The AI Interview: Jerry Ting, VP of Agentic AI at Workday

When GPT-3.5 arrived in late 2022, enterprises rushed to experiment with generative AI. Marketing departments promised agents that would transform work and engineering teams built features that generated text. The gap between promise and implementation, however, soon became a problem.

Jerry Ting describes this period as a false start: “The marketing went from that to agents, but the way that engineers actually use GPT and foundational models was more feature-based generative AI."

Jerry looks after agentic AI at Workday, where he moved after the company acquired Evisort, the contract AI platform he founded, in 2024.

Workday has accelerated its AI strategy through acquisition. The company completed its US$1.1bn purchase of Sana, a Stockholm-based AI firm, in early November and acquired Flowise, a low-code agent builder with more than 42,000 GitHub stars, in August. The company has also announced plans to acquire Pipedream, an integration platform with more than 3,000 pre-built connectors.

What Workday calls agents now differs from what the term meant two years ago. The company has built systems that perform job functions rather than assist with them. “Agents are now doing part of what the labour force does,” Jerry says. “And that’s what we call at Workday a role-based agent.”

Agents in payroll, for example, handle administrative tasks that follow established processes but consume time. “We have an agent that handles the worst parts of payroll that an admin doesn't want to do,” Jerry says. “That’s going beyond what software can do – it’s doing the job of the person using Workday, so that person can do more.”

Workday Agent Builder addresses build versus buy decisions

Chief information officers ask Jerry the same question in different ways. Should they build AI capabilities internally or buy them from vendors? “Right now, when I talk to CIOs, it’s a build versus buy conversation,” Jerry says. “And then they try to make me defend why they should buy from us.”

Agent Builder represents Workday’s answer. The platform lets organisations create custom agents for specific requirements whilst maintaining security and compliance standards. Every organisation reaches a point where off-the-shelf solutions stop fitting.

“We know you’re going to have to customise,” Jerry says. “There’s that last mile that's industry-unique, organisation-unique – do it with us. Do it in a secure, compliant way. Don’t do some homegrown thing and then hack your way back into Workday.”

Jerry runs a centralised AI team at Workday that evaluates use cases and determines which agent types fit which tasks. The team exists to prevent a pattern he sees in IT departments.

“They’re not as nuanced about the use case – they just try to automate the human, and then you get really strange results,” he continues. “You breach trust with your stakeholders, and then you’re stepping back on your AI deployment.”

How Workday’s Pipedream and Sana acquisitions extend agent reach

An agent that only accesses Workday in a ‘walled garden’ approach has limited use. Work happens across platforms, with employees using Microsoft Teams and Slack and IT systems running on ServiceNow, for instance.

Workday’s Pipedream acquisition addresses what Jerry identifies as a fundamental constraint, with the platform serving more than 5,000 customers and offering connectivity to applications including Asana, HubSpot, Jira, Recurly and Slack. Jerry offers employee onboarding as an example, where a customer might need badge creation integrated into the workflow.

“We can call out to ServiceNow and say, ‘As part of employee onboarding, we can get that badge for the employee and do that as part of a business process that lives in Workday’,” he adds.

Workday also completed its acquisition of Sana in early November after announcing the US$1.1bn deal in September. Sana serves more than one million users across hundreds of enterprises with two products. Sana Agents enables organisations to build AI agents using a no-code interface, while Sana Learn provides learning management with AI-native features. The platform indexes company information from sources including Google Drive, SharePoint and Office 365, making it searchable through AI.

Jerry describes the acquisition as addressing interface problems: ”The thing we all complain about with Workday is that there are sometimes too many clicks. Logging in as a father expecting a kid, I just need to do one thing. I don’t need to do 50. I want to do that one thing quickly and get out. Sana lets you do that – it simplifies the user experience.”

Workday positions Sana as what it calls the new front door for work, a single interface that connects systems, data and actions. The platform will serve Workday’s 75 million users under contract across more than 11,000 organisations.

The three acquisitions work together. Flowise provides the tools to build agents, Pipedream provides the connections to other systems and Sana provides the interface. Jerry says the company plans to introduce capabilities from all three during 2026, possibly in the first half of the year.

Human oversight remains central for payroll verification

Workday divides agents into two categories based on risk. The first handles processes where errors have financial or legal consequences. The second monitors and analyses continuously without human oversight.

For high-risk processes, humans make final decisions. At its Rising EMEA conference, Workday demonstrated an agent that identifies budget availability and recommends spot bonuses. The agent performs the analytical work – reviewing budgets, checking eligibility criteria and preparing recommendations – but stops short of execution. The agent suggests and the human decides. “For a lot of business processes today, it’s good to have human verification because getting payroll wrong is not funny,” Jerry says. “You’ve got to get it right. We take responsibility very seriously.”

The second category of agents operate with greater autonomy. Workday calls these ambient agents. They run continuously, performing work that organisations need completed but cannot realistically assign to humans on a constant basis.

“An example is the audit agent that we’re building,” Jerry goes on. “Ideally, a CFO can check the financial accuracy of the general ledger at all times, but nobody just sits there and stares at it.

“These ambient agents are always on. They do this type of work that humans don’t do today, but it adds more value from a software perspective.”

Multiple agents collaborate on complex enterprise tasks

Workday builds agents that focus on single functions, with specific specialisations in payroll, compensation, planning and more. These are powerful alone, but combined are a different prospect.

“Where I get excited is when a payroll agent works with a compensation agent and a planning agent – now you’re getting a team of analysts working together for one user,” Jerry says, with multiple specialised agents collaborating rather than one agent attempting to handle everything.

The combination of memory, context and collaboration produces systems that approach problems through reasoning.

“You have a context layer where the agents are intelligent and reasoning, doing chain-of-thought reasoning,” notes explains. “How do humans think about a problem? I don’t think we’re that far away.”

Whether this constitutes progress towards artificial general intelligence depends on definitions, with Jerry focusing more on theoretical milestones rather than practical capabilities. “I think about agents doing really big parts of things that we don’t want to do,” he says. “Depending on how you define it, we have a lot of that today.”

What’s the difference between functional AI and science fiction? “When you make it about HAL from 2001: A Space Odyssey, you start to conflate these issues,” Jerry says. “But in terms of whether AI can benefit the world in a very generalisable way – I think we’re there.”