet
The operating claim is straightforward: if the semantic layer is built once and made reusable, teams can shorten agent development cycles across multiple workflows. Pathuri said an agent that previously took three or four weeks to build can now be created in a few days when the necessary business context is already defined.
That does not remove the data work. Pathuri said enterprises still need a clean and connected data layer, plus evaluation and production traces to keep humans in control of the system. The interview did not disclose Microsoft spending, revenue impact, headcount or any new product tied to the work.
Market forecasts are rising with the same assumption. Promethium.ai projects the enterprise knowledge graph market will increase from roughly $1.9 billion today to nearly $10 billion by 2032, driven by demand for governed context in AI deployments. Forecasts in this category remain directional, but they show where vendors are positioning the next layer of enterprise AI infrastructure.
SiliconANGLE said Pathuri spoke with theCUBE’s John Furrier as part of an AI Luminaries interview series. TheCUBE disclosed that it is a paid media partner for the Neo4j AI Luminaries series and said sponsors do not control editorial content.
This story draws on original reporting from SiliconANGLE.
