Workday Plans Honeydew Buy for Governed AI Data Queries
Workday plans to acquire Honeydew, compiling governed definitions into live Data Cloud queries so AI agents, dashboards and models produce consistent answers.

Why it matters
- Workday announced the planned Honeydew acquisition on November 18, 2026, at Workday Rising EMEA in Barcelona.
- The transaction is expected to close in the fourth quarter of fiscal 2027, ending January 31, 2027, subject to customary conditions.
- Honeydew’s semantic compiler would turn governed business definitions into deterministic queries before requests reach AI models.
- The integration would apply definitions directly to live Workday Data Cloud data and carry permissions with each query.
- Workday expects improved accuracy and efficiency, plus lower token use and costs.
Workday expects its planned acquisition of Honeydew to close in the fourth quarter of fiscal 2027, ending January 31, 2027. The company announced the deal on November 18, 2026, at Workday Rising EMEA in Barcelona. Subject to customary closing conditions, the acquisition would place Honeydew’s semantic compiler inside Workday Data Cloud.
Workday says the compiler will give AI agents and business-intelligence tools the same governed business definitions. The goal is consistent answers across agents, tools and models, even when each interface starts with a different rule. The proposal moves the choice of definition into a shared layer instead of leaving every interface to resolve it independently.
Workday illustrates the problem with a headcount request. One dashboard may count only full-time employees. A chat experience may include contractors. An AI agent may count approved but unfilled roles as already staffed.
Each answer can be valid under its own definition. Together, though, they leave a business without a consistent number. In Workday’s account, the problem is not one bad calculation. It is the absence of a shared rule for choosing among valid definitions.
What does Honeydew change?
Honeydew’s semantic layer lets teams define business metrics once and use them across business-intelligence tools and AI agents. Its compiler turns those definitions into precise, governed queries. The layer can capture and map multiple definitions rather than forcing one meaning onto every metric.
Honeydew then selects the applicable definition based on the person or agent asking, the specific use case and the situational context. That selection happens in the shared layer. A dashboard, chat experience and agent would no longer each resolve the definition independently before producing their own version of the answer.
This arrangement gives Workday a place to enforce business meaning before an answer appears. In the headcount example, the shared layer would decide which population the requester intends to count before any interface turns that interpretation into a result. That is a different job from teaching a model the definitions through more prompt text.
How does the semantic compiler change AI queries?
Workday describes the current alternative as custom code that intercepts large-language-model requests. Teams also hardcode business rules into data pipelines. A third method, schema-stuffing, packs data dictionaries, SQL join rules and definitions into prompts.
The company says that work can have to be repeated when teams switch models or platforms. With Honeydew, business logic would move out of the prompt and be compiled into deterministic queries before a request reaches the model. The model would still receive the request, but it would no longer have to improvise the metric’s underlying calculation.
“An AI model shouldn’t be improvising the calculation behind a payroll number,” said Gabe Monroy, Workday’s chief technology officer.
His line captures the intended split. Prompts can describe a metric, while the compiler enforces the calculation and its permissions. Workday expects the approach to improve accuracy and efficiency. It also expects lower token use and lower costs, while preserving definitions when customers change tools or models.
Why does live Data Cloud matter?
Workday distinguishes the planned deployment from semantic layers that operate on data already extracted into a warehouse or lake. With those systems, answers can be only as current as the latest refresh. The proposed Honeydew integration would instead apply customer definitions directly to live data in Workday Data Cloud.
That design lets agents use current information such as the same day’s ledger. It also lets a governed metric run against live enterprise records rather than a separately refreshed extract. Workday presents this direct-to-data approach as a material difference from semantic layers built around warehouses and lakes.
Freshness changes what a business metric can mean in practice. A rule for the current ledger cannot rely on an older extract. Under Workday’s plan, definition logic and source data sit together in Data Cloud, reducing the gap between business meaning and the state of the business.
What changes before a query runs?
Workday already governs agent identity, verification, access and behavior. Honeydew would add another control point at the moment a natural-language question becomes a database query. That control would sit before data access, rather than appearing after an agent has already acted.
Permissions would travel with business definitions. Each person or agent would see only permitted information, without separately managed data extracts. Workday says invalid or unauthorized requests would be blocked before a query runs, rather than flagged afterward. The ordering matters because it lets the system stop an improper operation before it touches data.
Yasmeen Ahmad, senior vice president and general manager of Workday Data Cloud, described enterprise context as the central bottleneck for enterprise AI. She described the planned integration as a “living semantic graph” providing autonomous agents with governed business context.
For Ahmad, context is not background text. It governs which interpretation and data can support an agent’s next action.
What does this mean for Workday’s agent strategy?
The planned acquisition puts business definitions at the center of Workday’s agent strategy. One layer would determine how a question is interpreted, which data can answer it and whether the request is allowed to run. Governance would extend from an agent’s actions to the answers on which those actions depend.
For companies deploying AI agents, the practical stake is whether different interfaces can return governed answers without rebuilding business logic and access rules each time. Workday’s proposal pairs the metric with its access rights and sends both through the same layer. That could reduce the fragmentation Workday describes in model requests, data pipelines and prompts.
What has to work after the deal closes?
The proposal still faces customary closing conditions before it can reach Workday Data Cloud. Its first test is consistency across those interfaces. A headcount answer should come from Honeydew’s shared definition logic, rather than three separate interpretations of full-time staff, contractors and approved vacancies.
Workday also expects the permission check to happen before query execution. Live Data Cloud records must supply current answers, including the same day’s ledger where applicable. Customer definitions need to survive a move to another model or tool, while the integration improves accuracy and efficiency and reduces token use and costs.
For Workday customers using AI agents, the practical promise is controlled access to a consistent answer. The company is placing business definitions between a question and the data, then carrying permissions along the same path. If the system performs as expected, Honeydew could turn enterprise context from a prompt-building task into infrastructure for autonomous agents.
The acquisition will be judged on that operational layer: consistent definitions, live answers and rejected queries, all enforced before an agent acts.
Source: unite.ai
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
744 articles
Related articles
- Microsoft merges coding and productivity AI into one Copilot app
- Collate's Ontology Studio Gives AI Agents a Formal Map of Business Meaning
- OpenAI Says Agents Have Replaced Chatbots as Its Default Work Tool
- ServiceNow Makes OpenAI a Preferred Intelligence Layer for 80 Billion Workflows
- AI Users Work Up to 90 Minutes Longer Than Colleagues, ActivTrak Study Finds