I recently celebrated my one-year anniversary as CEO of Stardog. Anniversaries are useful opportunities to reflect—not just on what has changed within a company, but on what you have learned about the market.
They’re also a reminder of who made the year possible.
I’m grateful to the Stardog team, who welcomed me and spent years building technology the market is only now catching up to. To our customers, who trust us with some of their most important data and hardest problems. And to our partners and investors, who share our conviction about where this market is headed.
I joined Stardog because I believed the market was moving rapidly toward a problem the company had spent more than a decade solving.
Stardog built an enterprise platform around knowledge graphs, ontologies, semantics, and reasoning, technology designed to give both machines and humans a richer understanding of data and, more importantly, the relationships and meaning within that data.
My thesis was: AI was going to make context exponentially more important.
What is an enterprise context layer for AI
Large language models are remarkably good at understanding language and generating responses. But an AI model does not inherently understand the unique language of an enterprise: its customers, products, suppliers, assets, policies, definitions, metrics, relationships, processes, and business rules.
That context has to come from somewhere. That is the role of an enterprise context layer. An enterprise context layer gives AI systems and agents a shared understanding of business data, definitions, relationships, rules, and meaning across the organization.
An ontology-driven semantic layer can serve as that enterprise context layer. It creates a machine-understandable representation of the business, connecting information across disparate systems and explicitly defining the relationships and rules necessary to answer complex questions.
Without sufficient context, AI is forced to infer meaning from incomplete information, increasing the potential for incomplete, inconsistent, or fabricated answers.
Observations from my first year on enterprise context layers for AI
After one year, I underestimated two things. First, how quickly the market would move toward this architecture. Second, how many vendors would suddenly claim to provide “context.”
I recently attended a major data industry conference where it seemed that almost every vendor had the word context prominently displayed somewhere in its booth. That is actually a very positive development. It demonstrates how quickly the market has recognized the problem. But it also creates confusion.
During the past year, the conversation has progressed well beyond knowledge graphs and traditional semantic layers. The industry is increasingly talking about enterprise context layers: infrastructure that provides AI systems and agents with a governed understanding of enterprise data, business meaning, relationships, rules, provenance, and operational state. This is no longer a fringe architectural concept.
Gartner has described the universal semantic layer as a strategic foundation for scaling AI safely and effectively. It has also stated that context layers should become a core component of data and analytics infrastructure and predicts that organizations prioritizing semantics in AI-ready data can significantly improve agent accuracy while reducing costs.
The major data platforms are moving in the same direction. Databricks now describes Genie Ontology as its “unified context layer.” Microsoft Fabric is integrating ontology directly with AI agents, giving those agents governed access to business entities, relationships, definitions, rules, metrics, and source mappings.
The market debate is no longer about whether AI needs context. Increasingly, the question is how enterprises should architect that context, and who should control it.
That leads to several observations from my first year.
1. Build your context layer with the end state in mind
If you are not considering the long-term implications of your context architecture today, you may be creating unnecessary cost and rework tomorrow. Enterprise context should not be recreated independently for every AI agent or application.
Imagine defining customer, revenue, product, supplier, risk, order, or thousands of other business concepts separately for every agent your organization develops. Now imagine maintaining those definitions as the business changes. That doesn’t scale.
The value of an ontology increases when business knowledge is defined, governed, and reused across applications and agents.
Organizations should therefore think beyond the first AI use case. The objective should be a reusable semantic control plane that can ultimately support the enterprise’s broader AI strategy.
2. Your data landscape will continue to change
Another lesson has come from conversations with large enterprises. A customer will occasionally tell me that “all of our data is in Databricks,” Snowflake, or another strategic platform. A few questions later, “all” usually becomes “most.”
That isn’t a criticism of those platforms. It is simply the reality of the modern enterprise.
Data lives in operational databases, warehouses, lakehouses, SaaS applications, legacy systems, documents, external data sources, and increasingly real-time systems. Companies acquire other companies. New applications get deployed. Architectures change. Today’s strategic data platform may coexist with tomorrow’s.
Enterprise data is inherently heterogeneous and dynamic. That creates an important architectural principle: Your enterprise context layer should not be constrained by the location of your data.
