{{item.videoDuration}}
{{item.title}}
{{item.text}}
{{item.videoDuration}}
{{item.text}}
Key takeaways:
AI pilots often stall because existing data architecture was designed for reasoning human analysis, not AI agents that depend on consistent business meaning and context.
The Gold Layer remains critical, but AI needs new layers of data architecture above it: semantic definitions, knowledge graphs, and indexes to help agents interpret and connect data.
A shared ontology can replace fragmented departmental definitions, reducing ambiguity and the risk of AI scaling inconsistent interpretations.
Governance shifts from controlling predefined dashboards to governing what agents can access, interpret, and combine at runtime.
Most large enterprises are sitting on an AI problem they struggle to name—pilots that demo well stall on the way to production. But the issue is rarely in your model alone. It’s also in your data.
Data readiness used to mean your numbers were clean and certified. For most organizations, that was the “Gold Layer” of data lake architecture: Bronze for raw data, Silver for cleaned and standardized, Gold for business-ready data and insights. That was enough to build comprehensive dashboards that could help you make important decisions, but that’s not enough for AI. The Gold Layer remains the certified layer for trusted data consumption, but its role is expanding: it now needs to serve not only human-driven analytics, but AI systems and agents with governed, reliable data.
What was the Gold Layer originally designed to do?
In medallion data architecture, Gold was the destination. It existed because business users and business intelligence tools couldn’t reasonably query raw operational data scattered across source systems. So, engineers built a clean, modeled, pre-aggregated data layer where the warehouse ended and the business began. Your finance analysts could open Power BI, point it at a Gold fact table, and have your quarterly numbers in seconds. They never thought about Silver or Bronze layers because they never had to. Gold was the standard because it reduced complexity for humans.
How does AI change the role of the Gold Layer?
For your business to make decisions, you need faster, deeper insights. Your data may be more connected than it’s ever been, but your analysts no longer point a BI tool directly at a table. They simply ask a question in plain language, which an agent interprets through a semantic layer, which queries Gold Layer data underneath. The human never actually sees Gold, never writes SQL against it, and often doesn’t even know it exists. But the agent still depends on it completely. For the documented truth, the pre-computed aggregates, the consistent business rules, and the point-in-time numbers a regulator will accept. The work Gold does didn’t disappear, but its audience changed from a person to a process.
AI can traverse raw, diffuse data at scale humans can’t, but it doesn’t have the institutionalized knowledge or human judgement to handle semantic ambiguity. What it needs is a new layer of organization. Your Gold Layer remains important, but it stops being the top story of your data architecture. A three-part penthouse now sits on top—a semantic layer that publishes certified metrics from your Gold data, a knowledge graph that connects Gold and Silver data with relationships and reasoning, and a vector index that extends your reach to even raw Bronze data.
But to build this new story, your architecture needs to use a standardized ontology through all your data layers. If a term is ambiguous, an agent doesn’t throw an error, it just guesses and hallucinates a confident, wrong answer. A basic schema may record where data sits and what type it is, but it rarely carries enough business meaning to reconcile ambiguities the way a person would. As a result, a semantic layer has become widely recognized as essential AI infrastructure.
And while the interface has changed, your governance burden hasn’t. For decades, your data team controlled the experience by deciding what a dashboard exposed—which filters, which cuts, which definitions. That control was a quiet form of governance. The moment a user can simply type “How is Jane Smith’s book performing?” the guardrails are gone.
Now you govern which concepts and metrics an agent is permitted to invoke at runtime, a shift from build-time to query-time control. And consumption stops being a single query. An agent traverses a knowledge graph, calls a semantic view, retrieves a document, and composes an answer. The serving layer is no longer a technology. It’s a coordinated set of governed surfaces.
Why do AI agents need a shared ontology and semantic layer?
A shared ontology also helps solve the current issue of fragmentation. Traditionally, Gold was a set of departmental data marts.
Finance has a mart with specific definitions.
The commercial team has a separate mart, with overlapping (but subtly different) definitions.
Actuarial has its own data mart, bound to its own assumptions.
Human analysts learn those differences over time and reconcile them through experience and context. If leadership asks a cross-functional question, someone has to manually reconcile the differences. That’s not ideal, but it’s tolerable because each analyst only ever queries their own mart.
But AI agents don’t have that context. They can’t walk down the hall and ask a veteran analyst for institutional knowledge. When an AI system adopts the wrong definition, it scales that mistake across workflows, recommendations, answers, and critical operational decisions. And it does that confidently, at machine speed.
That’s why you need a shared ontology. To build trust in AI and reduce decision risk, your various marts converge around your Gold Layer: governed fact tables that are consistent across business contexts, not competing.
Does AI replace the Gold Layer?
No, AI doesn’t replace the Gold Layer. But it becomes an execution artifact generated from semantic definitions and ontology, rather than a hand-built construct. Data and processes still need to be certified, documented, materialized, and defensible to an auditor. Those jobs don’t disappear just because the consumer changed. If anything, they matter more, because an agent’s answer travels further and carries more weight than a single analyst’s dashboard.
What does change is the way Gold is built. It’s no longer separately maintained, bespoke departmental schemas with inconsistent definitions and weeks of engineering to surface every new metric. Instead, Gold becomes a derived layer, increasingly generated from the semantic layer’s definitions and the ontology’s concepts.
For example, when the semantic layer defines “in-force face amount” as “the sum of face amount where policy status is in-force” and the ontology defines what “in-force” means, the materialized Gold object can be generated from that same definition. That’s the endgame worth steering toward. Semantic definitions become a form of control, Gold becomes a layer for execution, and shared ontology governs both.
Gold becomes the certified machinery that makes an agent’s answers fast, governed, and auditable. The work it does grows in importance. Its visibility drops to zero, which, for a foundation, is exactly the point.
How AI changes the Gold Layer
How should organizations modernize the Gold Layer for AI?
The organizations pulling ahead in AI aren’t the ones with the most data or the flashiest copilots. They’re the ones that decided, deliberately, what their data means—once—and made every layer above Gold reference those shared definitions. If your agents are guessing, the fix rarely comes from a better model or a sharper prompt. It comes from an ontology, a semantic layer, and a Gold Layer that finally agree with each other.
Here are four questions worth asking before your next AI initiative goes to production.
Get that foundation right, and AI stops being a demo that stalls and starts being a system you can put your name on.
How can your organization unlock your data’s potential?
{{filterContent.facetedTitle}}
{{contentList.loadingText}}
{{contentList.loadingText}}
Follow us
