Physical AI and the Autonomous Job Site with Caterpillar SVP & CDO Ogi Redzic

0:01Hi there, and welcome to Enterprise AI Innovators, a show where top technology executives share how AI is transforming the enterprise. In each episode, guests uncover the real-world applications of AI, from improving products and optimizing operations to redefining the customer experience. I'm Evan Reiser, the founder and CEO of Abnormal AI. And I'm Salmo Tamidi, a general partner at Greylock Partners. Today I'm talking with Ogi Redzic, Senior Vice President and Chief Digital Officer at Caterpillar. Caterpillar is a 103-year-old industrial company operating across construction, mining, and power and energy, with data flowing in from assets across the globe.

0:38What makes Augie's perspective interesting is that Caterpillar sits on both sides of the AI wave. It's supplying the prime power that new data centers run on, and it's putting AI to work across one of the largest machine fleets on the planet. A few things stuck with me from this conversation. First, AI is bringing new technicians up to speed fast. Caterpillar supports equipment going back 56 years, so they built an AI system on top of 50 ,000 documents that lets a technician point at a machine and ask a question and get the answer on their phone. What used to take years of apprenticeship now happens in the field, in real time.

1:09Second, the CAT AI assistant works in 55 languages and can coach an operator right inside the cab, so a first-timer knows a safety feature like eFence exists and how to use it. Instead of making customers learn a stack of apps, Augie's team flipped it. You just say what you want, and the outcome comes to you. And finally, Caterpillar shipped their AI assistant in 10 months, and a launch that size used to take years. Augie is clear about why it was possible. Building the Helios data platform and getting the data structured came first. AI speed doesn't come from chasing the shiny use case. It comes from the foundation nobody sees.

1:43Well, Augie, thank you so much for joining us today. I think everyone knows kind of about Caterpillar, but do you mind sharing a little bit about kind of your role there and for maybe like the one person in the audience that might not be as familiar with kind of the business and the company and the brand. Do you mind sharing a little bit about the organization? Sure. Maybe I'll start with the company first. So I joined CAD about eight years ago. And frankly, when I joined, I knew about construction equipment. Everybody knows about construction equipment. You play with toys when you're a kid. I knew about mining because occasionally you do see the large mining trucks driving around in some video or in a movie.

2:18but I didn't know about power and energy business, which is really important for us as well. So a lot of engines, we make engines, we make turbines, we make a lot of equipment that powers, oil and gas industry. So it's a really important segment for us and obviously one that's directly tied to AI because data centers are power hungry and you see these days that a lot of companies are buying primary power, not just backup power directly from companies like ours. So Caterpillar really operates in these three distinct segments is construction, mining, and then power and energy. What I also did know how amazing the company is when I first joined.

2:54I mean, you kind of know of Caterpillar, but you don't really know about what Caterpillar does. And then you see how impactful work is and how global it is. You know, that's quite impressive. And that's why we have people like me, we come in. I didn't necessarily think I would have a long-term gig at Caterpillar when I first joined. You know, you don't expect to stay in companies for 10 years when you join them. but you've kind of fallen in love with the product and with people and that's what I say. And maybe tell us a little bit about your kind of role today. I think you mentioned your backgrounds in tech and data.

3:25Like every company is a tech and data company and I bet you're most of the surprised I think how sophisticated you guys are in using tech and data inside Caterpillar. Maybe you can share a little bit more. Yeah, yeah. We're actually quite proud of what we built. I run a team called CAD Digital and within that organization we have, I'd say it's an organization of digital builders we have a lot of software engineers that's probably the primary job you'll see inside a purpose of cad digital is to build software and tools and capabilities services for our dealers and customers so i'm kind of market facing a role within caterpillar and my team is really focused to bring the best that we can to customers that use our equipment and machines every day and and to our dealers which obviously an incredibly important part of the ecosystem This team is about 3 ,000 people.

4:17We also take care of Caterpillar Marketing and Brand. So CAD brand, it's something that we have responsibility for. So something that we cherish. It's incredibly important. The company has been around for 100 years, believe it or not. We just celebrated our 100th anniversary last year. So that was a really big thing for us. And, you know, we expect to go strong for many years to come. So again, my team, you know, these 3 ,000 people that we have are really focused on building what I call first foundational elements. So this is data. This is, you know, our platform. We call it Helios. We collect data from about 1.6 million connected Caterpillar assets around the world.

