As HR technology moves from AI-powered recommendations to agents that can execute multi-step workflows, the risks around employee data, privacy and accountability are becoming harder to ignore. A November 2026 report from employment law firm Littler notes that agentic AI can act across workplace systems, while governance frameworks are struggling to keep pace: only 13% of organizations surveyed believed they had the right governance in place. The report underscores why access controls, human oversight and clear accountability need to be designed into workplace AI. 

At HR Tech in Las Vegas, RPI Tech Connect host Chris Arey speaks with Melissa Olson and Jackie Dudas about what they heard on the conference floor. Their conversation moves beyond the latest AI demonstrations to the questions HR and technology teams need to ask: what data is feeding the system, what can an agent access or disclose, how should its behavior be governed, and where does human judgment still matter?

In this episode, Melissa Olson and Jackie Dudas share their takeaways from HR Tech 2026, including why clean data, shared definitions, security controls and careful testing should come before broad AI deployment. They also discuss the importance of looking beyond vendor demonstrations and keeping the employee experience human.

This series was recorded live at HR Tech 2026 inside the HR Executive studios on the expo floor in partnership with the WRKdefined Podcast Network. 

⁠Make sure you’re subscribed to the full series⁠ and ⁠visit ⁠⁠HRExecutive.com⁠⁠ for the news, analysis, and insights shaping the future of work.

Show notes

What should HR teams check before adopting AI agents?

HR teams should understand the data, permissions and governance behind an AI tool before allowing it to interact with employee information or carry out workplace processes. The discussion highlights how conversational AI can introduce risks that are easy to miss in a polished demonstration, including employees entering sensitive or unrelated information, agents exposing details they should not disclose, and inconsistent data producing unreliable answers.

Melissa points to the work that happens behind the scenes: cleansing and normalizing data, agreeing on organizational terminology, and applying consistent metadata tags. These foundations help an AI system retrieve the right information and operate within intended guardrails. Jackie raises a related concern about how much an employee might share in a conversational exchange and what the system could learn or reveal across the organization.

The conversation also makes a case for asking vendors what actually powers a branded AI assistant, rather than accepting its name or demonstration at face value. Organizations need to understand the underlying software, security model and data protections. The practical approach is to define the business outcome first, then prototype, test and verify the technology before expanding its authority.

Key takeaways

- AI is becoming a standard part of the HR technology conversation. Melissa describes the conference message as an evolution from AI-first and AI-native software toward agentic capabilities. Jackie observes that AI use has become so widespread that, in conference sessions, nearly everyone raises a hand when asked whether they use it. The challenge for HR teams is shifting from whether AI is relevant to understanding where it belongs and what it should be allowed to do.

- Conversational AI creates new questions about employee information. An employee might use an AI assistant to begin a leave process, but the same open-ended conversation could invite unrelated or sensitive disclosures. Jackie asks whether information shared by one employee could influence responses elsewhere in the organization, or whether an agent might reveal confidential details such as a pending promotion. Those risks make data access, disclosure rules and governance essential design questions.

- Data preparation is a major part of AI implementation. Melissa stresses that organizations need to cleanse, normalize and rationalize their data before expecting AI to work reliably. Teams across the business also need to agree on terminology and metadata tagging so information can be found and interpreted consistently. This is not simply an IT task: shared definitions across departments help improve search, analytics and the quality of AI outputs.

- Vendor demonstrations should be followed by questions about what is under the hood. A polished AI assistant may have a memorable name, but buyers still need to understand the actual software behind it, how it accesses data and what protections are in place. The conversation points to questions about security architecture, personal information, governance layers and safeguards against unauthorized access. HR and IT teams should evaluate the underlying system, not just the experience shown on stage.

- AI adoption should be deliberate, and it should not erase the human experience. Melissa advocates moving slowly, prototyping and testing before giving AI broad access or responsibility. Jackie reinforces the need to identify the outcome an organization wants before choosing a tool. Both also recognize that technology should improve employees’ experience, not make human support harder to reach. Automation has a place, but organizations need to decide where a person should remain available.

About the guests

Melissa Olson is Executive Director at RPI Consultants. She works with organizations on Infor technology and workforce-related challenges, and has appeared on RPI Tech Connect in previous conversations about skills-based work and workforce planning.

Connect with Melissa on LinkedIn

Jackie Dudas, Principal Consultant at RPI Consultants, joins Melissa for this special HR Tech conversation. The discussion draws on her observations from the conference, particularly around employee data, AI governance and the questions organizations should ask before adopting new technology.

