Starburst Podcast: Bridging Data and Intelligence: Starburst’s AI-Driven Mission

Welcome, everyone. I'm Ron Powell, Executive Director of the Analyst Consultant Advisory Network, also known as ACAN. Today's podcast is a follow-up to our recent private forum with Starburst, and it gives you the opportunity to hear directly from one of the most forward-thinking leaders in our industry.

I'm thrilled to welcome Justin Borgman, CEO and co-founder of Starburst. Justin has spent over a decade at the forefront of data and analytics innovation. Before founding Starburst, he co-founded Hadapt, where he helped pioneer SQL on Hadoop, and he later led Hadoop product efforts at Teradata.

Now at Starburst, Justin is helping shape the next generation of data architecture, one that's AI-native, real-time, and designed to simplify access, governance, and collaboration across the modern data stack.

Welcome, Justin.

Thank you for having me. It's great to be here.

Ron: Justin, let's dive in. Can you start by giving us a high-level view of Starburst's mission and how the power of AI is being brought to life by your customers?

Justin: Absolutely. At our core, we are a data platform that provides fast access to all the data an enterprise might have, regardless of where it lives. Now we see customers wanting to store as much data as they can in data lakes, leveraging open formats like Iceberg. But there will always be data silos, and so being able to federate queries out to other data sources is part of what makes us unique.

And the combination really allows us to play this interesting role as an abstraction layer above all the data in the enterprise with fine-grained access controls and providing that access to the analyst. And historically, that analyst has been writing SQL queries or leveraging BI tools, but increasingly, that analyst may be an AI agent or an application that leverages AI and is accessing this data platform. And what's so exciting right now is that we're in this transition to the world of AI.

Ron: In our private forum, you mentioned that Starburst is moving from data systems to knowledge systems. What does that transformation mean in the context of AI, and how it changes the role that Starburst plays inside the enterprise?

Justin: First, I would clarify that Starburst is not another LLM company like OpenAI or Anthropic, but rather we're a data platform company. And what we think is critical about the role that we play is that we want to be the rocket fuel for the AI projects that you're trying to complete.

And what we mean by that is the provider of data. Your AI is only as good as the data that it has access to, and we have access essentially all of it within the enterprise. And so, with that background, I think what we see is that our enterprise customers are moving from workflows which were about asking specific queries of the data to now having a more conversational interaction and automating entire decision processes, deploying logic to complete a task.

So, what we're trying to do is not only introduce natural language as an interface but also allow our access to bring context to bear to the challenging questions that people may have. So, this is part of a larger arc as traditional systems of record become systems of intelligence or knowledge systems in our view. And in this next phase, AI doesn't just predict, it automates.

And the previous model of a more human-led data consumption becomes a more machine-led data consumption by AI agents. And so that's the role that we can play. That's what we're excited about is really being the provider of that rocket fuel, if you will, to these new AI-driven workflows.

And natural language has really come a long way. Now there's a real push right now to connect structured enterprise data with LLMs.

Ron: How is Starburst becoming the bridge between raw data and AI-driven insights?

Justin: AI is only as good as the data that fuels it. And what we're providing is access to that data, both in the cloud as well as on-prem and in hybrid scenarios. In fact, part of what we've announced is this idea of an air-gapped agent as one extreme example of what an enterprise can accomplish with our technology to manage data sovereignty, data privacy, and various regulatory constraints that might dictate keeping all their data within an air-gapped environment. And we can operate within that with our deployment.

But more to the question about structured data and LLMs, we're providing access to both structured and unstructured. And that includes access to tabular data, both in traditional relational database systems as well as in Iceberg tables. And we do see a move towards more and more storage in these open formats like Iceberg within a data lake or lake house, as well as part of our announcement here is introducing a vector search capability.

And that now allows you to combine both structured and unstructured data and bring that unstructured context to enrich the structured tabular data that you already have access to. With Starburst AI, you're introducing agents, vector search, and AI workflows all built on top of Iceberg.

Ron: How does architecture help simplify RAG workflows and bring AI closer to the data enterprises already have in their lakes?

Justin: I would say the focus on Iceberg is important because it gives customers one uniform open format that they have access to.

And it really eliminates the vendor lock-in that has plagued the data platform, database, and data warehouse industry for a very long time. So many proprietary systems end up locking you in. And then, of course, the prices go up and so forth.

And so, we see this movement towards leveraging these open formats as much as possible and Iceberg becoming the dominant one. And so, the reason we focused on allowing you to do vector search and store these embeddings in Iceberg and build your entire workflow around that is that it simplifies your architecture. And it simplifies it on a low cost, attractive cost performance platform.

This allows you to do your RAG workflows entirely within that one architecture, which is going to simplify your data management as well as your access controls. And again, give you the most bang for your buck as you build that platform. So, we think that's key.

And I will say this is an important point. We were led here by our customers themselves. We saw some of our leading customers doing exactly this, where they chose Iceberg as that foundational layer for their open format and then started to build these RAG workflows on top of that.

And in working with them, we developed this technology to essentially do that vectorization, create those embeddings all within this open architecture, which we think over time will become the dominant architecture for how people build their AI applications.

Ron: Speaking of customers, one of the big moments in the ACAN private forum was hearing about Citibank's enterprise-wide adoption of Starburst and even their investment in the company. That's a strong vote of confidence from a highly regulated institution. What do you think convinced Citibank that Starburst is foundational to their data and AI strategy?

