Workday and NVIDIA leaders explain why enterprise AI must be open, specialized, and trusted gbefore it can run payroll, finance, and HR at scale.
Ghadeer Redler
Head of Multimedia, Storytelling
Workday
AI can draft an email, summarize a meeting, and answer almost any question you throw at it. But when the task is payroll, finance, or compliance, “close enough” isn’t good enough.
In this episode of Future of Work, Stephanie Felix, Sr. Principal Functional Architect at Workday talks with Workday’s Vice President of Product & Technology Josh Zywien and NVIDIA’s Sr. Director or Enterprise AI Marketing Erik Pounds. They break down why enterprises are shifting toward open, specialized models, why trust is the real barrier to AI at scale, and what changes when CHROs and CFOs feel confident handing agents their most sensitive work.
Press play to hear the full conversation.
Josh Zywien: You have this delta between what the model can do, what its—what the potential of what what can be done, and then how it can actually be applied in the enterprise because you cannot—the the probabilistic nature of AI is you can't get this stuff wrong. You can't get payroll 98.5% right. It has to be 100% right every time.
Narrator: Everyone is talking about what AI can do. Almost no one is talking about what happens when it gets your paycheck wrong. Welcome to The Future of Work podcast. Today, Workday's Josh Zywien and NVIDIA's Erik sit down with Stephanie Felix to unpack why enterprises are quietly walking away from giant one-size-fits-all models, and why the real barrier to AI at scale isn't intelligence—it's trust. Stay to the end for a prediction about the next 12 months that might just surprise you. Let's get into it.
Stephanie Felix: Welcome, everyone. Erik, let's start the conversation with you.
Erik Pounds: Hey, thanks for having us.
Stephanie Felix: We're seeing a big macro shift across the AI ecosystem where enterprises are moving away from closed general-purpose models towards specialized open-weight models like NVIDIA's Nemotron family. Can you tell us more about why, from an industry perspective, open model—is becoming such a strategic priority for tech leaders?
Erik Pounds: Yeah, I mean the adoption of open models is driven by a few things. One is, you know, they want to be able to trust the model, control the model, own their AI, run it where they need to, and then be able to take that model and and really make it their own by customizing the model, fine-tuning the model to build specialized agents—agents are that are really, really good at doing one thing and one thing, you know, really really well. And then efficiency. As we're all using AI, we know that, you know, we're the the the number of tokens we're generating is keeps climbing, climbing, and climbing, and we're looking to get as much as possible out of our investment. And by using open models can help you do that. And often these open models are being used alongside large frontier models. So it's not often as much an either/or, it's a it's agents using these systems of models where both types of models are used.
Josh Zywien: Yeah.
Stephanie Felix: Yeah, that totally makes sense, and I've heard so much at Rising, like customers' very unique, specific requirements. So, I'm sure that's resonating a lot. All right, next question. Building on that point, Workday Workday customers hold a company's most sensitive data—it's people and it's money. How do open models change the trust and governance conversation when bringing AI into those core systems?
Erik Pounds: Yeah. A lot of it is about just running the AI where your data is. So sometimes it's okay for the data to leave where it lives at rest or or or be used outside the platform, then come back into the platform. Sometimes you need to stay stay within its, you know, its four walls inside the platform. And and I'd say that's, you know, that's one of the the biggest reasons. Then the type of model that you introduce that data to, when it is a specialized model, it's really focused on doing that task. And and if that's the task you want that data to be interacted with, then you can have a little bit more control and governance over that.
Stephanie Felix: Right, totally makes sense. Cool. All right, JZ.
Josh Zywien: All right.
Stephanie Felix: Many technology leaders struggle with the execution gap—the distance between AI capability and measurable measurable business performance. How does this partnership with NVIDIA shift the paradigm for our joint customers when IT teams can build custom, domain-specific agents that run safely on one platform?
Josh Zywien: Yeah, I think the the challenge that we've seen over the last few years is we all use ChatGPT, or we use Claude, or Gemini, and we experience what these things can do. And it's magical, and it's it's broad, and and it feels like it can solve any problem. Then you go in the enterprise, and particularly with Workday, we deal with people, money, and sensitive IT data. These are things that run the enterprise, you can't get them wrong. And so you have this delta between what the model can do, what its—what the the potential of what what can be done, and then how it can actually be applied in the enterprise because you cannot the the the probabilistic nature of AI is you can't get this stuff wrong. So, payroll is always a great example. You can't get payroll 98.5% right. It has to be 100% right every time.
