I recently sat down with Juan Sequeda, Principal Researcher at ServiceNow and co-author of Designing and Building Enterprise Knowledge Graphs, to discuss what’s happening in enterprise data and AI. Listen to the full podcast episode here.
Juan brings a unique perspective, having lived through the acquisition of his company Capsenta by data.world, and most recently data.world’s acquisition by ServiceNow. These moves signal an important market shift: metadata and context are no longer optional infrastructure.
Metadata is Operational and Context is King
Five years ago, we watched the data stack fragment. The modern data stack promised flexibility, but it was incredibly complex. Every feature became its own category, its own product, its own procurement cycle. Teams spent more time stitching tools together than solving business problems.
Now the pendulum is swinging back. We are seeing consolidation across the metadata and semantic layer space because enterprises understand that data is a means to an end, not the end itself.
Rather than treating semantics and metadata as documentation, AI is forcing us to flip that paradigm. Data and knowledge together form what Juan calls the “enterprise brain.”
Our own research proved this. When we tested LLMs on complex schemas such as TPC-DS, they consistently failed without semantic context. Juan’s research at data.world showed the same thing: natural language interfaces and AI systems cannot work reliably without semantic guidance. You cannot trust LLMs to compute “gross margin” consistent or accurate if if has to traverse hundreds or thousands of database tables. That non-deterministic approach doesn’t work and undermines trust.
This recognition drove the biggest platform announcements of 2025. Snowflake and Databricks launched their own semantic layers. Google repositioned Looker as a semantic layer, and Microsoft positioned Power BI the same way.
Everyone is now in the semantic layer business because context is king. Without it, AI cannot understand business. Semantics becomes the mechanism that makes answers trustworthy, explainable, governed, and accountable.
How AI Is Changing Analytics, and Why Governance is Critical
Model Context Protocol (MCP) is driving a significant shift in the analytics space. When I connected Claude, ChatGPT, and Gemini to AtScale’s MCP server, I learned to ask more open-ended questions. Instead of saying “show me gross margin by product,” I learned to ask questions like “tell me why gross margin is fluctuating and where.”
When LLMs are connected to a semantic layer, AI can get creative (think non-deterministic) in generating its exploratory questions, but the answers to those questions are deterministic because they are answered by the semantic layer and its underlying knowledge graph.
Juan and I agreed that dashboards are not going away anytime soon, but we will build fewer of them going forward. Most enterprises are already drowning in analytical swamps with hundreds of dashboards that nobody uses. With AI, we have an opportunity to focus on what actually matters. With a semantic layer, AI can explore data independently and generate insights from the results. Headless agents can then drive action from those insights and make the leap from descriptive analytics to prescriptive outcomes.
Yet Juan raised an important warning: we are heading toward agent sprawl. Every team will build agents, and every workflow will have automation. Without governance, this will create chaos and risk repeating past mistakes.
Context infrastructure will serve as the control tower for managing data, analytics, and AI agents. It defines who can access what, which definitions apply to which domains, and how agents should behave within guardrails.
What This Means for Careers
What does it take to thrive in this environment? According to Juan, three things matter:
Systems thinking. If your job is turning input A into output B, AI will automate it. The value is understanding how systems fit together, where bottlenecks exist, and what second-order effects look like.
People skills. These are not soft skills. They are essential skills. You need to understand incentives, translate between technical and business teams, and navigate organizational dynamics.
Business understanding. You need to know how companies make money, what drives industries, and what problems actually matter. Combine that knowledge with AI tools, and you become a force multiplier.
Persistence Pays Off
Juan and I both agree that this moment validates those who have been building semantic infrastructure over the past decade. Those ideas and technology were always sound. What changed is that AI made the cost of not having governed semantics impossibly high.
Enterprises cannot afford to let definitions drift or retrain models every time business logic changes. The companies that built critical infrastructure early are ready. Those who treated it as an afterthought are scrambling.
AI needs context, and context needs governance.
That is what 2026 will be about.
SHARE
2026 State of the Semantic Layer