Three years ago I attended the Databricks Data + AI Summit for the first time. I remember walking the floor and hearing the same set of problems circling through every conversation: too many data silos, inconsistent governance, data engineers and data scientists working from different versions of the truth. The answer Databricks was building toward was elegant: a Lakehouse architecture that unified structured and unstructured data in one place, governed by Unity Catalog, so that every team worked from the same trusted foundation. At the time, it felt genuinely transformative.
Landing in San Francisco for DAIS 2026, what struck me first was the scale. The event has grown into something that mirrors the industry itself: larger, faster, and harder to take in all at once. More than 30,000 attendees, dozens of simultaneous sessions, and a keynote hall that felt closer to a stadium than a conference room. The pace of change in data and AI over those three years has been extraordinary, and Databricks has matched it almost announcement for announcement.
But what struck me more than the size was the shift in the conversation. Three years ago, the primary consumer of your data was a person: a data analyst running a query, a data scientist training a model. The challenge was making that data clean, accessible, and governed. Today, the primary consumer of your data is increasingly an AI agent. And agents have very different needs than humans do.
This is the problem that dominated DAIS 2026, and it is the one I came looking for answers on.
AGI Value Through Context, Control, and Choice
During the keynote, Databricks CEO Ali Ghodsi posed a question to the audience: “Do you think AGI is here?” He expected more hands. Instead, he got a cautious, half-hearted ripple across thousands of seats.
His response was to demonstrate just how capable today’s models have become: the things they can reason through, generate, and explain. His implicit argument: AGI is already here. But I think the audience was answering a different question. Not are models smart? but are they useful, in my business, on my data, for my problems? And to that, even Ali cannot yet give a confident yes.
This is the real gap. Models have become extraordinarily capable on general knowledge. But enterprise value requires something more specific: models that understand your data, your terminology, and your business context. A model that doesn’t know what “revenue” means under your company’s particular definition, or which product lines count as core versus tail, will give confident, fluent, and wrong answers.
“We don’t need more intelligence. We need more context.”
Ali Ghodsi, CEO, Databricks, DAIS 2026 Keynote
That framing is the lens through which every major announcement at DAIS 2026 should be read. Databricks organized their entire product strategy around three principles that follow from it:
- Context: connecting AI to the meaning of your data, not just the data itself
- Control: governing AI agents and their costs with the same rigor you apply to data
- Choice: avoiding lock-in across models, clouds, and data sources
These aren’t marketing pillars. They represent three genuinely hard problems that every enterprise deploying AI at scale will have to solve. Here is how Databricks is approaching each.
The Enterprise Context Layer: Genie Ontology
Three years ago, the breakthrough was the Lakehouse: a single, governed data layer for everyone. Clean data, unified governance, one source of truth. It was the right answer for the problem of the time.
But clean data is not the same as understood data. An AI agent querying your Lakehouse today can retrieve rows and run calculations. What it cannot do, reliably, is understand that “net revenue” in your finance team’s model excludes returns processed after day 30, or that “active customer” carries three different definitions across three business units.
This is the problem the enterprise context layer is designed to solve. Databricks’ answer is Genie Ontology, a semantic and knowledge layer that sits above your curated Lakehouse data and your knowledge base. It captures business definitions, the relationships between concepts, and institutional knowledge, combining them into a context that agents can draw on when answering questions. The result is not just more accurate answers, it is also faster answers, because agents spend less time searching for context on the fly, and more cost-effective answers, because fewer tokens are consumed in the process.
If the Lakehouse was the right foundation for human data consumers, Genie Ontology is the right foundation for agent consumers. Together, the two form what I believe will become the standard enterprise data and AI architecture within the next two to three years.
Purpose-Built Agents: The Genie Family
Databricks didn’t just announce a platform. They announced a suite of purpose-built agents designed for different personas inside the enterprise, a move that signals a clear belief: one-size-fits-all agents don’t work at enterprise scale.
The agents introduced at DAIS 2026 include:
- Genie One: a single coworker agent for both technical and non-technical users. The goal is a unified entry point, whether you’re writing SQL or asking a natural-language question about last quarter’s performance.
- Genie Agents: purpose-built for business users. These are the agents a sales director or supply chain VP would actually use, designed around business tasks, not technical workflows.
- Genie Code: automation for data engineers and ML practitioners. Rather than replacing the engineer, Genie Code automates the repetitive work of building and maintaining data and ML pipelines, freeing technical teams for higher-value problems.
- App Builder: enabling teams to build data-backed applications without deep engineering investment.
- Genie Zero Ops: perhaps the most practically compelling announcement for enterprise operations teams. An AI agent that detects, assesses, remediates, and verifies pipeline issues automatically. The goal is to reduce the maintenance burden enough that engineering capacity shifts from keeping the lights on to building new value.
I had the opportunity to put this to a personal test. During a partner training session, I built a Genie-backed Databricks application from scratch, with no prior setup. It took 90 minutes. For someone who has spent years watching clients wrestle with weeks-long development cycles for comparable tools, that is not a minor data point. It is a signal about where the productivity floor is headed.
Control and Security: AI Unity Gateway
One theme ran as an undercurrent through nearly every session I attended: democratization has arrived, and it brings opportunity and risk in equal measure.
The same ease that allowed me to build an agent application in 90 minutes means that analysts, operations managers, and product teams across your organization are doing the same thing, or will be soon. This is genuinely exciting for enterprise innovation. It is also a governance and cost problem that most organizations are not yet equipped to handle.
Three years ago, Unity Catalog solved a version of this problem for data: one place to govern all your data assets, enforce access policies, track lineage, and control costs. AI Unity Gateway is the equivalent for your AI estate.
Think of it as a control tower for every AI model, agent, and application running across your organization, regardless of which team built it, which model it uses, or which cloud it runs on. It gives you visibility into what is running, what it costs, and whether it is behaving within policy. It lets you enforce the same governance standards on AI that you have spent years building for data.
As the number of agent applications inside any enterprise grows from dozens to hundreds, this kind of centralized control will shift from a nice-to-have to a non-negotiable.
Join Us and Databricks on This Journey
Databricks has always moved fast. What was different at DAIS 2026 was the clarity of direction. Every announcement, Genie Ontology, the agent suite, AI Unity Gateway, and the continued investment in Lakebase, maps back to the same thesis: enterprises will not extract value from AI by deploying smarter models. They will extract it by making their data and their organization ready for agents to work on.
The companies that win in the next three years will be the ones that solve the context problem, govern their AI estate before it governs them, and avoid the lock-in that comes from building on a single vendor’s closed stack.
At DS Stream, this is exactly where we are focused. As a Databricks partner, we work with organizations to turn these capabilities into measurable business outcomes: not proof-of-concepts, but production deployments that drive real value. Our team includes some of the most certified Databricks practitioners in the market, with experience delivering projects for Fortune 500 clients across retail, CPG, and beyond. We are currently rolling out our first Databricks-native solutions, built specifically for the CPG and Retail industries, designed around the context, control, and choice framework that DAIS 2026 confirmed is the right architecture.
If you’re thinking about where your organization stands on this journey, whether you’re building a context layer, getting ahead of agent governance, or simply trying to make sense of what all of this means for your roadmap, we’d be glad to have that conversation. Reach out to our team at DS Stream. Let’s figure out your next step together.
.png)
.webp)

.webp)
