Building governed agentic AI with Unity Catalog and Iceberg

Workday, the leader in enterprise cloud applications for finance and human resources, serving more than 10,000 organizations worldwide, needed to scale AI agents across its platform without fragmenting data or eroding trust. By partnering with Databricks to build a universal data layer on Apache Iceberg™ — governed end-to-end by Unity Catalog — Workday transformed a multiplicative data-silo problem into an additive, scalable architecture that brings agents to the data rather than replicating data to agents.

From Data Silos to a Universal Data Layer — Solving the N×M Problem

As Workday began deploying AI agents across departments — from procurement and supply chain to finance and HR — the team quickly recognized a familiar challenge resurfacing in a new form. Each agent operating with its own independent data store created point-to-point connections that multiplied complexity exponentially. With 100 agents and 100 data systems, the architecture looked like what Phoenix Majumder, Senior Director of AI Engineering and Platforms at Workday, calls "a very complex spaghetti". Meaning it was impossible to govern, observe, or monitor with confidence.

"We realized we cannot operate like this," explains Majumder. "If you have point-to-point connections at scale, someday some agentic workflow will give you wrong information or make a wrong decision, and the whole trust plane around AI will come down the drain. One hundred good things, but one bad thing and the trust erodes."

The challenge was not merely technical; it was existential for AI adoption. Workday's leadership understood that trust, once lost, is nearly impossible to rebuild. As Majumder notes, "Even in the early days of BI, you could generate hundreds of dashboards, but one dashboard will give that wrong piece of information to someone sitting in a C-suite of the organization. They will mistrust all the dashboards that come into their desk. And that might happen in the agentic space, too."

Rather than accept this trajectory, Workday made a deliberate architectural decision: convert the multiplicative problem (M × N) into an additive one (M + N). The team needed a centralized, open, and forward-compatible data foundation. One that could serve as a single source of truth while remaining interoperable as agentic frameworks rapidly evolve. The guiding principles were clear: centralization of data, centralization of governance, and making that layer the place where agents come to consume data rather than replicating data outward.

Building the governed agentic data foundation with Unity Catalog and Iceberg

Workday's solution centers on two key technologies working together: Apache Iceberg™ as the open table format and Databricks Unity Catalog as the unified governance layer. Together, they create what Workday calls its "universal data layer": a foundation where AI agents access current, governed data without creating additional replicated stores.

Apache Iceberg™ provides the transactional guarantees that enterprise AI agents demand. Its ACID compliance and snapshot isolation give agents a consistent, stable view of data while they reason and act, which is critical when multiple agents query the same datasets simultaneously. But Iceberg's most transformative contribution, according to Majumder, was cultural: "Iceberg brought oil and water together. Data science and ML engineering teams want to innovate and experiment really fast — hypothesis testing and validation is very important to them. But platforms and data engineering teams, their work is fairly deterministic. Iceberg was a format equally embraced by two opposing schools of thought. Oil and water mixed; they fused together."

Unity Catalog serves as the centralized governance control plane, delivering fine-grained access control, end-to-end lineage, comprehensive auditing, and consistent policy enforcement across all data and AI assets. Majumder illustrates the necessity with a concrete example: "An account executive who works in the North Territory shouldn't be able to ask the agent to tell them the account status or health of the South Territory. The agentic interface should know the entitlement of that individual and block the information they are not entitled and authorized to see. You shouldn't be able to prompt inject it and get information you are not authorized to see. That would be a debacle, especially when operating with highly sensitive data."

"Unity Catalog was a key contributor to achieving true enterprise scale for us when it comes to compliance, security, observability, lineage, tracking, and monitoring," says Majumder. The architecture also incorporates long-term memory (stored in the Iceberg layer as the persistent DNA of agents) and short-term session memory (vector stores for contextual interactions), with reasoning traces captured for auditability, ensuring that if an agent's output is ever questioned, the team can "open the hood" and trace the rationale end-to-end.

The modular design philosophy is intentional. "We have to be very deliberate to design our systems modularly," Majumder explains. "We have to think like we are builders of a Lego toy. You take the parts and build a toy. But equally, if a part breaks down in future, you should be able to swap it very quickly. That's very important because we're still in uncharted territory."

Accelerating Business Outcomes Across the Enterprise

With the universal data layer and Unity Catalog governance in place, Workday has unlocked measurable business value across multiple functional areas. The architecture enables use cases spanning finance close-cycle acceleration, early attrition-risk identification in HR, supply chain and procurement contract negotiation optimization, and improved retention modeling and sales planning accuracy. As Majumder describes: "You could accelerate your finance closeouts or cycles. You could look at your attrition rates way ahead of time. You can improve the way you do supply chain management, ensuring that you could get the best out of your negotiations. Then obviously marketing and sales use cases where your retention and your sales planning can improve a lot."

The architecture's true power lies in its ability to scale safely. Workday progressed from single-agent pilots to coordinated multi-agent deployments across departments, all without creating technical debt or governance gaps. The unified approach eliminated redundant data pipelines, and the modular design means models can be swapped out, cost controls can be applied at natural choke points, and new agents can be onboarded rapidly because the foundational data and governance layers are already in place. Data integration complexity was reduced from a multiplicative (M × N) to an additive (M + N) model. A fundamental architectural simplification that compounds in value as the number of agents grows.

"We are just getting started," says Majumder. "If 10 is the best state, we are probably at a 3. That's what makes it exciting. We have to cover seven more steps." Looking ahead, Workday is focused on three priorities: expanding interoperability as organizational data grows more complex and comes from more sources; establishing registries and communication protocols for agent-to-agent interaction (including MCP servers); and deepening governance automation and observability to monitor and steer agents over time — treating them as "digital employees" whose reasoning can be analyzed, corrected, and improved.

The foundation built with Databricks, Iceberg, and Unity Catalog positions Workday to scale confidently from today's coordinated agent deployments to tomorrow's fully autonomous enterprise workflows. As Majumder summarizes: "It's not just about building things from the ground up. It's about how you monitor, observe, and automate those components. The history of software engineering is repeating itself. These agentic interfaces are proliferating everywhere, but for them to truly work coherently, they need to start talking to each other."