Thomas Leurent: We've had our brain as a species for like 600,000 years. We've only started to build good bridges 2000 years ago, or even 200 years ago, because we needed to have the models and the physics in order to do that, right? So that's what the physics bring to ai. It's like everything we've built, we've built not because we had our brain. But because we had the physics, Narrator: you are listening to Augmented ops where Manufacturing meets innovation. We highlight the transformative ideas and technologies shaping the front lines of operations, helping you stay ahead of the curve in the rapidly evolving world of industrial tech. Natan Linder: So we're back at Augmented ops. Today, we have Thomas Lauren. Hello Thomas. How are you? Thomas Leurent: Nathan, I'm great. How are you? Natan Linder: I'm good. It's kind of interesting things that do not really sound like ai, you know, like structural performance management. It sounds between something I have to know about and something I never want to know anything about. Like structural performance management. That's fair. That's fair. But Thomas, before we dive into this and like what your company X Excels is doing, tell everyone a little bit about yourself. Naton, we've, Thomas Leurent: we've known each other for a few years. We're MIT network and we started Xcels back in 2012. Fun fact, we incorporated a company, the company in Switzerland, uh, instead of of the us we pulled the technology out of MIT, between corporate in Switzer. Natan Linder: What is the main area you operate and what, maybe you should take a second to define this idea of structural performance management before we dive into the business. Thomas Leurent: Yeah. I mean, we forget how much everything that we use is grounded in, uh, in the physical space. In Natan Linder: in the physical world. Thomas Leurent: Yeah. Yeah. And so if you fly an airplane, you know, you're assuming that this thing is gonna land. Right? But how much do you know in real time about the fatigue consumption of your airframe? Right. Well, that's not an industry we operate in, but we, we did want a prize for exactly this kind of a use case with the American Institute of Aerospace along. This was with a Professor Wilcox a while ago. That's a use case everybody can relate to. Like does, do They know, they know how much fuel is left in the tank, that's for sure. Do they know how much life is left in the, in the steel? Natan Linder: Theoretically, you know, all the Roll Royce and GE engines now come with hundreds of embedded sensors that send information real time to control centers. But the catastrophic failures unfortunately still happen. You know, I'm, I think it's okay to admit that you love to watch true crime and, uh, plain investigation type of YouTube, uh, stuff. 'cause it's just like interesting. You know, those things. Um, Thomas Leurent: remember this guy that at some point, like there's a, there's a roll source engine. I think that the, the blade flew off and it kind of took part of the cain. Natan Linder: Yeah, Thomas Leurent: right. So the guy was exposed in flight because the, the blade essentially flew up. And that's where we come because as you say, they, they actually come, we don't work in aerospace to be clear, but they, they actually, uh, bring all that data. On what's happening, but how much life exactly is left in the blades. Natan Linder: Mm-hmm. Thomas Leurent: That's a different story. And that's where physics-based, structural performance management actually really helps. Natan Linder: So it's a lot of physics, probably math and applied math. So algorithms and data. Where do you apply it today? What's your focus? Thomas Leurent: Yeah. We're super focused on, uh, on the big kit in the, in the energy industry, so the legacy energy industry, oil and gas. Natan Linder: Mm-hmm. Thomas Leurent: When I say big kit, I mean something like floating barge the size of three or four aircraft carriers, right? Mm-hmm. That's called TSOs. We've hit huge value points in that space, but also the largest downstream plants in the planet, so either refineries or other type of plants in downstream, and in those things, what we forget was people feel up at the pump and they think it's, it's all easy, right? Yeah. But we now realize over the past couple of months that it's not right. It's actually very hard to get this oil in your tank. What that industry has done very well is actually build huge plants which produce very reliably and sweat the steel as much as it can. There's a lot of steel in the ground. Mm-hmm. And we understand that steel structural performance measurement better than any time before. Natan Linder: And we're talking here in Augmented operation, Augmented ops. You know, we talk about, you know, assets don't live in a void. They've, they live in organization and even energy producing assets like require humans to figure out what to do with them, you know, to check yields and figure out, I don't know, quality maintenance, things like that. What has this approach contributed? If you can give a couple of concrete examples, it would be great. 