Justin: I think it is really a fact that we have access to all the enterprise's data, and that makes it a very strategic platform for any enterprise who chooses to deploy it. If you're a large, complex, multinational business like Citibank, with over 100 countries that they operate in, some very strict regulation by various authorities in multiple countries, having to create a system that can work enterprise-wide in the complexities of that environment is challenging. And the fact that we were able to do it is something that gave them enough confidence to not only go enterprise-wide but celebrate this achievement with them and celebrate that partnership.

Because as you likely know, it is not common for a bank to be willing to speak publicly about their use of a vendor's technology. So, I consider that a major milestone in and of itself, and it's a function of the incredibly strategic nature of this partnership in terms of being able to deliver fast access to data, regardless of where it lives, across all kinds of different silos and across many different use cases, everything from anti-money laundering to fraud detection to risk analytics to Customer 360. So many use cases that leverage data that spans multiple data sources that they're now able to access directly.

And then when you add the context of AI, and I would say every large financial services institution is thinking about their AI future, now data access becomes even more critical because once again, your agents will only be as useful as the data that they have access to. You're also enabling less technical users to work with AI, utilizing SQL. They don't have to know Python or other languages.

Ron: How do you see this democratizing access to AI and changing the role of the analyst across the enterprise?

Justin: We're trying to democratize it really in two ways. First, the fact that a SQL developer can now interact with data and build their applications is important, because SQL developers tend to outnumber machine learning engineers with Python expertise on a ratio of 10 to 1 or more. And being able to bring these capabilities to the SQL developer expands the audience dramatically within an enterprise and makes it a lot easier and faster to take advantage of AI in the applications that they're developing.

The second point that I would make is with the natural language interface that generative AI now helps provide, we're able to bring these capabilities directly to the fingertips of a non-technical user entirely. I'll give you a quick example. In a different bank, we're working again in an enterprise-wide capacity where the CEO of the bank, this is a very large publicly traded bank, can ask questions of that data himself via a chat interface.

It's essentially like chat GPT, but for the bank's data themselves. And behind the scenes, we're leveraging our data platform to get access to all the data that they need. Some data is vectorized, and we've created those embeddings that can be leveraged.

Other data is curated through our data products functionality, which allows you to add business metadata to the underlying raw data itself to enrich and provide more context. And again, this is now creating that interactive experience that many of us have when we use chat GPT on our phones, but now for your enterprise data. And I think that's the key, is really closing that gap between a chat interface on internet data and a chat interface on your enterprise data, which again is going to be private, highly governed, and not something that you want leaving the four walls of your enterprise.

Ron: With all the focus on data quality and AI observability, how do you see Starburst evolving next? Especially as more companies move toward iterative AI-driven development workflows?

Justin: I think governance is the central pillar of our AI strategy. It was one of the first enterprise features we built seven and a half years ago when we were first getting Starburst off the ground. It's one of the reasons many open source Trino users became Starburst customers in those early days, and it continues to be a pillar for us in this AI future that we're moving into.

That means applying the same fine-grained access controls—both role-based and attribute-based—including row-level and column-level security, data masking, and of course, query auditing. It also means being able to track which users accessed which data, what queries they ran, what results they saw, and having the ability to review all of this after the fact for compliance, forensic analysis, or other oversight purposes. It's a tightly governed system, and we think that is the key to creating a platform that can be used at scale within a serious enterprise, supporting that auditability as well as governance and the access controls because that allows you to also debug when things do go wrong, like hallucinations or the potential for IP leakage. Being able to trace the data's journey essentially from source to destination is what's going to simplify the impact analysis and the troubleshooting, and we think is table stakes for this type of platform.

Ron: We're hearing a lot about the need to prove ROI quickly, especially with AI investments. How is Starburst helping customers innovate while still showing clear business value and impact right out of the gate?

Justin: This is one of the reasons we built our own agent on top of the platform, is while we think of ourselves first and foremost as a data platform for others to build their own applications and agents, we felt it was necessary to include an agent out of the box, included with the platform. And the reason is it becomes a great way of demonstrating value almost immediately, particularly if you use it in conjunction with our data products functionality, which allows you to curate high-quality data for access by the agent.

The conjunction of those two together allows you to get value almost within minutes of using the platform. If you already have data products prepared, you just connect your agent to the data products and now you're able to run queries with great business context and get value out of answering those questions right away. That was important to us, providing something out of the box to really showcase the power of the platform and deliver value quickly.

And then from there, customers may say, you know what, we're going to build our own agent. It's going to be a special purpose for anti-money laundering. And we're going to go deep there where we have domain expertise and build an agent.

But now I've got at least a starting point. I've got essentially a template that will help me now build my own custom agent more quickly down the road. And over time, this idea of agent-to-agent communication is likely to become a real thing.

And, you know, now the agents that customers build can also talk to the agent that's already built inside of Starburst. And we'll start to see that reality down the road.

Ron: Justin, I want to thank you for such an insightful conversation.

It's clear that Starburst isn't just keeping up with the AI-driven evolution of data architecture, but you're helping to lead it. And from simplifying RAG workflows to making AI more accessible across the enterprise, it's exciting to see how you're bridging the gap between raw data and real intelligence. And to our listeners, thank you for joining us.

If you're part of the ACAN community, this is just one of many deep dives we offer into the strategy shaping the future of data and analytics. We hope today's discussion gives you new ideas and maybe even a few questions to bring into your own client conversations.

Until next time, I'm Ron Powell.

Thank you for listening.