In this episode of the AtScale Data-Driven Podcast, Dave Mariani speaks with Juan Sequeda, Principal Researcher at ServiceNow, about why enterprise AI and analytics are reaching a breaking point without semantic context. The discussion explores how metadata and semantics are shifting from documentation to operational infrastructure, and why large language models and AI agents cannot deliver trusted, explainable analytics without governed business definitions.
Key takeaway:
AI systems are only as reliable as the semantic context they operate on. Without governed metadata and consistent definitions, AI-driven analytics will drift, hallucinate, and fail to earn enterprise trust.
Meet our Guests
Juan Sequeda
Principal Researcher at ServiceNow
Juan Sequeda is the Principal Scientist at data.world, having founded Capsenta, a company acquired by data.world, based on his PhD research at The University of Texas at Austin. His expertise is creating knowledge from complex data, particularly designing Knowledge Graphs for enterprise data management. He has received numerous awards for his research, including the NSF Graduate Research Fellowship, and recognition at conferences like ISWC. Juan is actively involved in academia and industry, serving on editorial and advisory boards, organizing conferences, and contributing to standards development. He balances his scientific interests with business strategy and customer engagement to bridge the gap between research and industry needs. He recently released his Benchmark to Understand the Role of Knowledge Graphs on Large Language Model's Accuracy for Question Answering on Enterprise SQL Databases.
Dave Mariani
Chief Technology Officer, Founder, AtScale
Dave is the founder of AtScale and is the Chief Technology Officer. Prior to AtScale, he ran engineering and data at Klout and Yahoo! where he built the world’s largest multi-dimensional cube.
Transcript
Dave Mariani:
Hi everyone, welcome to another episode of the AtScale Data-Driven Podcast. Today my special guest is Juan Sequeda. Juan has been on the podcast before, but he has a new title now. He is a Principal Researcher at ServiceNow. Juan was previously the Head of the AI Lab at data.world. He holds a PhD in Computer Science from the University of Texas at Austin. He is a co-author of Designing and Building Enterprise Knowledge Graphs, and he is the co-host of his own podcast, Catalog and Cocktails. Juan, good to see you again. Welcome back to the podcast.
Juan Sequeda:
Thanks Dave. Glad to be back here and glad to have this discussion with you. Super excited.
Dave Mariani:
We are going to talk about some of the things that I predicted are going to happen in 2026. It has been a crazy time. Lots of changes. AI has eaten the world. There have been a lot of acquisitions in our space, including your company, data.world. For those listeners who have not had the opportunity to experience an acquisition, what has that been like for you?
Juan Sequeda:
I am thrilled, excited, and honored to be part of this acquisition and really excited to be part of the data.world team and everything that has been accomplished, and also joining the ServiceNow team. Personally, I joined data.world in 2019 when data.world acquired my company, Capsenta, which I started from my PhD work where we were doing early semantic virtualization and early knowledge graph work. It has been really cool to see how my PhD work went into data.world and now into ServiceNow.
What is interesting is what has happened in the market over the last year. There has been a lot of consolidation. Five years ago, the pendulum swung from large monolithic systems to the modern data stack. Then the pendulum swung too far, and every feature became its own category and its own product. At the end of the day, data is a means to an end for customers. They did not want to spend so much time and money going through procurement cycles and stitching everything together.
The pendulum started to swing back. Metadata has traditionally been something on the side, but it has grown in importance. Every feature became its own thing, but all of it is important. Large platforms are realizing that metadata cannot be a separate thing. Metadata is at the center. Metadata brings knowledge, semantics, and meaning together. Because AI is everywhere, AI needs context and metadata.
We are seeing many acquisitions in the metadata space. Some players think there will be an independent metadata or context layer, but we are also seeing convergence of platforms. Systems of record, systems of analytics, and systems of action are converging. ServiceNow is a system of action. Because AI needs context, metadata is becoming a first-class citizen.
Dave Mariani:
Metadata has often been treated as documentation, something nice to have.
Juan Sequeda:
Yes, but metadata is operational. It provides context for how the business works. You need data, but you also need knowledge. Together, that is what builds the enterprise brain. We have lived in a data-first world and treated metadata and semantics as second-class citizens. AI is forcing us to recognize that this knowledge is the foundation.
Dave Mariani:
You published research showing that natural language interfaces and AI systems cannot work reliably without context. We replicated that research using TPC-DS and found the same thing. LLMs have a really hard time navigating complex schemas, and you just cannot trust them to do that reliably because they are non-deterministic in nature. You cannot allow definitions like gross margin to change from query to query.
Juan Sequeda:
One hundred percent.
Dave Mariani:
That recognition has driven big platform moves. Snowflake announced a semantic layer. Databricks announced a semantic layer. Google repositioned Looker as a semantic layer. Microsoft positioned Power BI as a semantic layer. Everyone is now in the semantic layer space because context is king.
Juan Sequeda:
Without context, AI does not understand the business. Governance becomes critical. You need guardrails so answers are trustworthy, explainable, and accountable.
Dave Mariani:
When I connected Claude, ChatGPT, and Gemini to our MCP server, it changed how I interacted with analytics. Instead of asking, “Show me gross margin by product,” I learned to ask, “Tell me something about gross margin.” The questions are non-deterministic, but the answers are deterministic because they are governed by the semantic layer.
What do you think this means for traditional dashboards?
Juan Sequeda:
Dashboards are not going to fully go away. You still need core dashboards, like a speedometer in a car. But we will reduce the number of dashboards. Right now, we have analytical swamps.
This is an opportunity to focus on what really matters. When you ask a question about gross margin, there is a reason. We should focus on the why. AI can help explore possibilities and generate insights. Then agents can help drive action. That is the real shift—from insights to action.
Dave Mariani:
That sounds like prescriptive analytics, which dashboards alone never delivered.
Juan Sequeda:
Exactly. We will have fewer dashboards, focus on real business problems, brainstorm solutions, and drive action. AI will do the mundane work so people can focus on outcomes.
Dave Mariani:
Where do you see tooling going? Will there be new interfaces, or consolidation around large platforms?
Juan Sequeda:
We are seeing convergence of systems of record, systems of analytics, and systems of action. Traditionally, these were separate. AI is bringing them together. Some platforms will try to bring everything inside. Others will emphasize optionality, letting data live where it is while centralizing context.
Context must be centralized because it needs to be governed. It should be defined once and reused everywhere.
Dave Mariani:
That aligns with what we see with MCP. Applications connect to multiple context sources rather than retraining models.
Juan Sequeda:
Governance is critical. Context is not just definitions. It is knowing which definition applies to which user, domain, or role. We will need automation, workflows, and agents to manage that.
We are also heading toward agent sprawl, similar to dashboard sprawl. We will need a control tower to manage data, analytics, and AI agents.
Dave Mariani:
This feels like the moment we have been working toward for a long time.
Juan Sequeda:
It is. This is persistence paying off.
Dave Mariani:
Before we close, what does this mean for hiring and tech jobs?
Juan Sequeda:
Three things matter. First, systems thinking. If your job is turning input A into output B, AI will automate it. The value is understanding how systems fit together.
Second, people skills. These are not soft skills. They are essential skills. You need to understand incentives and translate between levels.
Third, business understanding. You need to understand how companies make money, what drives industries, and what problems matter. Combine that with AI tools, and you become a superpower.
Dave Mariani:
I completely agree. Juan, thank you for joining me again.
Juan Sequeda:
Thank you very much.
Dave Mariani:
Thank you to everyone listening to another episode of the AtScale Data-Driven Podcast.
Juan Sequeda:
Cheers.
