Podcast: Why agentic AI is revolutionizing enterprise asset management

Key takeaways

Agentic AI enables autonomous decision-making, acting as a digital coworker to boost safety, efficiency, and productivity.
AI-driven incident reporting improves compliance by detecting unreported safety issues and prompting preventive measures.
Multimodal AI tools support frontline workers with voice and chat, bridging skills gaps and enhancing field safety.
AI co-workers optimize maintenance, inventory, and inspections, shifting plants from reactive to proactive operations.

In this episode of Great Question: A Manufacturing Podcast, Thomas Wilk is joined by Chris van den Belt and Berend Booms of Ultimo, an IFS company, for a conversation on the rise of agentic AI in enterprise asset management. The discussion explores how AI is moving beyond traditional copilots to become autonomous digital coworkers that enhance safety, streamline maintenance, and support frontline workers in dynamic environments. Together, they highlight real-world use cases, from improving incident reporting to optimizing preventive maintenance and inventory management.

Below is an excerpt from the podcast:

PS: Hi everyone, and welcome to a new episode of Great Question: A Manufacturing Podcast, brought to you by Endeavor Business Media’s Manufacturing Group. I'm Tom Wilk, the chief editor of Plant Services, and today we have with us two returning guests from Ultimo, an IFS company. We have Chris van den Belt, who's the head of product management, and his colleague, Berend Booms, who's the head of EAM Insights. I'm going to let Chris and Berend tell all the listeners why we're here today. So Chris, something very exciting happened this week.

CB: Thank you, Tom. Great to be here again—absolutely. So, to introduce myself, I'm in product management, focused on the roadmap of Ultimo—the product strategy—and, indeed, exciting times. A lot is happening in the AI area. My job is to keep working on the product, developing the product together with developers, and our chief architect who calls me twice a day to show cool stuff; and ideate together with customers to come up with great solutions where AI is really helpful. And indeed, we have launched our first agentic AI use case. But let me first hand over to Berend.

BB: Yeah, thanks for that. Like Chris, I'm also super happy to be back—thanks for having us, Tom. So, my name is Berend, I'm the head of EAM Insights over at Ultimo. And really the reason that we're joining you today is because we think that we've found the start to this new phase in enterprise asset management. At Ultimo we plan to lead the way as category leader. What we're doing is putting a lot of emphasis on AI, and specifically agentic AI, right? And we recently—as of this week—reached our first big product milestone, a big release, and we're here today to share how a shift in working alongside AI can really help organizations to work in much smarter, safer, and faster ways. And how AI can start taking care of routine tasks so that human expertise—and human experts—can be applied where it matters most, and where they're needed most.

PS: It feels like we're getting to the point in the application of AI where we're starting to see some real, tangible changes in the way people are going to do work. There's been this promise for three, four, or five years now, but use cases like the one that you describe this week, and which were outlined in an Ultimo press release, really are pushing the needle.

Before we get to that specific innovation, let's talk about agentic AI itself. I'm sure a lot of people listening know ChatGPT—it's a flavor of AI, generative AI. So for those who might be new to the topic, or just want an overview, how would you define “agentic AI”?

CB: Let's assume everybody knows ChatGPT: it responds to a prompt the user initiates—you ask a question and you get a response. Agentic AI, on the other side, is an AI system that acts autonomously. It can make decisions and take actions to achieve a goal without constant human interaction. So in that sense it’s entirely different from ChatGPT, for instance, where the user take the initiative. It can plan; it can reason through problems; it can execute multi-step tasks independently.

Let me give you a couple of examples. You can think of a digital home automation agent that manages your lighting and heating by learning from your routines, or a digital coworker that manages your calendar, rescheduling when conflicts pop up. So to summarize: it’s autonomous, and it takes initiative to achieve a goal. I think that's the best way to describe agentic AI.

PS: What's new about applying this type of AI to enterprise asset management? You know, last November the three of us talked about how CMMS and EAM systems in general helped companies move toward more proactive maintenance modes. But this is a different order of technology coming in. What specifically is new about this?

CB: It’s the approach—how AI is applied. Let’s first take one step back. The challenges we try to address are basically the same. It’s still about knowledge loss; balancing corrective and preventive work / proactive work; and increasing safety. But the approach is different.

So, first of all, we're talking about AI‑powered features—copilot features inside the user interface of a software product. They provide smarter tools to the workforce and make them more productive.

But agentic AI, on the other hand, is another way of thinking: it facilitates autonomous decision‑making in asset management. You could see it as digital co‑workers who work alongside human teams to augment their skills. So in that sense, it’s a different approach.