'cause the mind goes to this term that, uh, has been floating, I think for more than a decade. And I gotta tell you, we love buzzwords in this show. Uh, it it goes to like, the mind goes to digital twin. Thomas Leurent: First thing I say is what really gets me excited is when I can see a pretty clear path to a billion dollar value unlock. Like, and, and pretty fast. Okay. So I'll give an example. Natan Linder: Yeah. Thomas Leurent: Just in February she went public in, uh, in one of the, of the conference saying, well, Axo has helped us realize over half a billion dollar in value for an FPSO. So this, this barge, which is the size of three or four aircraft carrier, and that's because we could help the FPSO, uh, skip a dry dock. Okay. And so what does that mean? That means that, that, that facility, which you can imagine in the present circumstances is this even more would be at stake. But this FPAO, of course, is a critical asset. And then the inspection program, which was a fixed inspection program, was telling them, well, you have to unplug everything. All the wells unplug all the tiebacks and everything. And, um, the, the risers from the CED to the FPSO. Bring the FPSO to Korea and then dry dock it, and then do month of work on it. And then you replug everything. So you've got six months of lost production. Natan Linder: Mm-hmm. Thomas Leurent: You've got all the work that goes with it. This is, this is a lot of value. Right. And what happens is that those FPSOs are regulated by class societies. And those class societies understand that what we do, structural performance management, the. Technology before its finite element analysis enable to understand the steel. Yeah. But because we do it in a way which is more holistic, more detailed, we often unlock spare structural capacity. And that's exactly what happened in that case. So we, they, they did the analysis, which was more detailed than anything that ever done before the class society was load register. Everything was absolutely checked and rechecked to make sure everything was standard compliant, class compliant, and at the end they're like, okay, there's so much life left in the steel. We'll just, we'll just keep operating. I need Natan Linder: to know and how, how does it actually work? I mean, do you go to the barge and kind of sample it? Do you have live sensors? Because I know you have the models and like you work in different phases of the equipment's operations. So maybe talk about that a little bit. Of course we're trying to understand. Thomas Leurent: Yeah, Natan Linder: physical ai. I think when I say we, I mean the royal we, the planet. You know, physical ag now is all the rage, right? And I think like talking about how. Technologies like this work really grounded. Um, well, so how technically does it work? Thomas Leurent: Yeah. Let's start with how we build the model Natan Linder: mm-hmm. Thomas Leurent: And the workflow. And then I'll go to the core technology that enables all that. So how we build the model, we do have to build the model for the asset. And so what we don't do is we don't cut corners. Natan Linder: Mm-hmm. Thomas Leurent: Because those are high risk environment. We do not cut corners. We tend ly build a model, but we can do this for hundreds of kits on a, on a, on a very large plant. You know, it scales is a point. Once we've built a model, we impact from all the inspection and from the sensors, from all the inspection. We impact the real state of the assets. So this is really what. What's called a condition based model. And then after that we plug all the sensors streaming that so that we can replicate everything that happened to the asset. So the sensors, even if we build the model when the asset is 10 or 15 years old, they've had the sensors from since the beginning. Natan Linder: Mm-hmm. Thomas Leurent: So we then recompute everything that has happened to the asset. Like every cycle, every web. So Natan Linder: you go back in history. So that means you have to do you plug into like historians and kind of traditional systems and get all the data and then go back in time and Thomas Leurent: Exactly. And then we impact every cycle that the asset went through to understand the stress and the fatigue. And you say, well, but that's insane because actually that model and that computation must take, well, that's the thing. Like the core tech was about accelerating finite analysis and feas. Is what has designed the entire mechanical world. And when I say accelerated, I mean that for this case, for example, we run it about a hundred thousand times faster. Natan Linder: Mm-hmm. Thomas Leurent: Than FEA would do it. And that's where AI comes in. Because what we've been doing, just like Tulip Hass on AI forever in a sense. Natan Linder: Yeah. Thomas Leurent: What we've been doing is we've been blending machine learning with FEA since 2000, and then we've kept improving all that for 20 years. Yeah. Natan Linder: So the models you've built know how to, in a statistically significant way, output, very concrete recommendation about steel performance characteristics. That you can trust. So, so that means you have means to evaluate this model and like to provide a recommendation that you can stand behind, like you said, without cutting corners. And in a way, this is, this is kind of a form of ai. I think that, uh, you know, if it's not gen AI and it's not in, in some frontier model, people are like, oh, that's, you know, not important. But, uh, you're, you're talking about unlock of billions of dollars of value. Where, where does that meet like the customers? Reality