HRMS News logo

Workday buys Honeydew to move AI logic into deterministic queries

Workday buys Honeydew to move AI logic into deterministic queries

Workday announced on November 18, 2026 that it will acquire Honeydew, a semantic layer for AI and business intelligence, according to ETHRWorld. The deal is designed to improve AI accuracy by shifting business logic out of prompts and compiling it into deterministic queries that run against live Workday data.

Honeydew CEO David Krakov said the core challenge was never getting AI to write database queries but getting it to understand the true definitions behind company data. Workday CTO Gabe Monroy added that an AI model should not improvise the calculation behind a payroll number. With Honeydew, the model states what it needs and Workday builds the exact query the same way every time, within each person’s or agent’s permissions.

Why semantic layers matter for HR data

Most semantic layers sit on data already extracted into a warehouse or lake. That means answers are only as current as the last refresh. For payroll and HR teams, a stale answer can be worse than no answer. A headcount figure pulled from yesterday’s extract may miss overnight terminations or new hires.

Honeydew applies a customer’s definitions directly to live data in Workday Data Cloud. Agents can act on answers from today’s ledger rather than a delayed copy. For a payroll manager asking about gross-to-net calculations, the query uses the same logic the payroll engine uses, not an approximation generated by a language model.

The distinction is practical. A large language model can write SQL that looks plausible but calculates overtime incorrectly. A semantic layer enforces the organization’s actual definitions for earnings codes, cost centers, or leave balances before any query runs.

How governance changes at the query layer

Honeydew adds governance at the moment a question becomes a query. Workday permissions travel with the definitions, so each person or agent sees only what they are allowed to see. There are no extracts to manage and no separate permission model to reconcile.

Invalid or unauthorized requests are blocked before any query runs, not flagged after the fact. That is a meaningful shift for HR teams handling sensitive data. A manager asking an agent about compensation for a direct report should not trigger a query that pulls salary bands for the entire department. The semantic layer blocks the request before it executes.

Paired with Workday agent governance, Honeydew helps ensure agent responses are as governed as their actions. For organizations deploying AI agents in HR service delivery, this closes a gap between what an agent can do and what it can say.

Workday buys Honeydew to move AI logic into deterministic queries (illustration)
Photo: Maj. Gen. Sarah Zabel discusses the Air Force Journey to Agile Software Acquisit by Todd Maki, Public domain

What this means for payroll accuracy

Payroll is the highest-stakes use case for this acquisition. A payroll number is not a suggestion. It must match the general ledger, tax filings, and employee records exactly. When an AI agent answers a question about a payroll figure, the calculation behind that answer cannot be improvised.

The deterministic query approach means the same question produces the same answer every time, assuming the underlying data has not changed. That consistency matters for audits, compliance reviews, and employee trust.

For finance leads, the benefit is auditability. A query compiled from governed definitions can be inspected and reproduced. A prompt-generated answer cannot. When a payroll discrepancy arises, the team can trace the exact query that produced the number.

Practical steps for HR teams

The acquisition does not require immediate action from Workday customers. Honeydew’s capabilities will be integrated into Workday’s platform over time. However, HR and payroll leaders can prepare by reviewing their data definitions now.

Common mistakes include maintaining inconsistent definitions across systems. If the definition of a full-time equivalent differs between Workday and a downstream warehouse, any semantic layer will inherit that conflict. Teams should document how key terms are defined in Workday and resolve discrepancies before relying on agent-generated answers.

Another step is auditing permission models. Because Honeydew enforces Workday permissions at query time, gaps in role assignments will surface quickly. A person who should not see compensation data but currently can will be blocked once the semantic layer is active. That is a feature, but it may surprise teams that have not kept permissions current.

Area Current state After Honeydew integration
Query source Prompt-generated SQL Deterministic compiled queries
Data freshness Warehouse extract, refresh-dependent Live Workday Data Cloud
Permission enforcement After query runs Before query runs
Business logic Embedded in prompts Governed definitions
Audit trail Limited for AI answers Reproducible queries

What remains unclear

The announcement does not state when the acquisition will close or when Honeydew capabilities will appear in Workday products. It also does not disclose financial terms or how many Honeydew customers will be affected.

Workday has not said whether Honeydew will remain available as a standalone product or be folded entirely into the Workday platform. HR teams evaluating semantic layers independently should watch for integration timelines and product roadmaps from Workday.

For organizations outside the Workday ecosystem, the acquisition signals a broader industry direction. AI accuracy increasingly depends on governed data definitions rather than better prompts. That principle applies whether the system of record is Workday or another platform.

Sources: ETHRWorld

Photo: Streamax AI technology applications in the railway industry (32642645637) by International Railway Summit, CC BY-SA 2.0

administrator

Related Articles