As AI agents grow increasingly autonomous, organisations that lack the governance frameworks to manage them risk rogue agents, hallucinated responses, and a rapid erosion of trust in their AI deployments.
Without the right context, guided compliance, and sanctioned execution, AI agents can run loose and “lawless”, said Shan Moorthy, Workday’s Asia-Pacific CTO.
Organisations’ underlying business processes are deterministic by design, so that they can deliver consistent and predictable outcomes. This is essential particularly for HR and finance systems, which need to be accurate at all times and where there is zero margin for error.
While this makes AI powerful tools, it also makes the technology unpredictable and fundamentally lawless, Moorthy said, on the sidelines of Workday’s Elevate 2026 Singapore, Workday’s annual customer event.
The lack of predictability presents tremendous risks for finance, HR, and legal teams, where hallucinations can result in inaccurate financial reports, unauthorised transactions, and misinformed decisions, he noted.
And when AI agents have limited context from which to draw, with controls and policies that lack a proper framework, organisations run the risk of having agents that do not operate consistently in compliance.
To achieve the precision and accountability they need for critical workflows, businesses need an AI infrastructure embedded within a trusted architecture.
This will ensure they have a probabilistic AI platform capable of delivering deterministic outcomes, Moorthy said.
He pointed to solutions such as Workday’s Agent System of Record, integrated with its agentic AI platform Workday Sana, that are built to help companies plug gaps in governance and risk compliance when they deploy AI agents.
The Workday Agent System of Record lets businesses manage their AI agents on the same platform on which they manage their employees, providing a unified agent analytics infrastructure that delivers full visibility of the organisation’s hybrid agentic workforce.
Businesses need an AI architecture that gives them 100% explainability and 100% observability, Moorthy said, especially as a growing number of AI agents are deployed to handle tasks and take actions on behalf of employees.
Workday Sana provides an infrastructure that ensures agents run on the necessary guardrails and within trusted business processes, including approvals, access controls, and data security policies.
Sana can generate answers that are personalised and “policy-aware”, said Wong Pei Woan, Workday’s Asean senior director of solution consulting.
It understands context and is able to look at an employee’s calendar and company policies, for instance, and generate a monthly expense report that is compliant, Wong said at Elevate Singapore.
The Workday Agent System of Record also provides a dashboard that indicates the number of AI agents that have been deployed within the organisation as well as agents that are the most active. This spans all agents, regardless of whether they were built by Workday, the enterprise customer, or its partners.
It also enables users to choose which skills to activate for each agent and set up the necessary security profile for the agent, Wong said. The agent then will automatically inherit the access policies required for it to perform its task.
The dashboard further shows the value of each agent, so the organisation can decide whether to shut down underperforming agents or double down on those delivering high value.
Organisations need to move from AI that assist, to AI that actually do the work, said Jess O’Reilly, Workday’s Asean general manager.
They need context-aware AI assistants that operate not only autonomously, but also safely, O’Reilly said during her opening address at Elevate Singapore.
AI needs context to be useful, she said, noting that general-purpose LLMs (large language models) can help create tasks and automate routine work at scale and speed, but do not have context on how an enterprise operates.
“They can’t, and won’t know how your work gets done,” she said.
This is where Workday aims to help its customers with their agentic AI deployments, she added.
Unified platform for better visibility, faster response
When done right, AI can yield significant gains in speed-to-market and power impactful decision-making.
Razer worked with Workday to consolidate its people processes across 19 global offices, creating a single source of truth for more consistent and efficient decision-making.
By embedding AI and machine learning into its HR strategy, the gaming peripheral company was able to enhance its talent acquisition and employee experience.
It ensures Razer remains agile, allowing it to bring on new skillsets more efficiently to support its global growth.
“Workday continues to be at the forefront in terms of what AI and machine learning can bring, positioning us for innovation, agility, and growth,” said April Wan, Razer’s global HR head.
Scoot, the low-cost subsidiary of Singapore Airlines, also tapped Workday talent acquisition and development tools to consolidate its employee data onto a unified cloud platform.
With Workday Extend, Scoot’s HR team is able to create tailored, concierge-style applications on-demand, allowing employees to take ownership of their own career development.
It provides the airline with an accurate view of employee sentiments and needs, and ensures all HR strategies are supported by data-driven insights, rather than intuition.
“Workday’s product roadmap was a big plus for Scoot. It demonstrated that there was another forward-looking organisation, always looking beyond the ‘now’,” said Ivan Chuah, Scoot’s HR director.
Customise AI to your business requirements
JLL Singapore has used Workday Extend to build custom applications, including “Easy HR”, which lets employees generate complex reports through natural language queries.
On the recruitment front, the integration of Workday’s HiredScore AI has overhauled how the real estate firm attracts and selects talent. Quarterly hiring is up 64%, and time-to-fill has fallen, driven by AI-powered candidate grading and automated lead surfacing from existing databases and partner networks.
“With HiredScore AI for Recruiting, we have transformed our hiring processes, dramatically increasing recruiter productivity while also improving candidate quality,” said Ajay Pimpalshende, JLL Singapore’s director of product management.
Rakuten also brought in Workday to transform its diverse and complex organisation structure into a globalised, data-driven business.
The e-commerce operator implemented Workday Human Capital Management and Workday Prism to establish standardised data across unique, localised markets, and provide executives with real-time insights.
The platform serves as a global integration platform, streamlining the integration of new acquisitions and supporting skills-based talent management across 30 countries.
In addition, Rakuten is leveraging generative AI to accelerate routine tasks and guide employees through complex processes, ensuring a thoughtful and ethical approach to data usage.
“For us, Workday is more than an HCM system. It’s really a global integration platform,” said Rakutan’s global HRIS Regina Krug.
“We’re at an intersection where organisations have access to the tools, but not necessarily a strategy about what they’re doing with the tools,” Moorthy said. “Most are still in the experimental stage and trying to get AI to operational.”
To add to that challenge, they now have to expand their AI deployments to include agentic in a significant way.
CIOs have been tasked to figure out their organisation’s AI strategy even amidst a lack of use cases from which to draw and build on, he noted.
And when there are use cases, these do not necessarily have the regulatory and compliance frameworks and controls that their organisation needs, he added.
There are big gaps businesses must resolve if they want to transition from experimentation to production, Moorthy said.
Workday designed its solutions to meet such requirements, he noted, including providing the guardrails that enterprises need to ensure their AI agents remain “lawful” by nature, while still delivering the operational efficiencies and value that probabilistic AI models promise.