when you have to explain to them, you know, this specific type of physical ai, like how are they seeing it vis-a-vis the excitement, the noise, the paranoia out there with, uh, generative ai and what do you guys do about it? Thomas Leurent: Yeah, that's a great question. So the first thing is, as you said, we are standard compliant, which is key in a high risk environment. So anything that can lucin. There's no stable state for that. Right. So, Natan Linder: but wait, Tom, who's, who's regulating you? Like, just for the audience who don't know exactly, you know, they don't read the regulation books for energy market. Thomas Leurent: The use case I gave the FPSO, they are class societies like, uh, Lloyd Register, American Bureau Shipping. They have a partnership with American Bureau of Shipping, for example. Natan Linder: Mm-hmm. Thomas Leurent: Those guys actually are, are always checking that what is done with those assets is right. Right? Mm-hmm. And, and if you're in class as an operator, it means that. You know, you're doing the right thing basically to make it short. Natan Linder: Yeah. And what is their leash on your clients? They can like shut down an operation if they're not happy. Like what, what, yeah. Thomas Leurent: They can, they can force a dry dog. They can, uh, yeah. Mm-hmm. This is exactly that, right? Mm-hmm. Now, on the other side, the operators also are autonomous in downstream because what they rely upon is standards. So API, American Petroleum Institute as a, as a big standard like, and which is completely embedded in our product, right? This is API 5 7 9, everything that is. Steel that runs red hot and is pressurized. Steel is part of that standard, and, and we, you have to be compliant with the, with the standard because it's a high risk environment. Again, when you talk about the noise with ai, anything that hallucinates is not allowed. Natan Linder: Unacceptable. Thomas Leurent: That space. Yeah. Natan Linder: Yeah. Thomas Leurent: Rightfully so. Right? So now where do they place us? I think where, where there is a really great complement is what we sometimes call. We may be a tool as a service. Of course, we use machine learning to accelerate all these, but it's serious mathematics that don't cut corners. But then we may be a tool as a service for a bigger AI pro program and an optimizing program that actually decides what it's gonna do with a plant. Needs the physics in order to make those decisions. And, and the comparison I like to take is AI is trying to replicate what we do with our brain with decision and recognition and so on, but we've had our brain as a species for like 600,000 years. We've only started to build good bridges. 2000 years ago, or even 200 years ago, because we needed to have the models and the physics in order to do that. Right? So that's what the physics bring to ai. It's like everything we've built, we've built, not because we had our brain, but because we had the physics. Natan Linder: That's a perfect segue to Jensen and just previously in GTC declared this year as the year of physical ai. So I'm kind of on a quest to collect people's definitions for that. Thomas Leurent: Yeah. You are more the specialist on that broad space than I am, but what I would say is it's got context. It relies on data, but it, it always relies on data and on rules at the same time. Right. So it's also rule-based ai and it's got an understanding of the rules behind the data. Natan Linder: Mm-hmm. So I think about physics, you know, which describes the real world and the ability to, you know, simulate, synthesize it. So obviously very related to the stuff you're doing, but also the stuff that is harder to simulate, like, for example, combination of events. Maybe some come from automated systems, but also some come from humans that still operate in the physical world. That, that's kind of been like my interest where, you know, in your field right now, you see humans interacting with, uh, what someone might look from the side and say like, Hey, you know, this is physical ai. Where, where do you see that? Thomas Leurent: So Tulip will have like cameras monitoring the space and so on. Yeah, and that's, I mean, for us it would be more. For example, one of the classic case in offshore is, uh, there's a boat that that actually collides with another. Boat or FPSO platform or whatever. At that point, you want to understand what happens, right? Mm-hmm. So you Natan Linder: forensics and can understand the Yeah, Thomas Leurent: yeah. And, and understand how much structural capacity you have left after that event, right? Natan Linder: Mm-hmm. Thomas Leurent: But I mean, unfortunately recently we've had many cases where, you know, we, we have many customers in the region, um, of the GCC. Of course, first and foremost, we think about the safety of everybody there. Yep, Natan Linder: of course. Thomas Leurent: But we've had cases where there's been impact from the, the, the events in the region. But I'll tell you though, what's, what's actually the most surprising is the biggest impact is that when you have to stop a plant, that's what people don't realize at the moment in the region when you have to stop a refinery from running to cold. Without doing it properly. Natan Linder: Mm-hmm. Thomas Leurent: It's a big deal because things corrode very quickly, so we finally