4:56So that all kind of flows into this core platform. We also collect data from dealers. Think about invoices and machines. Think about work orders for repairs. Think about other information that dealers connect from how machines operating in the field, inspections, oil samples. That all goes into this core platform. And then we build a set of services in that platform, Helios, that kind of serve multiple applications. So it was one of the first things that we did in CatDigital was to bring all this together. Because I'm sure you talk to others. Enterprises usually have multiple siloed apps and underlying data.

5:35And that's what we have in Caterpillar. So our first effort was really kind of to combine all that into one platform that you can have a common set of services. Every time you say, I want to show an asset on the map, you can do it in the same way. There's no reason why different apps or different teams should do things differently. If you want to, you know, see a customer of Caterpillar, well, I should come from one central repository. It shouldn't be, you know, each team's responsibility to have their own customer database. a lot of your kind of engines and turbines are kind of what's helping close the gap between like everything being set up to like getting the data center online and so like how is that like how is some of these you know kind of like the trends and kind of the data center builds on how that's affecting your business and i'm also curious like how have you been able to like adapt and react to that so quickly to kind of you know meet the demands and um let me tell a little more about that yeah i mean it's a very good question i i think we're we're playing in the ai world both both on the kind of consumption of AI side, but also on being able to support the further growth and just the ability to do AI globally.

6:38As you mentioned, our turbines and our gensets are incredibly important for continued growth in this space. The demand is incredible. If you look at our public statements, you'll see that our backlog continues to grow. I mean, we have more and more demand, and we're addressing that by increasing our capacity. I mean, you really have little choice. There's a lot of demand. And we can and will be producing even more to meet that demand. What's interesting is that you could look at gensets, you know, 10 years ago, we were in data center business as well. But there was primarily being used for backup power because companies that were putting new data centers out there could rely on being connected to the grid.

7:19But that is no longer always guaranteed, as you know. And a lot more of the deals that we engage in these days will be about prime power. So when we're a primary source of power for at least a number of years before customers could connect to grid or until grid catches up with the demand. And this creates some distinct challenges, opportunities for the customers. They come to us and we help them address them in the near term. Do you want to share a little bit about just like what's the day-to-day experience like of the kind of Caterpillar employee, maybe someone on your team using AI? And like, how do you think about transforming how they work day-to-day and how has that changed kind of the average employee experience?

8:00We're now looking at doing entire projects, leveraging AI and completely different from the whole setup that we have for a vast majority of projects still. How will that play out? I mean, that's for me is a seismic change, right? That will change how we do things entirely. So, okay, so it's a big one. Now, how do we do AI in individual opportunities, right? So you will have, for example, our condition monitoring advisors. They used to have to have looked to, you know, seven different applications to pull data together. Now we'll give them AI and allow that data to all be merged for them automatically.

8:36Example from, you know, our team, the technology team that's working on AI autonomy. You're applying AI to physical world to make our machines do task autonomy and do bigger even autonomy all by themselves. So tech creates a world of opportunities, right? Opportunities for us, human machine interface, how you coordinate the work on a job site. I mean, all those are entirely new opportunities and there will be products or services around those that will change how work gets done in construction and in mining. I mean, we've had mining autonomy now for decades and have hundreds of millions of kilometers driven without any safety incidents reported.

9:22But that's in mine sites, right? Mine sites are somewhat isolated. You know, you primarily have your trucks moving material, being loaded and dumping materials in a known environment. You get to construction sites very different. So applying AI in the world of construction is a different challenge altogether. That's one of the examples where we're applying some of the partnerships, some of the, you know, things that we've learned and then applying in the real world to address those. And then, you know, one that's very dear to my heart is something we call AI assistant, where we're trying to really disrupt ourselves.

10:00I mean, when we say, OK, how are we using AI? That was part of your question to get better. Well, we have incredible applications we've built over the last, say, 10 years. Right. I have an application for e-commerce. We have an application for equipment management. We have an application for inspections. We have a number of very good applications. but if you're a customer do we want you to invest your precious time in learning all of our applications like i mean you can in your own life you want to use the right functionality at the moment where you need it you don't want to spend time learning the app it's not about learning the app it's about the outcome which you want to get out of an application so we looked at okay what is ai going to do for us and our customers and and we said well for one maybe we can actually get ahead of our customers having to learn about our digital ecosystem and put what we call Cat AI Assistant in front of them so that they could be prompted with the customer can be asking questions using language and getting answers when it matters but also we can be coaching operators inside the cab like how do I do something in cab of our machines you can actually coach him if he's not doing something properly, you can tell them how to modify that behavior.

