Vernon Yai brings a sharp focus to the intersection of data governance and the burgeoning world of autonomous agents. As enterprises move beyond simple retrieval-augmented generation (RAG) models, the need for a business brain—or ontology—becomes paramount to ensure AI doesn’t just guess but actually understands. In this conversation, we explore how platforms are attempting to solve the trust deficit in AI by creating a shared, authoritative understanding of organizational data, and why the race to become the “system of record” for agents is the next great frontier for the modern enterprise.
How does the implementation of a ranking system inspired by PageRank transform the way an organization identifies and trusts its most authoritative business definitions?
The genius of using an algorithm like PageRank in a data environment is that it stops treating all information as equal. In many organizations, a single business term might appear in hundreds of different dashboards, queries, and documents, making it nearly impossible for an AI to know which one is the “truth.” By using a ranking system, the platform can automatically extract context and organize it into a living graph that weighs several critical factors. It looks at who created the information, how widely it is used across the company, and how recently the data was updated before deciding which definition the AI agent should rely on. This move from a flat list of data to a weighted hierarchy ensures that when an agent answers a question, it is leaning on the most authoritative and certified assets available. It creates a sensory layer for the business where the most “trusted” paths are illuminated, much like how the early web identified the most relevant websites based on their connections and reputation.
Traditional RAG and vector searches often struggle with nuanced business context; how does a unified context layer specifically address the trust deficit currently preventing widespread AI adoption?
The single biggest barrier to enterprise AI adoption is that decision-makers simply do not trust AI outputs enough to act on them without a manual human check. Older approaches, such as standard RAG and vector search, are essentially just pulling back whatever text looks mathematically similar to your question, but they don’t actually understand the business logic behind the words. A unified context layer, like an ontology, provides the specific meaning that a standard catalog cannot, allowing one definition to feed every agent so you stop getting three different answers to the same question. By grounding answers in governed definitions that have a clear lineage back to the source, we can finally attack that trust deficit head-on. It turns the AI from a sophisticated search engine into a reliable advisor that speaks the actual language of the company, which is the only way executives will feel comfortable letting these systems operate autonomously.
Analysts have warned that an ontology cannot fix a “messy” foundation; what are the hidden risks for a CIO who tries to layer this technology over existing, poor governance structures?
There is a very significant danger that implementing a sophisticated ontology layer without having your data house in order will simply speed up your existing mess. An ontology is not a magical cure for poor lineage, weak data ownership, or fragmented permissions; in fact, it can amplify those issues by making incorrect definitions more accessible to autonomous agents. Most enterprises currently lack the data and governance readiness required to implement these layers effectively, which can lead to agents pulling incomplete data or applying the wrong logic to a workflow. The hard part for leadership isn’t just creating the ontology once, but maintaining its accuracy as the business changes and evolves over time. Without clear metric ownership and a robust process to resolve conflicting definitions, the project risks becoming just another stale metadata project with a more sophisticated name that ultimately fails to deliver value.
With several major vendors pursuing similar context-layer offerings, how should an enterprise decide where to build its semantic foundation to avoid confusion and platform lock-in?
The current market is becoming increasingly crowded with offerings like Horizon and the IQ family, which can lead to a great deal of confusion as every vendor slaps a different name on essentially the same idea. My primary advice is to follow the “data gravity” mantrmost enterprises will find the most success choosing the platform where their data already resides. If your data lives in a specific lakehouse environment, that environment’s native ontology is likely your most efficient path forward. However, CIOs must look beyond just immediate functionality and assess how open and portable these definitions are, especially in multi-platform environments. You want to ensure that your business definitions can move across different analytics tools and AI platforms without getting trapped in a “semantic lock-in.” Snowflake, for example, is focusing on open semantic interoperability to differentiate itself, which is a key consideration for companies that want to remain flexible as their data stacks evolve.
We are seeing a shift toward data platforms becoming an “agentic system of record.” What does this evolution mean for the future of how enterprises govern and deploy AI agents?
We are witnessing a shift that mirrors the historical transition when ERP systems became the system of record for transactions and data warehouses became the system for analytics. The next battle is over which platform becomes the authoritative source that AI agents read from, reason over, and act through. This “control-plane” strategy aims to create a single place where enterprise data, governance, business semantics, and agent execution all converge. When a platform manages everything from the raw data to the agent’s logic, it can ensure that the AI operates safely and at scale while respecting all existing permissions and lineage. This evolution suggests that data platforms are moving from being passive storage units to becoming active, “agentic” systems that don’t just hold the truth but also enforce how that truth is used in real-world actions. It’s a complete reimagining of the data stack as the actual brain of the digital enterprise.
What is your forecast for the evolution of verifiable execution in enterprise AI?
I believe the next major battleground is not just providing context, but ensuring verifiable execution. Over the next two years, we will see a shift in focus from “how smart the agent is” to “how accurately the agent performs the task.” While ontologies help the agent understand the business, they don’t yet guarantee that the agent won’t skip rows, apply the wrong math, or misunderstand a complex manual workflow during execution. The industry will move toward systems where every action taken by an AI can be verified and audited against the organization’s governed logic in real-time. We will see the rise of “verification layers” that act as a final check, ensuring that the agent’s output isn’t just a plausible-sounding answer but a perfectly executed business transaction. The winners in this space will be those who can prove their AI agents are consistently accurate and fully accountable for the actions they take.