is supposed to operate. Right? Yeah. It's supposed to push all the product through. Natan Linder: Yeah. It wants to operate. It's like, it's almost like an artificial organism. Thomas Leurent: Right? Exactly. Exactly. And the moment it stops, if you haven't done it properly, when you do it properly, 'cause they want to work on it, they stop it. Right. They get all the product out and they, they put it in an inert state. So they may put, I don't know, nitrogen in it to make sure that it doesn't corrode. The way it's been stopped at the moment in the GC all those refineries is, they've been stopped with the product still in it. Natan Linder: Mm-hmm. Thomas Leurent: That is not how it's supposed to be. So those assets are in a state that they were not designed for. Natan Linder: Mm-hmm. Thomas Leurent: And seeing that corroding and so on. So there's huge impact throughout the entire asset as this happened. And that's why when they say it's gonna take a year to restart it and so on, that's not just that there's been an external event, it's just the fact that you stop that refinery, uh, immediately in an emergency. Natan Linder: What's the role of humans in that situation? What do humans do in a refinery? I'm, I'm asking because I think like people kind of like to divide the world to discrete Manufacturing and process, you know, continuous Manufacturing and discrete, they assume there are humans and then process, they assume there are no humans. And I think maybe you have less humans, but it's kind of also a bit of a myth. You know, there are, there are humans and they do need to make all sorts of decisions and, you know, use data and like. Figure out how to do things better, faster, or, you know, prevent risk in a way that is, um, more, I don't know, you pick the trait, you know, uh, effective, sustainable, so on and so forth. So what's your take on the role of humans in process industries? Thomas Leurent: I mean, this is, this is a great use case because right now, um, there is no great optimized, uh, scenario for what's happening now. Mm-hmm. I mean, humans have to make decisions right now. They have to step in. Really, first of all, understand what happened during, say, an emergency shutdown. Mm-hmm. Like this may have caused all sort of issues, just the act of shutting down. You may have got into special vibration mode that you're not designed for and those type of things. So they have to understand what happened. They have to assess the state of, uh, of the, of the asset and, and prioritize where they want to assess that. Then they, they have to really now decide how to operate the asset. Most of the asset will be operated in one form or another. They'll be operated outside the design envelope for month to come. Natan Linder: Yeah, probably also from the SOP envelope because like some, in some of those cases that I, I guess they were dealing with emergency shutdown. There was no SOP was followed. Thomas Leurent: Yeah, Natan Linder: it was just like emergency, like a real one. Thomas Leurent: Yes. Natan Linder: So you've seen that? Have we heard stories like that? Yeah. Thomas Leurent: Yeah. Yes. Natan Linder: That's crazy. Thomas Leurent: And frankly, I mean, first of all, kudos to the operators because you know, this is an industry where people are really used to dealing with emergency. They are drilled into this. We're talking about really best in class operators here that have. Incredible track record on, on, on safety, and they anticipated things very well, but they, they also knew how to push that post button like that, that emergency stop button at the right time. And so they've, they've not incurred. Any bad scenario compared to the event that was happening. So extremely intense. Uh, they know how to do that right now. They are very disciplined about doing the assessment before they go into another space. But yeah, they're gonna have to dis to completely operate outside of SOP at best for weeks and probably for month to come. Yeah. Natan Linder: Mm-hmm. Back to this discussion of, uh. Physical ai, which like the mind goes to maturing technological product, uh, experiences like for example, self-driving cars have a world, world models that are getting pretty good at understanding all things about, you know, road conditions. GPS combined with traffic, uh, sign language. Detecting humans and, and then operating the vehicle itself with sufficient like detection, uh, response time, and all the degrees of autonomy. So that's like one thing people are talking about. The, the other big things is, um, you know, robotic brain models of all sorts that would help you train and work in the real world and perform tasks. And mo mostly like cut short, uh. Effort to achieve successful task completion or reduce the number of errors and also the retrain loops. So the retrain time, you know, you're not training a robot once. And so a lot of gen AI techniques are being used there. And, um, traditional type of, uh. Things like, uh, you know, machine vision is, is, you know, Thomas Leurent: yeah. Natan Linder: Pretty mature. So like, and then in this world, like you're bringing in something that is kind of going from historical design to what is the state now with machine learning. You know, Tulip is like, well. In this world, we think humans are still a big part. So we need the same physical AI