A semantic control plane should be able to connect information across heterogeneous environments and support both virtualization—leaving data where it resides—and materialization when performance, operational, or analytical requirements make that appropriate. The context layer should remain stable even as the underlying data landscape evolves.
3. The proliferation of AI agents makes shared semantics more important
The rise of agentic AI strengthens this argument. We are moving toward an environment in which enterprises may eventually operate hundreds or thousands of specialized AI agents.
Those agents cannot each develop their own interpretation of the enterprise. If one agent defines a customer differently from another, calculates revenue differently from finance, or interprets supplier risk differently from procurement, organizations will create a new generation of semantic silos, only this time operating at machine speed.
The enterprise therefore needs a common source of meaning. Agents can be distributed. Enterprise meaning should not be.
A governed semantic control plane can give multiple models, applications, analytics systems, and agents access to consistent business meaning while allowing those systems themselves to remain heterogeneous.
4. Context is becoming an enterprise asset
One of my biggest realizations over the past year is that an enterprise ontology is much more than technical metadata. It can encode decades of accumulated institutional knowledge:
- How does this company define its customers?
- How are products related?
- How does its supply chain operate?
- How does it calculate risk?
- What constitutes revenue?
- What policies govern particular decisions?
- Which relationships matter when diagnosing a problem?
That knowledge is valuable intellectual property. As AI becomes increasingly embedded in business operations, the ontology and semantic models representing that knowledge may become one of an enterprise’s most strategically important digital assets.
This leads directly to another issue.
5. Architectural sovereignty matters
I strongly believe enterprises should maintain control not only of their data but also of the business meaning encoded around that data. I call this Architectural Sovereignty.
Data sovereignty asks: Who controls my data?
Architectural sovereignty goes further: Who controls my business definitions, relationships, rules, ontology, workflows, and the context my AI systems depend upon?
This distinction becomes increasingly important as major platform providers embed semantic and ontology capabilities within their proprietary ecosystems. There is nothing inherently wrong with using those capabilities. They may be exactly right for certain use cases.
But enterprises should understand the architectural decision they are making. If years of business logic, semantic modeling, and institutional knowledge become deeply encoded within a proprietary platform, migrating away from that platform may become considerably more difficult—even if the enterprise technically still owns all of its underlying data.
Open standards provide an important alternative. They allow enterprises to separate the intellectual property represented by their ontology from the technology platform currently implementing it. That creates portability, optionality, and leverage.
6. Independence becomes more valuable as the AI ecosystem evolves
The enterprise AI stack is evolving extraordinarily quickly. Models will change. Agent frameworks will change. Databases will change. Lakehouses will change. Applications will change. Some of today’s dominant technologies will be replaced by technologies that have not yet been invented.
Your enterprise meaning should not have to change with them. This is where I believe an independent enterprise context layer becomes strategically important.
Rather than making the semantic layer subordinate to a particular database, lakehouse, cloud, LLM, or agent framework, the context layer can sit across them.
One enterprise ontology. Multiple data platforms. Multiple models. Multiple applications. Multiple agents.
That architecture allows the layers above and below the semantic control plane to evolve without requiring the enterprise to continually recreate its understanding of itself.
The question I would ask every CIO and Chief Data and AI Officer
After a year of conversations with enterprises building serious AI strategies, I think the fundamental architectural question is becoming increasingly clear:
Does the enterprise want its business meaning and ontology embedded inside a proprietary AI or data platform, or does it want an independent semantic and context layer capable of serving every data platform, model, application, and agent?
Stardog’s position on an open, independent context layer for AI
At Stardog, our position is clear. We believe enterprise meaning should belong to the enterprise. We believe ontologies should be reusable across applications and agents.
We believe the semantic control plane should span heterogeneous data rather than require all enterprise data to reside in one platform. We believe open standards matter because the business knowledge encoded in an ontology is too valuable to become unnecessarily dependent on a single vendor.
And we believe the explosion of AI agents will make these principles increasingly important.
A year ago, my thesis was that AI would cause the market to recognize the importance of semantic infrastructure. That happened faster than I expected.
The next phase of the market will be more interesting. The question is no longer simply whether enterprises need context for AI.
The question is who will own it.