Connect with Jackie on LinkedIn

Connect with WRKdefined

Website · LinkedIn · TikTok · Facebook · X · Substack

Powered by the WRKdefined Podcast Network. 

[00:00:00] AI can do these transactions, but what they're not showing is all the work we have to do in the back end. Data cleansing, normalizing, rationalizing all their data. It's kind of interesting how ubiquitous AI is at this point. We'll be in a session and the speaker will ask how many people are using AI and everybody's hands go up and it's like at this point it is the new normal. When you have that really good, clean foundation, the AI is going to excel.

[00:00:27] I think companies should not take for granted that what they're offering is what their employees want and they should constantly see what is out there that they can offer. Take to retain employees. Hello and thanks for tuning in to a very special edition of RPI Tech Connect. We are live on the floor of HR Tech in Las Vegas.

[00:00:57] Having lots of fun here, lots of great discussions. And today I'm joined by none other than Miss Melissa Olson, who is a returning guest to the show. And for the very first time on such a grand stage, Miss Jackie Dudas. McCrack goes wild. Big round of applause. Thank you both so much for joining me today. How are we doing? Doing good. Spun time. This is my favorite week of the year for content, for people, networking, for a lot of things.

[00:01:25] Because I'm usually delivering presentations. So to be able to sit back and learn. Excellent. Fun. Well, I appreciate both of your time this afternoon. I want to, the focus of this conversation is really just to kind of hear about the last 48 hours here at HR Tech. So first question is, was there anything that you heard in the last two days that either surprised you or that you weren't expecting? Melissa, we can start with you. Um, no.

[00:01:56] No surprises. No surprises because, you know, we've been getting the AI message for a couple of years now. So it's, again, an evolution of that same message. It's seen a lot of the software that, you know, they were already building AI first or AI native. And now what's next? Agentic. It's, yeah, we're seeing some of that. But I don't, so far, have not been surprised by a gay thing. So that's good. It's a good thing.

[00:02:25] This is also your fifth year in a row at HR Tech? It is. So you're in the know, in the scene. Yes. Try to. Very good. Jackie? I think it's less of a surprise once you sort of think about it a little bit more. But it's kind of interesting how ubiquitous AI is at this point. Like, we'll be in a session and the speaker will ask how many people are using AI and everybody's hands go up. And it's like, at this point, it is the new normal.

[00:02:54] And so, you know, at first, in that first session that I saw that happen, I was surprised by it. And then it's like, of course, that's the case. I use AI on a daily basis. So there's really no surprise there, more you think about it. Yeah. That's a really great point you make, too. So, like, this is my first HR Tech event. As a marketer, I've attended some other events this past year. And, like, AI is a theme no matter what industry you work in. Like, it is absolutely 100% a part of, like, everyone's job.

[00:03:22] If we were here in 1998 or 1999, everything would be about ACML, Web1, yeah, the Internet, all of that. So, you know, I think we'll be hearing this for another five years up east. This is the only message. Very true. There's a lot of talk about what AI agents can do.

[00:03:46] Have you heard anything or heard anyone say that AI could do something that you would need to see to believe? Jackie, we can start with you. Well, not necessarily. I have questions about what else AI can do that maybe, like, there is no intention for AI to do. I was in a session earlier, and they were explaining how easy it is for an employee to start, you know, for example, a leave process and get themselves out on FMLA.

[00:04:15] And it's basically a conversational exchange between AI and this employee. And I wanted, I had so many questions pop into my head about, like, what else can an employee feed AI that might be not work-related? And if the AI is learning what, you know, from the employee, from all the employees in the organization, can AI start rumors?

[00:04:39] Can AI begin to change the landscape of the workplace, you know, of an employee? Like, because in so many places when we implement HR, if an employee has a place to enter free-form text, a lot of our clients say, no, no, no, we really need to lock that down. And if they're able to have this free-form conversation with an AI agent, what, you know, what's in place to make sure?

[00:05:03] So I would like to see what's in place to make sure that it's not sort of like a free-for-all of information that gets injected into the system. So I will say, if the software is doing what, if they know what they're doing, there should be a governance, sure, layer under D. And that's what I would like to say. Right. And, you know, you've probably heard the MCP term thrown around a little bit in model context protocol. And usually it's in that layer. They are using MCP.