BB: And I think it's also exciting to see that this is not just about the technological advancements—even though those are super impressive. It’s also very much about a new way of working, right? Agentic AI introduces this new interaction model, if you will, where instead of sitting behind your desk, behind your computer, clicking through screens or reacting to dashboards, your frontline workers are now going to start engaging with AI in much more natural ways, almost multimodal ways, if you will, where they choose a means of communication that's familiar to them and feels comfortable. That’s going to give them support where and when it matters most. For them, that’s going to be in the field, in the heat of the moment, and on the move. It’s not static; it’s super dynamic—and that's fantastic. The last thing you want when you're out and about doing your maintenance work is to be bound to a workstation for your insights.

Imagine you're in a highly demanding environment where safety procedures dictate that you always work in full safety gear—perhaps you have a special suit on, or you have protective gloves on your hands. Typing on a PC, or on a mobile, on any device, really, is very difficult to do. So activating AI‑fueled insights through voice suddenly becomes interesting.

Or another example: you're out on a big oil rig. Your hands are constantly too dirty to really handle some of the devices. That multimodal way of working isn’t only great for these folks; it's great across the board, because you're suddenly able to work in a way that's familiar, that's safe, and that works for you—instead of being told how to work.

And another thing: this helps bridge the skills gap—some of the workforce challenges that we're seeing on both sides of the spectrum for workforce challenges. You have a newer generation joining the workforce that is digitally ready—they’re born digitally native, almost—and they have high demands and requirements when it comes to the way of working, and this is very supportive of how they like to work. On the other side of the spectrum, you have the more seasoned veterans on your maintenance teams. They have a lot of this archaic, tribal knowledge, almost—but they might be less digitally savvy. So you're addressing both sides of these challenges: the difficulty of finding skilled labor, and having a new generation joining the workforce, opening up that knowledge to the central workforce. Agentic AI is just a very big enabler, I find.

PS: Interesting. This reminds me of a survey we did with our readers on electrical safety best practices at their plants, where one of the questions we asked was: Do you have a response plan? Do you have an incident‑reporting plan? Near‑miss reporting? Obviously the answer wasn’t going to be anywhere near 100%, but I was surprised at how few actually had incident‑reporting plans. One of the things that your release makes clear about this technology is that it's designed to help drive improved incident reporting across several dimensions—event descriptions, staff involvement, injury details, things like that—and the AI sits there and scans for this information, right?

CB: Yeah, absolutely. Incident reporting is indeed a very important step toward compliance, but even organizations that have that process in place are dealing with under‑reporting, because safety incident registration heavily relies on manual incident reports. So whenever something happens—let’s give an example: a forklift hits a machine and the machine is damaged. What often happens at that moment is that the technician or an operator reports a failure or submits a work request to the maintenance department to repair the damage. But they don't even notice that it's actually a safety incident that should be reported in order to be able to take preventive measures.

That is exactly what this AI use case solves. It scans incoming work requests, and when it detects safety‑related issues, it autonomously reports a safety incident. It ends up in the same list of human‑reported safety incidents for approval by a safety manager, who can look at it, process it, and flag it as a real incident—or maybe a false positive—but it gives you the possibility to define preventive measures, and that is something that we have recently released.

We’ve turned it on for a couple of customers already who are willing to test the feature and provide us with feedback, and what we already saw is that quite a few incidents have been reported. Some of them were false positives, but others were not—weren't reported by humans—and preventive measures have already been taken, so I think that demonstrates the added value of such use cases. And I think it's a very natural way to get familiar with agentic AI, because it's something that we do on top of the process that's already in place—manual incident reporting. It doesn't replace it; it's an addition to what we already do, and the human is still in the loop, and even still in control.

BB: And it's also a great example, if I may add, of human‑digital collaboration. The fact that the AI agent is able to report the incident, but then you still have the human in the loop who analyzes and validates the reports—and some of them are false positives—but based on the input, safety measures are now taken. That's the perfect example of how man and machine come together to achieve better results.

PS: Let me ask the difficult change‑management question, which is: how do you get people accustomed to working with this technology? The machines is not the problem—the machine will still, at this point, do what we tell it to do. How comfortable do you sense that people are with this kind of change? Because we're not talking about a simple change; we're talking about bringing in an AI product which on some level will be responsible for health and safety—and that's the number‑one top priority in plants, is that side of the business. What's your sense of that, Berend?

BB: It's a fantastic question—a fair question to ask, Tom, and I think one of the most important questions. For me, the key lies in building trust, right? How you build trust—how you do that gradually over time—is going to make or break the success of this particular functionality and its application.