11:19If you're sitting at your desk, you can ask, okay, what machines should I be fixing today? Where do I have issues? An AI assistant would tell you that. So rather than learning, you can get all that information today through our applications, but rather than going through menus and figuring out how to ask the right question, you can just verbally say that and get the answer. And you can do it in 55 languages and get this anywhere you are in the world. Again, the benefit of being global can help operators everywhere around the world. So I'm kind of curious, when you were developing CAT AI, how does the experience of the technicians and operators in the field kind of influence that?

11:57And maybe tell a little bit more about how AI is changing how they do work and how they use Caterpillar. Yes. I mean, think about some of the biggest challenges our customers have. really you can boil down to three labor shortage productivity and safety so how can we help them how how can what we do help all their employees come home to their families right so talking about safety first well we have a lot of safety features on our equipment but in order to to get full benefit of those features you need to understand how they operate and you should be able to to to know how to engage with them at the right moment as an operator.

12:40So on safety side, for example, we have something we call e-fence. So that will tell you that, you know, boom and a stick of your excavator cannot, would not go too far out because you're doing a job next to the highway and you cannot go over that. Now you can do that manually, right? So you can expect your operator never to do something, but we created something we call e-fence, which will actually physically stop you from going past certain points, also going up. So if you go up, you cannot hit the wires that are above you and cause some damage. The AI assistant then we said, okay, well, we want to make sure that customer, the operator that comes in for the first time knows that the feature is in first of all, that he can do that.

13:24And then two, get some guidance on how to operate it. And then three, actually do it for him. Have him just tell you, okay, I want to go not more than two meters because that is where I see these wires above me. So you just say it, and then it will not go above that distance. It sounds very simple, but actually is very handy for people that are new to this. Look at the technicians, for example. Every year in the U.S., about 40 ,000, 39 ,000 roughly technicians either retire or go do something else. So we actually have a number of technicians that are leaving the workforce every year that we need to somehow offset.

14:01And, you know, as you know, in your own personal life, I mean, you want somebody that's experienced. You know, if you want somebody fixing your car that's actually got it before, you don't want somebody who's doing it for the first time. So how do you get these new technicians to be as capable as possible, as quickly as possible without having, you know, the old apprentice system, somebody to actually learn from? Well, that's where AI system comes in again. I mean, you can actually summarize. I have our system for supporting technicians has 50 ,000 documents and it has millions of entries. And it goes back to equipment that's, you know, 56 years old.

14:39I mean, we go back all the way to 70s with the equipment that we support. So you go on a field and you need to service an asset. Oh, OK, so this asset was last produced in 1995. How do I fix it? Right. Well, if I can bring you the knowledge, hey, this is a machine. ask the question, how do I, and get immediate answer on it, on your tablet or on your phone. As a technician, you will be much more empowered. You don't have to search through menus. You don't have to go and look and look and look and then to find what you want to fix on that machine. You can actually say, how do I replace hydraulic pump here?

15:14You now, and this is more exciting and just an example of recent advancements, with glasses that are getting better and we think we can do so much more with them. You can actually just look at things and we will recognize where we see leakage of some material or we see a wire that's crossed or we see some physical damage. Those are the areas that need to go and get addressed. Here's an example of how to, with augmented reality, actually guide a technician through the repair. That sounded like science fiction two, three years ago, right? but now it's yeah i mean literally i mean the way i like to kind of guide my team on what what to do we sit down and say okay what are the use cases we want to enable for our customers based on what we hear from customers and what we kind of expect that they will benefit from and we create videos aspirational videos like we actually had videos for you know how we want operator to be in the cab We had videos for technicians in the shop and outside of the shop.

16:16And what's amazing, Evan, is that we did these videos like two years ago. So, OK, this is the experience we want to enable. And then it's already here. It's happening so quickly. The product itself, AI assistant for us, we put it out there in 10 months, which before was unheard of. And a company like Caterpillar would product launches of that size would take years. and ability to actually go quick, it only happens because we did all the work that I mentioned at the beginning around Helios and our core data. So we have our data structured and organized and we can put AI relatively easily on top of that.