[00:05:32] There is some layer of data security, governance, but you're right. Trust but verify. Sure. Actually, right now we're at the don't trust but verify. Maybe we may trust. But I think everybody's kind of aware of that security side. But it is a Kengo robe. Sure. Seeing this already in some of the frontier models. Sure. So, you know, it's a valid point.

[00:06:01] We don't know how it will act. Nobody, we don't have a, you know, roadmap from somebody else that has done this already. So, you know, who's going to be the one? I mean, there's a lot of large organizations out there that might be the ones that will hear the news that it's gone rogue. What's going on? Yeah. Yeah. You said governance. I thought of governance. I also thought to cue the Terminator 2 music. Because if AI is, you know, you make a great point about you're able to give it so much information.

[00:06:32] Who's going to stop it from doing something and maybe it shouldn't? Or, I mean, even something as simple as one of the examples today was I want to promote so-and-so. And if there's another manager who says, you know, I'm about to do my review with so-and-so, would AI respond and say, well, they're about to get a promotion? Would they let the cat out of the bag? And if AI is learning from the inputs that it's being given, will it disclose information that it maybe shouldn't disclose in other organizations?

[00:07:02] I don't know. So I think that what the piece, it's like the cooking shows, right? It's like AI can do these transactions, but what they're not showing is all the work we have to do in the back end of data cleansing, normalizing, rationalizing all the data. And then everybody in the organization having to agree on the metadata tagging.

[00:07:27] I'm going to get super techie here, but I feel like that is the stuff that we're not hearing or I haven't heard in a lot of the presentations is the hard work around getting your data to a place and getting everybody in the organization to agree on what the vernacular is.

[00:07:48] So we'll go into the metadata tagging so that you can train your AI and you can put guardrails in and it will act the way you intend for it to act.

[00:08:02] So that's the, again, the stuff that a lot of people's eyes glaze over on, but it's also, that's an important question to ask a lot of these vendors or, you know, we work with Infor and I really try to start understanding how they're working on that,

[00:08:21] really understanding the AWS bedrock foundation and how that secures data and how that protects Black Hat from coming in, Black Hat agents for coming in and making PII personal data and pushing it outside of an organization. There's a lot of scary stuff out there. So I will, I will still always be that person that's the doomsdayer, but I'm super excited about AI.

[00:08:48] I love this stuff, but I also want to be realistic about it and let's, let's go slow and prototype and test and verify before we decide let's do it all and give it all to AI. Like, so. You raised some really great points there and it reminds me of some of the themes and like overall messaging that I've picked up on in the sessions that I've attended.

[00:09:10] And it was so focused on like having organizations take a step back to think about what is the outcome that they want before even entertaining the tool. Like the tool, think about what it is you're trying to solve for before you bring the tool into the equation.

[00:09:24] And I think that data, data governance, data quality, data hygiene, as well as like the vernacular that you mentioned, like having agreeing on what terms you're going to use, how you're going to use them is such a critical step that like it's so easy to just rush in and want to, you know, explore and take advantage of all this new stuff. Like you do have to be diligent and move slowly and be, and, and act with caution. Cause like you said, this is very powerful stuff that we're dealing with. And it shouldn't just be IT messaging all of this, right?

[00:09:54] The vernacular has to be agreed upon by every area of an organization. 100%. That way, when you're doing searches, when you're asking it to bring you step back, it's tagged correctly. And you know the words to go and do your search and you know how to tag all those little pieces of data. So then it comes back and your analytics also, I mean, it's hard to do analytics, but like.

[00:10:17] That's important work that is worth the effort because when you have that really good, clean foundation, the AI is going to like excel. Yeah. Yeah. That was actually, I don't know, maybe this is one of your questions, but one of my favorite sessions that I've been to so far is the closest thing to like a cautionary tale about AI and how important having clean data. And to set your AI system up for success, how important that is. You can't just feed it.

[00:10:43] And I took some pictures of the presentation about the stats of like, and I would get them all wrong if I tried to regurgitate them now. But the stats of how much it tells you that's actually wrong, not factually correct, and unless you give it a clean data set to work from or some guidelines to work from. So that's the most important part too. And I don't. Story is old as time. Dirty data. I mean, it's true. It really is.

[00:11:12] So that actually wasn't one of my questions, but I'm glad you brought it up. And now I'm going to ask you any favorite sessions in the last two days? Um, oh boy. All of them? No, I don't know about. Well, you like the ones where you can see what's actually happening. I do. Or I like a lot of the thought leadership ones. So there have been a few, but there, I feel like there haven't been as many this so far yet.

