Databricks wrapped its annual Data + AI Summit on June 18 in San Francisco — and the message from the company’s leadership was hard to miss: the era of AI agents as first-class platform citizens has arrived.

More than 30,000 data and AI professionals gathered at Moscone Center, with tens of thousands more joining virtually. The product announcements were substantial, touching everything from data pipelines to governance to how agents are built and deployed. Here’s what actually matters and why it changes things for data teams.

What Databricks Announced

Agent Bricks — Agents Become a Native Platform Layer

Agent Bricks, Databricks’ comprehensive agent development platform, got a major expansion at this year’s summit. Since its launch, more than 100,000 agents have been built on the platform. The expansion adds support for any model, governed memory systems, and sandboxed execution environments — meaning agents can now operate in production with the same governance controls applied to data assets.

The underlying design principle: Choice (any model, any format, any cloud), Context (governed business semantics), and Control (governance over what agents do, not just what they access).

Unity AI Gateway

A unified governance layer for AI assets across the platform — covering models, MCP connectors, tools, and agents both on Databricks and externally hosted. Unity Catalog, which already governs data, now extends to AI in a meaningful way. For enterprise teams worried about who’s accessing what through which agent, this is the answer Databricks is offering.

Genie One and Genie Agents

Genie Spaces — Databricks’ natural language query interface — has evolved into Genie Agents: autonomous, shareable agents capable of independent action. Genie Ontology, an auto-learned context graph, underpins this with governed semantic understanding, improving accuracy as the system learns the business meaning behind the data.

Spark Declarative Pipelines and Lakeflow

On the data engineering side, Delta Live Tables transitions into an open Apache Spark standard with AI-native authoring. Lakeflow is Databricks’ next generation of pipeline orchestration. Both moves push the platform toward a more open, interoperable architecture.

Iceberg v3 GA and OpenSharing

Databricks moved Iceberg v3 to general availability and announced that managed Delta tables now support external engine writes — deepening interoperability with non-Databricks systems. OpenSharing, a vendor-neutral framework for sharing AI assets without data copying, is moving to the Linux Foundation. That’s a notable signal about where Databricks sees the open data ecosystem heading.

MLflow 3 for GenAI

MLflow 3 adds tracing, evaluation, and observability capabilities purpose-built for agents — including off-platform deployments. For teams running agents in production, this is the monitoring infrastructure they’ve been waiting for.

Why This Matters for Data Teams

The consistent thread through all of these announcements is that Databricks is treating AI agents not as a feature inside the platform but as a new software layer on top of it. That’s a significant shift.

For years, data teams operated in a world where their job was to get clean, reliable data into the hands of analysts and business users. Dashboards, reports, SQL queries — the outputs were mostly static. What Databricks announced this week reflects a different future: data pipelines that feed agents, agents that take autonomous action on business logic, and governance layers that span both the data and the AI that acts on it.

Teams that have built strong data foundations — governed data, clean pipelines, solid semantic layers — are now positioned to deploy agents that are actually trustworthy. Teams that haven’t will find that “just adding AI” to a messy data environment produces unreliable agents that nobody trusts.

The governance question has moved to the front of the queue. Unity AI Gateway signals that governance for AI agents is no longer an afterthought. Enterprise teams are going to be asked by security and compliance to demonstrate what their agents have access to, what they can do, and who approved it. Platforms that can answer those questions with auditability will win the enterprise deployment cycle.

Agent Bricks at 100k agents is a real number. That kind of scale means the platform is no longer experimental. Teams that are still evaluating whether to pilot agents should now be asking why they haven’t deployed them yet.

What This Means for Business Leaders

If you’re a business leader and your data team is running on Databricks, ask them two questions after this summit:

1. What’s our governance posture for AI agents? (Unity AI Gateway is the Databricks answer — make sure your team has a clear plan.)
2. What’s the first business process we can automate with Agent Bricks now that the platform supports it at scale?

The summit announcements weren’t announcements for announcements’ sake. They reflect Databricks responding to real enterprise demand for agents that are governed, reliable, and auditable. That demand has arrived.

For data professionals specifically, MLflow 3 and the Genie Agent evolution are the things worth studying first. These are the production monitoring and end-user deployment layers that turn data infrastructure into business automation.

---

Want to build the data and AI skills your team needs to actually deploy this stuff in production? Enterprise DNA’s learning platform covers Power BI, Python, SQL, and AI — for teams that want to move from data to action.