16:55But then the use case is what actually brings the value to the customer, right? So just because we can do something doesn't mean we should do something. We should only do it if it addresses customer challenge of some sort or creates clear customer value. So we then pick these examples and go after them. What's in the lab today? What are you envisioning right now that's going to help our customers in the next couple of years? I think because I could go on and on and on because our list is long of actual ideas. But I do think for the audience here, the most exciting piece in my mind is what we're doing with physical AI.

17:34I mean, it is what's happening on our machines to make them autonomous. It is the orchestration that's happening on a job site that is going to make that more autonomous. Because when you think about it, a lot of regular construction jobs, there's like a lot of, I shouldn't call it waste, but let's say not productive time. When you're waiting for something else to be done or somebody misread a drawing and moved this material here and should have really been over there. So you got a way to move that material on that other part of the job site. Trying to remove some of that waste, and I'll call it waste, but idle time, let's call it, and replace and leverage AI to actually improve how that work progresses on a job site is a tremendous opportunity for everything to go faster.

18:23Because, you know, again, you open with, there's a lot going on in data center business, and there's a lot that needs to happen from permitting to the actual construction to getting the energy, putting in production. If we can help fix some of these bottlenecks along the way, everything will and can move much faster. So for me, one of the biggest opportunities and that we're going after is really making our machines more autonomous so that they can do tasks themselves, making the human-machine interface more easy so you will not have necessarily operators with joysticks and moving things around from within a cab.

19:01You may be even commenced to the asset from outside. You may have some remote control, which we already have. And then you'll be about coordination between these different assets. So, hey, excavator, go dig me this trench that's two meters deep and three meters wide and five meters long. And let me know if you run into any very cables there. And then we'll order, please be ready to come here. And humans will have to still be there to orchestrate it. But I believe that over time, AI is increasingly going to do that as well. So I'd love to maybe just talk a little bit more about some of the autonomous machines.

19:37I'll share kind of a quick anecdote. So I had a customer flying to Las Vegas, right? And I was like, can you meet with me in Las Vegas? I said, no way. I don't have time. I was like, well, can I pick you up from the airport? I'll drive you to your hotel at least, right? And so I have a Cybertruck, a Tesla Cybertruck. And I picked him up. I turn on self-driving. He's like, whoa, this is pretty good, right? And so I think for a lot of people, they're just starting to realize how effective self-driving is in consumer vehicles. But my understanding is you guys have been using autonomous machines and mining for a long time.

20:09So it's not quite as new technology for Caterpillar. Can you maybe share more about that? And as some of the technology expands into other types of operations, like construction sites, what's some of the cool stuff you're doing and what's coming up next? You know, these jobs, if you put yourself in the shoe of a mining truck driver in remote sites in Chile or in, you know, Australia or Peru, somewhere far, far, far from here is town. Those are not the most desirable jobs. They're not easy jobs. They're well paid jobs, but they're really, really remote to distant. We don't have too many volunteers.

20:47So if you can automate that and really help with productivity, safety as well, because really if you know humans, the safety is obviously less of a concern. And you get incredible value back to customers because ultimately this is for them. I mean, this capability makes their productivity go high and their safety concerns go away. So now you've got to bring that to construction. But construction is much more complex. You have many more moving people. If anybody's ever been to construction sites, know how things are always coming. People are coming and leaving. There's deliveries. Work is progressing in different parts of the job site.

21:25So our efforts now on the job site are very much being targeted at making our machines more autonomous and more capable of doing tasks without humans operating them. And I believe that that will be the next three to five years. A lot of effort is going by us with our partners and making that happen. But then orchestrating those job sites, as I mentioned before, is where I think will be a real magic. Being able to actually look and control multiple machines, ensuring that things are moving in kind of in sync with one another so that the job itself is progressing linearly and not starts and stops like it often does today.

22:09So it's a huge opportunity for us, for the industry, and actually for the country and the world, because things are going to be more efficient in the future. And we would like to play a big part of that, of course. So I've only got five minutes left, so what I would like to do is end the episode with a bit of a lightning round to kind of get your quick kind of takes. And admittedly, it's a little bit annoying because I ask questions that are very hard to answer succinctly. I'm kind of looking for your one tweet response, right, if possible. So my first question is, what would be your advice to maybe other CEOs out there about how you measure success of a CIO or a CTO or chief digital officer in this AI era?