Really, if you think about it, you're suddenly working with this new digital coworker. But is that really that different from working with a new intern, let's say? Because when a new intern joins your organization, you don't throw them into the deep end. You monitor them closely. You feed them information. You give them the context they need. Plus, you're supervising them—you're checking their work, double‑checking their work sometimes. And that's because at that step, you don't have that trust. You don't fully trust their judgments; you don't fully trust the insights and the things they come up with.

Over time, as the intern learns, they improve, and they start making smarter decisions. And they also get more responsibility in return—because you trust them more, right? Eventually that intern grows up to be a junior colleague, a senior colleague, maybe even a trusted expert within the organization. And with each step you hand over more autonomy and responsibility because this person has proven themselves. That's exactly how I see the role of agentic AI in asset management. In the beginning it will be more like an intern—gathering data, making suggestions, and learning on the job from all the contextual information and the input you feed it. But you, as a human, are still in the loop. You're still in the driver's seat. You are making the final calls.

As this system we've designed observes more information, it learns your peculiarities, and every organization has them, it starts offering more consistent and more valuable insights. It's going to move up the ladder, from an intern to a junior, to a senior, and ultimately to more of an expert level of augmentation, of support. I don't think organizations are ready and willing—not just yet—to hand over some of that critical decision‑making to a digital coworker, at least not on day one, right?

But what we're doing is building AI that is going to earn trust over time. Because it is helpful; it is reliable; it is always available, right? Never gets sick. And you have this—as Chris called it—a human‑in‑the‑loop model, so you still stay in control, but you don't have to shoulder the entire workload by yourself. You're now suddenly supported by these digital coworkers.

PS: You mentioned there are some people trying this out, and I'm not going to ask for names, I know beta testing is a private matter. What's the response been from some of the teams that are using this product? Are they using it for the initial use case? Are they liking the results they're getting? Are they asking to have it applied beyond that first case to a different case?

CB: So first of all, we've been working closely with a couple of customers who co‑develop features with us. In terms of innovation, people often give you the advice: don't ask the customer what they want, because they'll only tell you what they already know—it doesn't help you innovate.

The approach we took is, we organized a customer panel, and we had some Teams meetings with them. But also, the chief architect at Ultimo and I visited those customers on site. We talked to a lot of people, seeing with our own eyes how they currently work. The way we collaborated with them is that we inspire them by showing things that could work—showing new technology, the possibilities of that technology—and sparking their creativity and ideas to come up with the best use cases, the greatest use cases.

One use case we've been working on, for instance, is to give people a daily maintenance briefing—a heads‑up at the beginning of the morning. When you wake up, go to work, or open your laptop, you want to know what happened during the night shift, for instance, and what happened the last day—to get up to speed, to become more proactive that day. Of course, you can go through all the shift logs that contain a lot of information, but maybe only 10 or 20% is relevant for the maintenance team.

What if we could come up with a daily maintenance briefing with the highlights—the biggest disruptions, or maybe recurring/emerging issues that popped up that night, that night require your proactive action? What we basically saw was that they almost got addicted to it in a couple of days, because when we stopped sending those briefings—which were basically an experiment—we got the question: Hey, where's that summary? In the morning meetings they already started missing it, because it contained more and more valuable information than what they got from their coworkers in the normal way.

PS: Wow, that's fascinating. And as someone who works in the media, I completely agree that sometimes your audience—or your users—don't know what they need. And then you provide it and it's like, Oh my gosh, how did we live without this?

CB: Yeah, that’s exactly what happens.

PS: So guys, we've talked about safety and maintenance as use cases for this technology, for agentic AI. Chris, could you talk about some other use cases that you've seen or that customers are requesting?

CB: Yeah, absolutely. Let's divide them into two different areas. One is: how do we get from reactive to proactive? And the other area is, how do we see digital co‑workers working alongside human teams to basically augment their skills?

From reactive to proactive—what we often hear when we talk to maintenance managers, maintenance engineers, planners—is that they react to disruptions; that they're stuck in the day‑to‑day mode; and that they spend time on reactive tasks and struggle to get from that reactive mode into a proactive mode where they analyze what happened—sit back and see what actually happened—and see how they can improve. And agentic AI is an enabler for us to turn that around: to provide insights proactively and to enable humans to collaborate with digital co‑workers. Let me give you two examples.