[00:11:41] And also, you know, honestly, I also have. Yes. Meetings with vendors and some client work. And I haven't been able to go to everything yet, but I still have some hope that I'm going. Tomorrow is another day? Yes. There's much to be done. Yes. Okay.

[00:12:04] Um, on the note of sessions, um, one of the great things about sessions at conferences like this is that there's a lot of engagement between the attendee, the audience, and the speakers. Were there any questions that people asked that you were like, man, that's a good question? Maybe you had one of your own? Um, I, there was a question where, um, there was a presentation where the AI was really at the forefront.

[00:12:31] It was explaining all the great things that the AI can do, which of course, you know, that's, that is the theme here. And it's, it is really wonderful and really cool to see everything it can do. But the question was, what is behind this AI? What is the actual, like, you've given it a name for your company, like a, you know, a catchy little name so that the employees can refer to it. But what is the actual software that's behind it? And so I liked that question because I, you know, I want to see under the hood a little bit. I want to see what we're actually talking about here. So I liked that question to actually find out, you know, you've given it this pull name, but what is it really? Yeah.

[00:13:01] You mentioned the naming thing. And this is a session I sat in and on that was a case study. A customer was talking about the implementation process of an AI tool and part of their, like, internal marketing strategy for, like, increasing adoption and helping with change management was actually giving it a fun name to make it seem more human and personable and, and, like, actually humorous. And so it's, I like that. I'm glad you bring that up. Thank you. Um, all right. So. Anthrop, I guess that's animal.

[00:13:32] Anthropomorphize. Yeah. Yeah. You know, we, we really, and I think I did that when ChatGPT first came out. I tried to give it a name and I tried to, people always call it chat. Still say, please, can you do this? You have to. And they don't remember that later. Yeah. So I think, yeah, that, that branding is like people need to still, the human is out sometimes, but we need to humanize. Yeah. Yeah.

[00:13:57] And then, again, where's, where's that, uh, inflection point where we decide we're not going to ask human resources, not agent to, agent to size every process. Because I get really frustrated when I try and get in touch with a company because I have a return. I have something, a question, and I have a series of chatbots and I can't, even Amber, you can't even talk to a human anymore.

[00:14:26] And that, that becomes very frustrating. So, you know, we're here at an HR tech conference and we're talking about all the tech, um, and, and a lot of, uh, a lot of organizations are, are trying to talk about the human aspect of it. But, you know, I hope that just doesn't get forgotten. Absolutely. In here. So. That's another big part of what I've been hearing this week. It's like the human element is still just as important as it always was. And we need to remember that. Mm-hmm. All right. Uh, last question here for, for you two.

[00:14:55] Sum up HR tech in one sentence. Jackie, go. Wow. One sentence. I, one word would be better. Okay. Sure. Don't say AI. I was going to say overwhelming, but it's overwhelming, not in a negative connotation. I think it's overwhelmingly good. There's a lot of positive out here. Uh, you know, also working from home. It's, it's, you know, not every Wednesday is around like a couple thousand people.

[00:15:22] So, um, I think it's, you know, a lot of information, a lot of inputs, sounds, noises, colors. People, you know, it just is over. It's Vegas. Noises. I, it's really wonderful. And I, it's so nice to connect with people again and be at a con, I haven't been in a conference in a few years. So overwhelming, overwhelmingly good. I think. Let's go. Let's go. I like the energy. Melissa, same question.

[00:15:44] Um, I feel like, uh, innovation is a word that comes to mind for, for this because when I come around visiting the, uh, booth, the vendor booths at the expo, it's things that I would have never even thought about. And whether or not AI is infused into it, just offerings and services for employees that have evolved with what employees expect today. We don't just all need the generic.

[00:16:13] We'll give you a reimbursement for your gym membership. And we will give you, you know, your HSA, which there's that here too. But there are also just really cool offerings that I'm like, yes, if my company offered this, that would be a differentiator for me to go to, to this company versus this company.

[00:16:32] And I think that, that even if AI is not a part of it to still have that creativity and innovation around how to make employees lives better where we're spending most of our waking hours. Um, I think, I think companies should not take for granted that what they're offering is what their employees want and they should constantly see what is out there that they can offer, um, to retain employees. So I really like that.

[00:17:02] And this, and this is a great place to explore that. Yes. Yes. Thank you both so much. And for those of you tuning in, this is RPI Tech and I will catch you next time. Thank you. Thanks all. Thank you.