Ali Ghodsi’s Data Intelligence Playbook: Turning Data into Agentic Advantage

**Dave Vellante** (0:00)
Last week, we said Snowflake was attempting to cross the Rubicon, vaulting from cloud data warehouse innovator into AI leader and intelligent application platform. This week, the competitive fire got a fresh accelerant. As usual, Databricks followed up Snowflake Summit and put forth bold claims of its own, emphasizing unified governance, data intelligence, simplified user experiences, transactional capabilities, which we're going to talk about, and of course, open data. Coming off the data plus AI summit, Databricks is no longer pitching just faster lake houses. It's unveiling really a three-tier battle plan. At the foundation sits a mature data and ML ops platform, the company's heritage. Now above that, a nascent block that we refer to as system of intelligence, which is this living governed map of metrics and dimensions and causal levers. And perched on top is the system of agency, where autonomous workflows, they promise to convert insight into action without human middlemen getting in the way, but of course with humans in the loop to guide agents and teach them along the way. Now if this model takes hold, we believe the center of gravity at enterprise tech shifts from storing data to encoding meaning, and whoever controls that semantic digital twin will dictate how every downstream agent behaves. Of course, ambition doesn't close deals, so to seize the high ground in our view, Databricks must prove three things. One, that Genie, which is its conversational AI interface, and Databricks 1, a reimagined user experience, can capture user intent fast enough to enrich the semantic core. Second, that Unity catalog metrics can lure in the ecosystem to contribute rather than siphon out context.
And third, that Agentbricks, the new MCP interface, LakeBase, which is a Postgres-compatible transaction engine, together form the components of Databricks' intelligent platform that that can become the epicenter of data workflows that push decisions back into systems of record at production scale. So in our view, this is more than a feature shootout with Snowflake. It's a race to deliver what we call useful enterprise AGI, which is behind the corporate firewall, where CEOs like Jamie Dimon sit on a trove of proprietary data that matters more than internet-scale LLMs, which are facing commoditization. The stakes are defining who owns the next decade of enterprise value creation in software. Hello, and welcome to this week's theCUBE Research Insights, powered by ETR. In this Breaking Analysis, George Gilbert and I continue our theme of crossing the data rubicon. Our focus today is mapping what we heard at Databricks Data and AI Summit into our system of intelligence and system of agency framework that George has been working on, and I've been reporting on now for quite some time. Now, last week, we argued that Snowflake's aspirations to cross the rubicon were represented by pushing beyond warehouse economics, and George, we're going to get into this. So good to see you again. Thanks. I think this is three in a row.

**George Gilbert** (3:19)
It's becoming a habit.

**Dave Vellante** (3:21)
It's a good habit. Okay. So we're talking about pushing Snowflake last week, pushing beyond warehouse economics into transactions, eventually into application logic. And our assessment is that Databricks is racing toward a similar destination with a different strategy. Databricks is expanding its ambitions with an even more expansive vision that it calls data intelligence. And the slide you're seeing now frames that ambition. So at the base, you see Databricks' lake house, which we're going to talk about throughout, and transactional services. That's no longer the story. They're really kind of table stakes. What's new are the layers on top. Framed in this new software architecture that we've been laying out for the last several weeks and months that we've been envisioning. And we think that Snowflake Data... Like Data... Sorry, like Snowflake, Databricks is methodically climbing Jeffrey Moore's hierarchy, first enabling generative AI to really be the user experience, and then unleashing agents that can both ask, you can interact with, and act on behalf of humans. All that, of course, governed by the Databricks Unity catalog. Now, in terms of semantics, both technical and business, if Databricks can truly own those semantics, as Ali Ghodsi implied, it positions itself not just as a data platform, but as kind of an operating system for enterprise data and AI. So George, that's the aspiration anyway. You were there at Databricks Data Plus AI Summit. The open question is, how baked is this? How far along is each layer really? And where are the gaps? What's your thought?

**George Gilbert** (5:00)
Okay, so the key thing is what you mentioned. And Ali said at Analyst Day, owning semantics is existential. This is the data intelligence. That's because data programs or drives AI. And it's an enterprise's ability to refine data, to reflect how its business operates, that will drive each business's own AI, their analytics, their agents. And you have to, you can have the world's best analytic engines and frontier model agents, but they only deliver effective intelligence and autonomous action based on the quality of the map that each enterprise creates with its own data. So agents will learn how to augment their human supervisors from this map, and the frontier lab models may directly power or distill the models inside these agents. But turning that into unique economic value belongs to each enterprise, and it's the enterprise vendors like Databricks that aspire to help companies build this map. That's why the true value of enterprise AGI will belong to enterprises that can build a 4D map or digital twin of their operations, what we're calling the system of intelligence. That guides agents and their human supervisors, not only about what happened, which is metrics and dimensions, which we're going to talk about in detail, but why something happened, what's likely to happen, and then what should we do.