Workday’s AI Bet Is Built on Data You Already Own
Workday spent roughly eighteen months assembling the pieces. The $1.1 billion acquisition of Sana Labs in September 2025 gave it an AI-native front end. The Pipedream acquisition added over 3,000 third-party connectors. Flowise contributed to orchestration. The March 2026 global launch of Sana from Workday is where those acquisitions converge into a product claim: a single AI layer that finds answers, executes tasks, builds dashboards, and automates workflows across HR and finance using data Workday already holds.
Workday’s strongest argument has nothing to do with features. It is that your HR and finance data already lives there. Payroll history, headcount plans, benefits records, financial close data – years of it, structured and governed, tied to real identities. Sana does not need to go find that data or connect to it through an integration project. It starts with it. Any AI agent your team builds from scratch must earn that context first. Most never fully get there.
The capability that separates Sana from a better search interface is execution. A user can instruct Sana to update a new hire’s home address and immediately surface the downstream tax-withholding impact – one instruction, multiple backend systems, no manual handoffs. That is a different category of tool than a copilot that drafts a suggested response and waits for approval.
Sana’s Core Pillars: Find, Act, Build, Automate
Sana advances work automatically through four key capabilities:
- Find: Delivers quick, cited answers to complex questions, eliminating manual searches.
- Act: Executes instructions, such as updating employee information, handling all backend processes.
- Build: Generates custom dashboards, documents, or assets from existing data.
- Automate: Uses no-code tools to run multistep workflows connecting Workday with external apps like Slack, Gmail, and Salesforce.
The Illuminate agents and Workday Build are how that execution capability gets deployed in practice. Illuminate agents are purpose-built for specific use cases – performance review processing, financial close – and are narrow by design. Workday has been deliberate about not positioning them as general assistants. That choice reflects a hard lesson from enterprise AI deployments: Broad general agents fail more visibly than narrow task agents because their failure modes are harder to contain. A general agent operating across ambiguous inputs is more likely to hallucinate or produce low-confidence outputs at the exact moments that matter. A narrow agent with defined data inputs and a bounded task scope fails less often and fails more predictably. Workday Build extends this logic to custom use cases, letting teams create purpose-built agents without starting from zero code, using Workday's existing data model as the foundation.
European industrial group Berner piloted Sana and achieved 90% employee adoption, retiring 400 generic ChatGPT licenses in the process. The value wasn’t having AI; it was having AI that understood their specific business context and security requirements.
Deployment timelines are more credible than the category average. A proof of concept can be operational in four to six weeks. Full production rollouts are being achieved in three to six months. That is possible because Sana runs on Workday’s existing data model rather than requiring a parallel data integration project. Organizations that have already gone through a Workday implementation are not starting from scratch.
Workday Build lets your team create custom AI agents without writing code from scratch. Workday positions it as the option for organizations that want more than the standard agents but don’t want to build everything themselves. The honest question is how much of Workday’s underlying systems Build actually gives you access to and how much stays locked. Vendors typically open just enough to satisfy the customization argument while keeping the core under their control. Get a live demonstration before assuming Build can do what your use case requires.
Our Take
This announcement is aimed squarely at HR and finance leaders inside existing Workday shops who need to show AI progress without taking on development risk. If that’s your situation, the case for leaning in is genuine. The data governance story is real, the timeline advantage over custom builds is real, and consolidating point AI tools onto a platform you already pay for is defensible to a CFO.
The Pipedream acquisition is the part worth watching closely. Orchestration across third-party systems is where most enterprise AI deployments quietly die, and Workday has bought a credible answer, tucking it under the Workday AI platform brand rather than leading with the acquisition name.
Every major platform vendor is making a similar bet right now. SAP is trying the same thing with Joule but is slowed down by having customer data split across old on-premises systems and newer cloud versions. Oracle is embedding advanced AI models directly into its HR and finance products, betting that a better model matters more than tighter integration. ServiceNow has picked Anthropic’s Claude as its core AI engine and is building controls around it. These are genuinely different strategies. Which one pays off depends entirely on where your data lives and how your organization actually works.
Before committing, get the Workday Build API surface in writing – not a roadmap, a working demo. And nail down the consumption pricing model before your organization is too embedded to negotiate. Workday’s renewal leverage is real, and agentic AI will create new billing surface area that most contracts don’t currently contemplate.