Stephanie Felix: Right.
Josh Zywien: You can run that across finance, closing the books, compliance, regulatory requirements, things like that. So I think closing the the gap starts with trust, knowing that these agents are going to do the right thing, follow the policies, permissions, rules of the business. And then I think number two is value. So it it comes down to ROI for the for the business. We all know that these these frontier models are are really incredible, can do really amazing things, but because they have this breadth of knowledge and this this corpus of of information, they have to—they're more expensive to operate. So for very specific, domain-specific tasks, it's actually more efficient to use these open-weight models for those things. And efficiency is good for us as builders of technology, but it's also good for the users and the the customers of it as well. They're the ones consuming it. The value proposition gets stronger for them while it gets stronger for us, too. So I think it's it's a win-win. It's not an either/or, as Erik said. I think it's there's there's lots of room and and need for these general-purpose models, but there's lots of work in an enterprise where a specific-focused model is is actually more valuable.
Stephanie Felix: Yeah. Yeah, Erik, I'm curious if there's anything that you would add, because we get a lot of questions in this category too, just about that gap of like actually realizing the value that you hope. Anything that you would add?
Erik Pounds: Yeah, I I would just say, when you when you have a problem that you're trying to solve, engage with the platform and and and build these a—use the agents that are there. And then and then now it's exciting with extend AI to be able to further build more specialized agents to get certain tasks done that may have been challenging in the past. I would say as the platform evolves, just motivating everyone to keep keep engaging. Platforms are going to keep involving and and build with these tools. It's amazing the results, you know, that that you'll you'll most likely see. And I think you both for integrating in the Nemotron open models, and it's exciting what you can build from from data to the harness to the models what you can build on on on the platform now with Agent Builder and the other tools in your role.
Stephanie Felix: Yeah. Yeah, awesome. Okay, then one more question, fun one, to close us out. So I'd like to hear from both of you on this one. But looking ahead over the next 6 to 12 months, as these open models and custom agents become a standard part of how enterprise teams operate, what's the biggest transformation you're most excited to see our customers achieve?
Erik Pounds: I'll dive in.
Josh Zywien: All right, go for it.
Erik Pounds: Really, it's the adoption of these of these specialized agents and agents working together. I know that you're enabling agents outside the platform to interact with agents inside the platform, and those teams of agents working together on our behalf under security controls is going to be really exciting. And and that that that growth is happening right now.
Josh Zywien: Yeah, totally agree. I think the thing that's exciting for me is our ability to open up the platform and allow people to work where they work. So, kind of working in Teams, or Slack, or Co-pilot. And and having the trust that doing that outside of Workday could do that with the reliability and and security that you would inside of Workday as well, is huge. And then I think for me, it's it's just the transformation of the work. We just have not seen enterprises truly adopt these things at scale, both because of the cost and the security—the reliability that these things are going to do things the right way. We've now kind of passed that frontier, and so I think we've we've been able to unlock that with the partnership with NVIDIA. And I think you're going to start to see more and more CHROs, CFOs be comfortable with this stuff operating on on really sensitive business processes and data. So it's what we see today is exciting; I think what we see in 12 months is just going to blow everybody's minds.
Stephanie Felix: Yeah, it's all moving so fast.
Josh Zywien: Yeah.
Stephanie Felix: Thank you both so much for sharing your thoughts on where enterprise technology is heading.
Erik Pounds: Thanks for having us.
Josh Zywien: Thank you. Yeah, thanks.
Narrator: That was Workday's Josh Zywien and NVIDIA's Erik on why the future of enterprise AI is open, specialized, and above all, trustworthy. The takeaway? The question was never whether your AI is smart; it's whether it's right every single time. Now we want to hear from you. Where could a specialized AI agent make the biggest difference in your organization over the next year? Share your thoughts in the comments. If you enjoyed this chat, please subscribe to The Future of Work podcast wherever you get your shows.
Report
A gap between effort and impact is the real ROI problem. Discover the solution.