Layered data architecture turns enterprise data into a system of intelligence
SiliconANGLE Ryan Stevens
Enterprise AI is leaning on knowledge graphs to connect messy data. The big shift isn’t speed; it’s getting answers leaders can trust.
Based on reporting by SiliconANGLE, Ryan Stevens — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Enterprise data is being reorganized into layers, and knowledge graphs are moving into the center of that stack. Tristan Baker, senior director and head of data architecture at Salesforce, says they’re becoming the piece that ties an AI agent’s question to the context needed to answer it. He made the case during a theCUBE interview at Neo4j GraphTalk, where the discussion turned to how companies are trying to build an enterprise system of intelligence.
The pressure comes from a familiar problem: business data is scattered across lakehouses, operational databases and customer profile stores, each built for a different job. Baker’s answer is not a single magic database, but a stack. Underneath are the systems that hold the raw data. Above them sits metadata that helps track where the truth lives and translates business terms into the right columns. The graph then connects the relationships and context that make the whole thing usable.
That matters because leaders don’t just want a number thrown back at them. They want to know how it was derived, what it touches and what else it’s connected to. Baker said nobody has solved that perfectly yet, and he was blunt about the danger: even the best tooling turns useless if the content feeding it is messy.
Governance may be the part people underplay most. Baker argued that access control can’t stay buried inside individual databases once metadata and meaning are lifted into a higher semantic layer. Legal and policy teams won’t write rules in database terms, he said; they’ll describe who should not see what. That pushes master data management and governance into the same conversation, which is probably where they belonged all along.
My take — AI-written commentary, not fact-checked reporting
This is the unglamorous truth of enterprise AI: the model is rarely the hardest part, the data plumbing is. Vendors love to sell magic, but the real work is mapping meaning, relationships and access rules without turning the whole thing into a compliance séance. Open or closed models matter less here than whether the company can stop its own data from lying to it.
Read more about this at: SiliconANGLE