Workday is creating a dedicated research organization focused on a problem increasingly central to enterprise AI: making autonomous systems reliable enough to operate inside real business workflows. Workday AI Research will investigate areas including agent memory, explainability, multi-agent orchestration and AI efficiency, while also expanding the company’s links with academic researchers through a new PhD fellowship.
Enterprise AI is moving from answering questions to taking actions. That shift raises a harder set of engineering problems: What should an AI agent remember? How can an organization verify why it made a recommendation? What happens when several agents collaborate? And, perhaps most importantly, how can enterprises ensure that sensitive information does not survive after it is supposed to be forgotten?
Workday is betting that answering those questions requires more than adding another AI feature to an enterprise software suite.
The company has launched Workday AI Research, a dedicated technical research team focused on reliable, trustworthy and efficient artificial intelligence for enterprise applications. The research will feed into Workday’s own AI development while contributing methods, evidence and evaluation approaches to the wider research community.
The move places Workday among enterprise software vendors increasingly investing in foundational AI research as agents become part of HR, finance and other business workflows.
The research program covers several technical problems that become more consequential when AI systems can act on behalf of employees. Workday researchers are examining persistent agent memory, explainability, multi-agent orchestration, reward overoptimization during AI training, recommendation systems and adaptive resource control.
Some of the company’s recent research has been accepted at major academic conferences, including the International Conference on Machine Learning, the International Conference on Learning Representations, the ACM Web Conference and the Association for Computational Linguistics.
One area of research focuses on AI agent memory. Instead of allowing an agent to retain everything it encounters, Workday researchers developed a selective approach intended to preserve useful information while filtering outdated, duplicated or unreliable details.
According to Workday’s testing, the method produced 12% higher precision, approximately 8% better overall memory quality and retained 97% of memories considered important. It also ran about 31% faster than the AI-driven comparison used in the study.
The numbers matter because memory is becoming a foundational component of agentic software. An HR agent that remembers an employee’s preferences, previous requests or organizational context can be more useful than a stateless chatbot. But indiscriminate memory can also create privacy, security and governance problems.
Workday’s research into machine forgetting highlights the other side of that equation.
Researchers found that instructing an AI agent to forget information did not necessarily remove every representation of it. In their testing, information could still be recovered from an older summary roughly one in five times. Complete removal required deleting summaries that also contained references to the original information.
For enterprise HR systems, where employee records can contain sensitive personal and employment information, that distinction is more than academic. It raises questions about how AI memory should be designed, audited and governed when agents interact with persistent enterprise data.
Workday is also researching multi-agent systems, in which different AI agents divide complex tasks rather than relying on a single general-purpose agent.
In one study, an exploratory agent investigated options, a second agent focused on compliance and a coordinating agent managed the workflow. The architecture improved accuracy by 5.8% in the company’s testing, while all final responses met the defined constraints.
That approach resembles an emerging enterprise AI architecture in which specialized agents operate under orchestration and governance rather than one agent being granted unrestricted authority.
The strategy comes as enterprise AI adoption expands rapidly but remains difficult to scale. McKinsey’s 2025 global AI survey found that 88% of organizations regularly use AI in at least one business function, yet nearly two-thirds had not begun scaling AI across the enterprise. Only about 39% reported an enterprise-level EBIT impact.
The gap between experimentation and measurable enterprise value is where research into reliability and governance becomes commercially important.
Workday is not alone in pursuing that market. Microsoft is embedding Copilot and agents across its enterprise ecosystem, while Salesforce is pushing agentic workflows through its Agentforce platform. Google, Amazon and other cloud providers are competing to supply the models, infrastructure and orchestration layers beneath enterprise AI applications. NVIDIA, meanwhile, remains central to the compute infrastructure supporting these workloads.
The competitive question for Workday is therefore not simply whether its models are capable. It is whether enterprise-specific research can produce AI systems that understand business context, respect access controls and deliver predictable outcomes inside HR and finance workflows.
Gartner estimates that worldwide AI spending will reach $2.59 trillion in 2026, up 47% year over year, while warning that enterprises are still favoring tactical AI initiatives over disruptive organizational change. Gartner also predicts that 40% of enterprise applications will incorporate task-specific AI agents by the end of 2026, up from less than 5% in 2025.
That makes Workday’s decision to establish a research organization strategically significant. As agents move into production, questions around memory, auditability, orchestration and resource efficiency become part of the product itself.
Workday is also attempting to expand the talent pipeline behind that effort. Its new Workday AI Research PhD Fellowship will provide $50,000 in annual research funding through an unrestricted university gift, mentorship and collaboration with a Workday AI researcher. Fellows will also receive early access to relevant career opportunities at the company.
For enterprise buyers, the immediate takeaway is not that Workday has solved enterprise AI reliability. The research results are early and largely represent specific experiments rather than proof that every production deployment will behave similarly.
The more consequential development is that enterprise software vendors are beginning to treat AI reliability as a research discipline rather than solely a product feature.
That shift could ultimately determine how quickly organizations allow AI agents to move from copilots that suggest actions to systems trusted to execute them.
Market Landscape
The enterprise AI market is entering an agentic phase in which software increasingly performs multi-step work rather than simply generating content or answering questions.
Gartner says up to $234 billion of enterprise application software spending could be exposed to agentic arbitrage through 2030, equivalent to roughly 20% of enterprise SaaS spending by that point. The firm argues that agents capable of completing work across multiple systems could weaken the traditional relationship between software interfaces, users and seat-based pricing.
For HR technology, this could reshape recruiting, employee service, workforce analytics, compensation and talent management. But the same autonomy that makes agents valuable also increases the consequences of errors.
Gartner estimates that by 2028 an average Fortune 500 enterprise could have more than 150,000 AI agents in use, while only 13% of organizations currently believe they have appropriate AI-agent governance.
That creates a market opportunity for vendors able to combine AI capability with enterprise controls, data governance and explainability.
Workday’s research agenda is aimed squarely at that problem.
Top Insights
- Workday launched a dedicated AI research team focused on agent memory, explainability, orchestration, efficiency and reliability for enterprise software.
- Research on selective memory suggests enterprise agents can improve information retention while reducing unnecessary storage, latency and potential data-governance exposure.
- Workday’s multi-agent research indicates specialized agents can divide reasoning and compliance tasks, potentially improving accuracy without weakening defined guardrails.
- Findings on machine forgetting expose a major enterprise challenge: deleting an AI memory may require removing information from summaries and secondary representations.
- A new PhD fellowship expands Workday’s academic research pipeline as competition intensifies around trustworthy enterprise AI and agentic software.
Join thousands of HR leaders who rely on HRTechEdge for the latest in workforce technology, AI-driven HR solutions, and strategic insights