22:50That's an easy one. I mean, I always start with this when I talk to my team. I will not measure you on just quality of your code or uptime of the services that you build. I mean, that's important. I will also measure you on that. But primarily, I'll measure you on business outcomes. and business outcome for me is you got to find a way that you're either improving to top line so that you can clearly be judged on contribution to the business and top line or in the cost savings that you've done so that we see a ton of bottom line. And usually that goes partly with your business partners. Make sure you share objectives.

23:24So you have your KPIs that are shared, but those KPIs are not going to be only related to code and deliverables. they will be tied to whether your code actually contributed to meaningful business results. So don't tell me that requirements were bad. I mean, that's one thing that I will never want to hear from any of my guys. And I would tell all my peers, please make sure you don't do that, because if requirements were not right, it's responsibility of your team, my team in this case, to actually work on those requirements up front and address them before you start developing and building stuff.

23:58Don't build something that is not going to work and then blame the guy who gave you requirements. Work with him up front to make sure you're working on the right things. And then judge both of you on business results that you got. Because, you know, digital teams, we can do so much. But ultimately, there's go-to-market. There will be people that are selling or working with customers, and they need to do their job as well. So you both need to challenge each other. But ultimately, you should share a fact. To me, that's the best way to show meaningful progress. If you had to give your quick three bullet points first, maybe one of your peers that's stepping into a new role in the AI transformation, what would be your three must-do projects?

24:37Foundation is a critical piece. So from a technical side, you've got to work on foundation first. Don't go after shining examples. You've got to make sure that you have a solid foundation. So spend time, effort, energy on that. Two, find areas that will actually be contributing meaningful value to the company and pick only a few that really matter because ultimately the digital world is a target-rich environment. You can go after so many things, but take three, four, five that really, really matter to the key executives and the board and make sure you get those done right. And third is get the right people around you.

25:13I mean, you got to have the right people and the right culture. We pride ourselves on quality of people and people that want to be part of this mission. And if you have individual heroes, people that are not team players, I mean, ultimately, they will not contribute to your success. They're there for themselves and you don't have them around. So make sure you get to a lot of good team players that believe in the division and mission you set out. Is there a book that you've read that's had like a big impact on you or your leadership or, you know, kind of how you think about things in the AI era?

25:44You know, AI era in particular, maybe I'd say general strategy book that really kind of influenced me is Art of Action by Bungay. It's really bringing some of the military strategy concepts into the business environment. It's the idea that you can never know everything up front, that even if you tell people to do something, what they'll actually end up doing will not be exactly what you told them to do. So there are these gaps with what you want and why you can't really plan centrally for all the possibilities. and how important it is to be very clear about intent and the ultimate objective that you want to accomplish, but then give enough freedom for your key lieutenants to come up with their own path of getting there and make sure they'll read back to you their understanding of what that original intent is.

26:37So giving enough autonomy to your team, yet making sure that you continue marching towards a shared goal that should always be there. It allows you to, in military world, survive or modify a strategy based on what your enemies do. But in real world, in our world, I mean really more about what competition and customers are expecting from you and your solutions. So how to adjust to that. So that would be the one that I probably want to highlight. What do you think will be true in the future of AI that most of your peers would consider to be science fiction? What do you believe is going to be true that other people would not?

27:12Yeah, well, you know, I was spending, again, I told you a lot of my early career time on autonomous driving when nobody around me believed that that would be happening. And, you know, I remember going to my bosses early on and saying, hey, we should be doing this. It looked at me like I was crazy. And, you know, not many years later, we are all driven in autonomous cars. You have one. I mean, I live in Phoenix. I've been making Waymo for years now. I view that the world that I operate in will be very, very autonomous in the next decade or so. And autonomous in a good way. Again, going back to safety, going back to productivity, going back to some of the challenges with getting the experience labor.

27:56And this world is going to need more things built. I mean, you touched on building. There's a huge wave of building and rebuilding that we'll need around the globe. And I think autonomy will definitely help in that. It's time to build. I appreciate you guys contributing to help us do that. Well, thank you so much for joining us today. And let's catch up again soon. Sounds good, Evan. Thank you very much. That was Augie Redzik, Senior Vice President and Chief Digital Officer at Caterpillar. Thanks for listening to Enterprise AI Innovators. I'm Sam Motamadi, a general partner at Greylock Partners.

28:31And I'm Evan Reiser, the founder and CEO of AppRomal AI. Please be sure to subscribe so you never miss an episode. Learn more about Enterprise AI transformation at enterprisesoftware.blog. This show is produced by Abnormal Studios. See you next time.