equation to include understanding of a human in the operation. I'm kind of talking through like a real world scenario through this lens of, uh, you know, broadening the definition of, of what's physical ai, how does it meet your reality? So Thomas Leurent: let me give a couple of example on the big sides, right? Yeah. There, there are indeed technology deployed on those sites too. Serve the behavior of workers and, and basically catch when, when things are not safe. Right? Yeah. Natan Linder: Mostly safety, I'd guess. Thomas Leurent: Yeah. And to give an idea, some of the sites we work on at the, at the height of the, the build phase, and by the way, there may be a lot of build coming up, right. May new pipelines or, and new but 52,000 workers. We're working on the site at the height of the build phase, right? Natan Linder: 52,000. Thomas Leurent: 52,000. And as Natan Linder: on one site, is it, this is onshore? Onshore or offshore? It's Thomas Leurent: onshore. It's on Natan Linder: onshore, yeah. Yeah. Thomas Leurent: And this is actually, this is, I mean, I'm taking example here in, um, in Qatar, you can, you can go and check on the Shell website built in 2012, 52,000 workers. I think it's about 40 FL towers worth of steel. But, but importantly, even at the time. Not a single casualty for that bureau space. Right. So I think the safety behavior has been Yeah, the culture. Natan Linder: Yeah. Thomas Leurent: Yeah. In the culture. But this is where it comes in, right? You do want to, to stay at that level of safety and, uh, if AI can help, this is, this is awesome. Now, when we go into the scenario planning phase that you refer to, of course a lot of our customers use us for that, right? So it's like, okay, so now that we have something that we can run. That is so algorithmically efficient that we can run thousands of simulation of something that would usually would've, we would've had to give this a service provider, wait for three weeks. Now we can do run thousands of simulation at close to zero marginal cost. They use that to understand how they're going to operate the asset. How much they can push the steel or modify what you call the SOP, the standard, the operating envelope, what we call the integrity operating window. Natan Linder: Mm-hmm. Thomas Leurent: And if you can do that, especially at the moment, and that ties back into the value, if you can modify the IOW at the moment, just as the industry needs flexibility, you can really produce the right product at the right time with the right. OPEX and CapEx. Natan Linder: So I think I have to bring out the kind of the elephant in the room of physical AI as I see it, if that's okay. Okay. Because I think in the past, say three, almost four years, we've gotten used to those, um. The foundational models and they're getting of course, more capable. So we are seeing visual language models. You know, we, we do a lot with cameras and our playback product that you can kind of wire up an operation and give a production system eyes and then have a profile or debugger tool. So, you know, all this stuff is being on one hand combined and it's progressing kind of fast. But the industries we operate in are actually. Pretty conservative and are nowhere near the amount of data that is locked up in industrial operation of all sorts. You know, think about, you know, the future a little bit and I think we're in the few years before, uh, a GI, and you know, I guess when that happens we'll have to do another episode to figure out like what the impact of that. But before, let's just say, I don't know, we have four or five or six years, who knows? May maybe earlier, but, um. My point is, is that we are in the engineering phase of these, um, large language models, small language models, visual models, that, that we understand how to build them. But what I think is not solved is like how do you get customers comfortable in sharing data? Because you can imagine, I mean, I'm sure you have other customers, not just shell that in an emergency situation or if billions of dollars are on the line, you know, maybe they start thinking a little bit differently on, is my data that proprietary, or do I want the advances of a, you know, powerful AI model that is suited. To my type of problems. You know, I'm just extrapolating, I'm sure you all have thought about that, but like what value or what more value and faster value you can provide to all your customers if they were able to come to, uh, an agreement on how to share data and, and create those models. So how do you see that play out in your industry? I mean, I'm really curious. Thomas Leurent: So first of all, when it comes to safety, operators have been good at this in the oil and gas industry. So if you, Natan Linder: so they are sharing data on safety? Thomas Leurent: Yes. Yeah, they are. And they're sharing technology, not just data. Right? So the limit is clear. In the oil and gas industry, historically, what it has been is that when you go into reservoir modeling, like the stuff that's in the ground. They are very proprietary about the data. Yeah. Both on the technology and the data side. Right? So everything that's in the reservoir super proprietary and that's not where we play everything that is above the reservoir. They've been, and also, frankly, for safety reason, you have to remember, this is an industry