Unity Catalog is what made the lakehouse viable for the enterprise. Before it, customers could consolidate data and workloads, but they still struggled to answer basic questions: Who can see this? Where did it come from? Which policies apply? Can we prove any of that to an auditor? Without those answers, regulated companies were never going to run the business on a lakehouse.

Koantek leaned into Unity Catalog early. We were one of seven launch partners for the Brickbuilder Unity Catalog Accelerators, and we began running UC migrations before most customers understood how important the upgrade would become.

Data + AI Summit 2026 was another important milestone for Koantek. We were named Databricks’ 2026 Lakebase Partner of the Year, our seventh consecutive year of Databricks partner recognition. But the announcement I keep coming back to is Unity AI Gateway. I believe we are watching the start of the same kind of shift we saw with Unity Catalog. Today, that ungoverned surface includes agents, MCP services, coding assistants, model calls, and API keys consuming budgets one request at a time. We intend to lean in early again.

Unity Catalog Is Moving Into the Runtime‍

If you read the announcement as a better LLM gateway, you will underestimate it. Databricks is moving Unity Catalog into the runtime. The Unity Catalog announcement describes the catalog evolving from “a system of record” to “a runtime decision-maker for AI.” The Unity AI Gateway post says it now “enforces runtime controls across model calls, tool invocations, and agent workflows.” The Agent Bricks team was even more direct: “every customer should be using Unity AI Gateway.”

My interpretation is straightforward: Databricks is expanding its control plane. Unity Catalog governs which tables, columns, and rows a user can access. Unity AI Gateway extends that model to what an agent can do during an interaction—which model it can call, which tools it can use, which data it can touch, whether it can write to another system, and when a human must approve the action. This is governance of behavior at runtime.

What Databricks Announced

The announcements matter in four practical areas.

Cost and budget controls. Customers can now attribute spend by user, team, model, and use case, then stop requests when a budget is exhausted. Ghodsi called cost “the #1 question we get,” and Axios reported that Databricks built these controls after customers ran up tens of millions of dollars in AI bills in a single month.

AI assets in Unity Catalog. Models, MCP services, agents, and skills are now catalog objects with access controls, lineage, and auditability. Databricks also announced managed connectors for Google Drive, Slack, GitHub, Jira, Confluence, and SharePoint, with governance built into the connection.

Contextual Service Policies. Administrators can allow, deny, or require human approval based on the user, the agent, the requested action, and the context. That matters when an agent is pushing code, writing to an operational system, or working with sensitive data. It gives the CISO a practical way to limit blast radius without blocking every agent use case.

Governed traces and security monitoring. Model and tool traces land in governed tables and feed Databricks security monitoring through Lakewatch, now backed by the Panther Labs acquisition. Coding assistants such as Cursor and Claude Code can be brought under the same control plane.

Why I Expect Adoption to Move Faster

Databricks reported $6.9 billion in annualized revenue, more than 80% year-over-year growth, $1.7 billion from AI products, more than 100,000 agents on Agent Bricks, and more than 14,000 organizations on Unity Catalog (CNBC). That last number matters. Those organizations already have the governance foundation Unity AI Gateway builds on.

Most enterprises running AI in production already have some combination of:

• Provider API keys managed team by team and sometimes left in notebooks
• Multiple model providers with no consolidated view of cost
• Coding assistants adopted one developer at a time, often without a formal security review
• AI spend that cannot be tied cleanly to an owner, budget, or business outcome
• No clear answer when the CISO asks what an agent can actually do

This is familiar. It looks a lot like enterprise data governance before Unity Catalog, except the risk is broader and the costs can grow much faster.

The pressure is also coming from different parts of the organization. Security leaders are worried about data loss and agent permissions. Finance teams are seeing open-ended consumption. CIOs are dealing with unmanaged tools and delivery risk. The consequences are operational: a bad data permission can expose a table; a bad agent policy can push code, write to a production system, or consume a quarter’s budget over a weekend. Security vendors including CrowdStrike, Okta, Palo Alto, Zscaler, and SailPoint are already making this part of the enterprise security conversation. Unity Catalog did not have this much organizational pressure behind it in 2021.

What Customers Need to Do Now

Most customers are starting with some degree of AI sprawl, so this is more than enabling a feature. They need to inventory model endpoints, API keys, agents, MCP servers, tools, and coding assistants; route the right traffic through the gateway; register AI assets in Unity Catalog; set budgets; define policies for sensitive actions; and capture traces in governed tables. Then they need to test whether the controls actually stop the scenarios the security team cares about. The harder work is deciding who owns the controls and how spend, policy exceptions, and incidents will be reviewed. That operating model is what makes the technology sustainable.

What Koantek Plans to Build

We plan to approach Unity AI Gateway the way we approached Unity Catalog: get in early, build repeatable methods, and help customers move from fragmented adoption to governed operations on a defined timeline. The first step is a short assessment of the customer’s AI estate, followed by a prioritized implementation plan for routing, budgets, policies, telemetry, and operating ownership. We are building accelerators around that work and talking with the Databricks partner team about how Koantek can lead here as we did with UC. There is not yet a named Unity AI Gateway accelerator program. I expect one will emerge, but we do not need to wait for it to start building the capability customers already need.
‍

Author's Take

Unity Catalog became mandatory because enterprises could not scale the lakehouse without governed access to data. I believe Unity AI Gateway is headed in the same direction because enterprises will not scale agents without governed behavior. Budgets, policies, approvals, and traces are not separate features around the edge. Together, they are the operating controls for enterprise AI.

In 2021, Unity Catalog looked like an important feature. A few years later, it became a buying criterion and a migration mandate. Unity AI Gateway is starting with more urgency because agents are already in production, coding assistants are already spreading, and the cost and security questions are already on executive agendas. That is why I think this announcement will matter long after DAIS week is over—and why Koantek intends to be early.

________________________________________________________________________________________________________

Eddie Edgeworth is CTO at Koantek and a Databricks MVP. Koantek was one of seven launch partners for the original Brickbuilder Unity Catalog Accelerator program and is Databricks’ 2026 Lakebase Partner of the Year.

Want to talk about getting your AI estate governed and cost-controlled? Reach out to Koantek or connect with Eddie on LinkedIn.

________________________________________________________________________________________________________

Sources referenced

• AI governance at DAIS 2026: What’s new with Unity AI Gateway
• What’s new with Unity Catalog at DAIS 2026
• Agent Bricks: Data + AI Summit 2026
• Governing coding agent sprawl with Unity AI Gateway
• Databricks Expands Brickbuilder Program to Include Unity Catalog Accelerators
• Manage budgets for Unity AI Gateway (docs)
• Databricks revenue growth tops 80% to $6.9 billion annualized
• Axios: Databricks moves to stop AI overspend
• Databricks Agrees to Acquire Panther

‍

‍