One is the optimization of preventive maintenance plans. The next example is inspections: they are added after finding issues, but they rarely step back and see if they're still needed—if they still lead to defects that will be discovered during inspections. Imagine a digital co‑worker with a goal to eliminate ineffective inspections and focus preventive maintenance where it really adds value. It could analyze the data, take action, and give proactive insights to the one who needs to approve it and get rid of (in this case) ineffective inspections. Again, the user is in the loop to handle the suggestion, and over time the suggestions can be processed automatically, where the user intervenes afterward when a decision shouldn't have been made.

A similar area is inventory levels. What we hear quite often is that there is no systematic or structured or data‑driven approach to optimize inventory levels. People basically say, "The level of overstock can be seen by the level of dust on the shelves," and "Yes, we also face stockouts when we face downtime." So there's huge potential—in cost savings and in downtime reduction—but where do we start? Do we have the parameters? Do we have the data available to change it? Again, a digital co‑worker could gather the required data and could come up with suggestions to fine‑tune minimum and maximum stock levels—again by analyzing data, taking actions. Basically the goal of that digital co‑worker is to reduce excess stock and ensure critical parts are always available.

PS: You know, Chris, it's funny: we just finished a podcast recording that focuses on business uncertainty at the moment. That's one of the drivers we're hearing from people in general—any technology that helps them manage their inventory and optimize it during extra uncertainty is really helpful, even if it just means looking for the best price or budgeting for potential shortfalls. You know, this use case is out there for sure.

CB: And the second area is, how do digital co‑workers work alongside humans to augment their skills? Let me give you an example of what it could look like. A machine breaks down and, instead of forcing an operator to log into a system, fill in a form, and submit a work request, the operator uses chat—or maybe even voice—to report the failure. The system automatically detects the urgency by asking questions, or by looking at the criticality of the asset, for instance.

When it is urgent, a planner agent could automatically assign the technician on duty. It could message or even call the technician and try someone else if he’s not available. When the technician starts working on the issue, of course the level of experience will depend. If it’s a less‑experienced technician who fails to find the cause, he would often call this very knowledgeable, experienced colleague who knows everything, who could help him repair the machine. But instead of doing that, a troubleshooting assistant could support by providing data—because it has access to the failure history, and even the failure history of similar parts or similar assets at other sites. Then, when the solution or direction is chosen, a warehouse agent suggests a replacement item if the original is not in stock, and when the technician has fed the solution back into the system, a reliability agent could then evaluate whether the PM plan should be updated.

So, it’s a combination of agents working together—agents triggering tasks for other agents—and actively communicating, proactively, with the humans on the floor.

PS: Let me close with this question, and maybe I can direct it, Berend, toward you. With all these innovations occurring, what is next? What’s on the horizon for this kind of technology? We’ve heard the use cases. How do you see this evolving in the long term?

BB: It’s a good question. If I look at the future, one thing I'm sure of is that we want to do things the Ultimo way—which for us, it comes with the vision, comes with the mission, and it starts with: where is the impact? Where is the impact going to be the biggest? How do we provide value in the fastest amount of time? And how do we grow from there? How do we build out the trust and build out the functionalities this way?

So really, what we're working with now is a stepping stone for us in a much bigger picture. It’s first safety—but the application of these agentic agents, agentic AI, applies to a much broader asset‑management spectrum, right? It could address challenges in, as you said rightfully, inventory management. It could support work‑order planning. You could do asset cataloging—smart asset cataloging—where you simply walk around the site, take inventory, you have your assets cataloged. It’s those kinds of things that I think are now within the realm of possibility.

And in all of these developments, critically, we’re tackling some of the biggest challenges in the industry. The foremost challenge for me is that loss of expert knowledge—all that tribal knowledge that’s about to walk out the door because of the median age of your workforce that are about to retire, and you’re losing that knowledge. Now, with these AI counterparts—or AI co‑workers, let’s say—we’re making all of that expertise accessible not just to someone in the department, but to everyone: junior, senior, current, or future employees. You’re really creating what is knowledge equity, for me, across teams.

When we develop what we develop, and as we move toward the future, it’s about building trust. It’s about not replacing people but augmenting their skills and creating a baseline level that is unprecedented—much higher than what was deemed possible before. And thereby we’re going to empower the broader spectrum, the entire workforce, with all the insights at the right time, at their time of need, delivered in the most convenient way, so they can excel at what they do best and apply their human expertise where it matters most.

About thePodcast
Great Question: A Manufacturing Podcast offers news and information forthe people who make, store and move things and those who manage and maintain the facilities where thatwork gets done. Manufacturers from chemical producers to automakers to machine shops can listen forcritical insights into the technologies, economic conditions and best practices that can influence how tobest run facilities to reach operational excellence.