where you had really bad accidents. Yeah. Back in the eighties and so on. And the industry cleaned its act and said, we are now gonna be a safe industry. We're gonna be very serious about it. Right. And at that point, one of the thing was that they're going to share more the technology, share more the data that everybody benefits from in the industry. So I think the behavior we've seen in this industry has been exemplary on the operator side of oil and gas. Now, I think when we go to the OEM side of the mechanical industry. That's a big deal. I think OEMs are basically shooting themselves in the food by not being able to share data enough. And there are ways to share data without sharing ip, and that's what they have to realize. Right. And I'll give an industry that Got it. Right. And, and of course, you know, and it's simply the, the electronics industry, right. And so in the electronics industry, they've been used with platforms like CAENs and synopsis to, to share the data of each piece of the system. That you can build those extremely, like basically those supercomputers in a pocket, right? And if you couldn't share the data of the system, they would never have been able to implement more slow to this extent, right? Natan Linder: Mm-hmm. Thomas Leurent: The mechanical industry got stuck on that. And a great example of that is actually synopsis buying Ansys. Nobody would've seen that 15 years ago, but that's because one industry is sharing data and therefore growing exponentially and the other one is not. Natan Linder: Mm-hmm. Thomas Leurent: I mean, that's one factor. Yeah. Natan Linder: So hypothetical example, not operational, but not about say, safety. Like you can imagine that you may want SPM data across all the rigs of all the energy manufacturers on the planet. So the next time they ask you to assess something, you have a a much better. Product to, to give them. Yeah. But they, but, but they might think like, I'm giving out data that is proprietary to how I build my rigs and how I design them and this and that. I don't want that out to train a general model plus what's in it for me. So I think that puts people with, uh, the product and the engineering capabilities, like, like your company and their direct customers. Potentially also the integrator in kind of a, an interesting position because I don't think on tropic and the likes are coming, like to help the problems you're solving for the industry anytime soon with their generic models. It's just impossible. Thomas Leurent: So the first thing is. With our customers, that data belongs to them, right? Yes. Yes. Contractually, that's how we operate, of course. And that's how we want to operate, right? Same, I think for, for today. Third one thing is we actually don't need it because we, we build those models for the operator. We have standards that we apply. Natan Linder: Mm-hmm. Thomas Leurent: And so it works out without the data sharing. Right. Uh, we don't need data sharing to create value. I think where we would have had that is when we go on the design side, for example, in the offshore wind industry. Everything in offshore wind is overdesigned to ridiculous extent. In fact, the industry before it crashed, used to be proud of it. Like, oh look, it's so heavy and you know, it's so beautifully heavy and it's like. It's not beautiful. It's over designed and it's costing CapEx and it's making you a known player in the energy industry because it, Natan Linder: you talk about the efficiency of, say a wind, like a windmill type of operation. Yeah. Thomas Leurent: I mean, an offshore wind turbine is over designed from the tip of the blade. To the foundation, deep in the ground. You know, the foundation is of a design, if it's floating by 2020 5% and the, if it's a jacket by 10, 15% and we've been at sail week actually showing this with partners like, uh, OEMs like lamp rail. Natan Linder: Mm-hmm. Thomas Leurent: They were selling 25% on the, on the transition piece. This is tons upon tons of steel, but it's all over designed and then they wonder why in the end people don't want to. Build more offshore wind. Well, it's basically too costly. And for this, if they were about able to, to share the data more between the blade manufacturer, turbine manufacturer, you know, even within the turbine manufacturer, the blade division doesn't share the data with the rest of the, of the company. That is all costing in the end in CapEx, and that's making an industry less competitive. And, and for me, the crash in offshore wind, one of the issue was this lack of data sharing. The other one was lack of anticipation on, on interest, right? And the third one, frankly, was standards imported from the imported from the oil and gas industry for an industry that doesn't have the same uh, risk. So we should have gone to new standards based on new technology. Natan Linder: Yeah, I want to end on a optimistic note. So, 'cause I feel like, you know, we're talking a lot about energy and I think the price of oil has been, we can say it's been on people's mind the past, uh, several weeks. So I feel like it's kind of like peak volatility, you know, all the stuff we've seen in the market as a result of the global conflicts, the. Complexity of the market, the adoption of ai, the operational restructuring, and like, you know, are, are they gonna create an alternative