Workday announced a four-layer Workday Data Cloud at Workday Rising earlier this month, giving AI agents direct access to live data in its System of Record under the permissions customers already use. The layers are Open Data Core, Data Access and Processing, the Context Layer, and Agentic Experiences.
Agent Lake, which holds the Open Data Core, and Live Data Query, the main capability in Data Access and Processing, are generally available. The Context Layer’s semantic models and Agent-Ready Tools for Data, part of Agentic Experiences, are in early access.
Workday says teams no longer have to extract and duplicate data, or rebuild security, before agents can use it. Agent actions beyond policy limits go to a person for approval.
Open Data Core and Live Data Query
Open Data Core, the first layer, brings operational and historical data together.
Workday’s transactional system is an object-oriented graph holding data, calculated logic, relationships, and effective-dated history, and Workday is writing more than 240 curated business objects from it, including Worker, Position, and Cost Center, into Apache Iceberg format in Agent Lake. Zero-copy federation makes them available to Google BigQuery, Salesforce Data 360, and Snowflake, all generally available, with Databricks in early access.
Contextual Security carries over to the lake and scopes access by a person’s relationship to the data: a regional HR director sees their region, while a compensation partner sees one pay grade across every region.
Data Access and Processing, the second layer, delivers that data to the tools teams already use. Its main capability, Live Data Query, which reached early adopters by June, routes each query across live transactional state and Agent Lake in sub-second time with no extract step. It runs on the open-source Trino engine, so analytical load stays off transaction processing, and exposes JDBC, BI, and Model Context Protocol (MCP) endpoints.
Workday reports early adopters ran more than 1,645 queries in three months, two to three times faster than legacy report-based extraction.
Context Layer and Agent-Ready Tools for Data
The Context Layer, the third layer, defines business terms once. Its Financial and HCM semantic models package Workday’s KPI and dashboard logic for metrics such as headcount and voluntary turnover. In Workday’s example, a question about regrettable attrition among EMEA people managers returns the same answer from any agent, because the model holds the definition.
Agentic Experiences, the fourth layer, is where agents act. Its Agent-Ready Tools for Data let agents discover schemas and ingest or extract data over MCP, and every call passes through the Workday Agent Gateway, which enforces runtime authorization, rate limits, and business policy. Enforcement sits in the gateway, so a limit does not depend on how an agent was prompted.
An action that exceeds a guardrail, such as a spending limit, goes to a human approval workflow or triggers a transaction rollback. Each agent’s identity, skills, and execution log are registered in the Workday Agent System of Record through Agent Passport, an early access tool that tests and monitors agents before and during deployment.
What This Means for ERP Insiders
Workday is treating data, meaning, and action as one stack. Open Data Core and Live Data Query supply the records, the Context Layer supplies the definitions, and Agent-Ready Tools for Data supply governed action. HR and finance leaders get all three from the same system of record, under the same permission model.
Metric definitions are becoming a platform asset. Workday is packaging its own KPI and dashboard logic in the Context Layer as semantic models that any agent can query. ERP teams that define turnover or headcount once then see the same answer wherever an agent runs.
Open access depends on the system of record’s controls. Workday extends its permission model to the platforms where customers already analyze data. BI teams, data platform owners, and agent builders meet one set of access rules instead of rebuilding them in each tool.