to the oil flowing out of, uh, whole moose and so on. What are you excited about? What give you confidence that we're actually resilient and, uh, the world is not gonna go crazy even more than it is right now. Thomas Leurent: I think resiliency is now a top item in the, in the energy CEO's agenda. Natan Linder: Right? Everyone is talking about that resiliency. Yeah. Thomas Leurent: I think in term of optimism, I, I'll tell you, I think one thing that is happening now is that people are realizing that things are complex. That what we thought were secure lines of supply are not necessarily that secure. Natan Linder: Mm-hmm. Thomas Leurent: And that we need to double down on diversification. And residency. And so what we're gonna see moving forward is people more conscious of where the energy comes from. Natan Linder: Mm-hmm. Thomas Leurent: We are gonna integrate that more into the equation of not only how government think, but how consumers think and how voters think. And that was far from people mind before us in the energy industry. We knew about it. Now everybody knows about it. Right. That's gonna modify behavior both in a consumption and voting, and that's, that's probably a good thing. Natan Linder: Are we gonna see diversification in a certain part of the world, do you think? Or it's just like globally? I mean. Obviously Asia Pacific has really impacted, and we all heard the news on how Japan is thinking about this or us is kind of tough to follow, but, and it's, there's been Venezuela as well, so are they gonna build a lot more facilities and how, how is that gonna impact, what, what do you think? Thomas Leurent: I think there's gonna be two things, molecules versus electrons. People will think again. Where do I want to get my energy from? Because I've got a choice. Mm-hmm. Or do I want to get both? Right. And then there is centralized, which is decentralized. Everything that's centralized is easier to take down. Natan Linder: Yeah. Thomas Leurent: These things that's decentralized. So you have a solar roof that's mo less likely. Okay. Natan Linder: Yeah. Thomas Leurent: In a sense, right? So you've got centralized molecule versus electrons and then you've doubled down on your infrastructure. You remove the critical path. So we've already done that in Europe with uh, after the Ukraine invasion. We started to really have more import terminals for, for natural gas, right, for LNG and simply bringing that redundancy in your system, even if you choose to, to stay in a centralized molecule system, you're gonna have more redundant redundancy than before because you need to, I mean, I wouldn't be surprised to see a pipeline that actually bypasses the straight of all moves within three to four years. You know, why not, you know, Natan Linder: via Saudi or something like that. Thomas Leurent: Yeah. And Oman and so on. I mean, the Saudi pipeline is active already, so that one was a, was a good idea, but they, they're probably gonna do more of those. Natan Linder: Yeah. Yeah. You know, the needs are just coming up. And I was kind of imagining you also touch on like, you know, the demand is not just coming, like, let's get back the supply up and running. It would be like, we don't have enough power to. Figure out this new form of compute we're all using on a daily basis called ai. You know? Yeah. And that will put another strain on the energy market Thomas Leurent: as well. But you, you know my take on that, right? I think algorithmic efficiency, exactly. The kind of algorithmic efficiency we brought to mechanical simulation. Mm-hmm. I think that's gonna kick in also on the AI side. So we're gonna move away from LLM to small language models in many cases, and, and then. We, we can't, this is not sustainable where we are now. So basically those l and m models are algorithm things to be very inefficient and that's gonna change. And so this is a far prediction, but if it happens, what we're gonna see is that there, there's gonna be, have been an of investment into all that computing Natan Linder: mm-hmm. Thomas Leurent: Capability because the algorithm will catch up in a way which is surprising and very powerful. Natan Linder: So that suggests over capacity in X amount of years in the future. Thomas Leurent: Yes, we are gonna be able to run. AI is much more cheaply than we think, I think. I think it's already kicking in by the way. Right. We've seen deep seek versus open ai. Right. That was one. One example, but that's not over. It's gonna keep going. Natan Linder: Yeah. Okay. We're gonna keep going to Thomas. Thank you so much for joining. This has been a good, good break from the noise of physical AI and kind of grounding it in like real industries and real. Situation the world is dealing with right now. So I really appreciate you coming on the show, and we'll catch up soon when you're in Boston or Vietnam or wherever your company takes you, and looking forward to that. Thanks a lot for joining. Thomas Leurent: Thank you so much. Bye. Narrator: Thank you for listening to the Augmented ops podcast from Tulip Interfaces. We hope you found this week's episode informative and inspiring. You can find the show on LinkedIn and YouTube or at Tulip dot co slash podcast. If you enjoyed this episode, please leave us a rating or review on iTunes or wherever you listen to your podcasts. Until next time.