Subscribe
Share
Share
Embed
Automation does not fail at the top. It fails in the layers underneath.
Sami Sultan, Vice President at Darcy Partners, ex BCG, and ex Shell wells engineer, returns for part two: a walk through the full Physical AI stack as it actually exists in heavy industry today.
In this episode:
The stack, layer by layer. Cameras reading rock as it comes off the shale shaker, computer vision flagging anomalies, AI agents that sense, plan, and act, and the actuation layer where a digital decision finally touches physical equipment.
Human in the loop. Why the geosteer keeps their job when a few percent of production is worth millions, and why safety critical calls will stay human for a long time yet.
The strange origin of mission control. How a failed military operation in 1980s Iran gave birth to joint command, then to the remote operations center, now the brain of rigs, mines, and factories everywhere.
Why full stack automation ventures go bankrupt. One sensor fails and the house of cards comes down. The aerospace lesson is that automation is a management system, not a feature.
Beyond the buzzword. Why Sami refuses to say digital twin, and why physics informed neural networks are winning the trust that black box AI cannot.
And the next frontier. Drilling techniques crossing into mining, continuous extraction, and the bridge that could finally make Western rare earth production viable.
Watch this space.
About the guest: Sami Sultan is Vice President of Oil and Gas and AI at Darcy Partners, where he leads technology scouting and advisory for the world's largest energy operators and utilities. He spent nearly a decade at Shell as a wells engineer, where he helped found Shell Geodesic, an algorithmic well navigation venture that applied AI to the subsurface years before it was mainstream. He holds seven patents, is a BCG alum, and earned his MBA in sustainability from the Yale School of Management.
Sami Sultan, Vice President at Darcy Partners, ex BCG, and ex Shell wells engineer, returns for part two: a walk through the full Physical AI stack as it actually exists in heavy industry today.
In this episode:
The stack, layer by layer. Cameras reading rock as it comes off the shale shaker, computer vision flagging anomalies, AI agents that sense, plan, and act, and the actuation layer where a digital decision finally touches physical equipment.
Human in the loop. Why the geosteer keeps their job when a few percent of production is worth millions, and why safety critical calls will stay human for a long time yet.
The strange origin of mission control. How a failed military operation in 1980s Iran gave birth to joint command, then to the remote operations center, now the brain of rigs, mines, and factories everywhere.
Why full stack automation ventures go bankrupt. One sensor fails and the house of cards comes down. The aerospace lesson is that automation is a management system, not a feature.
Beyond the buzzword. Why Sami refuses to say digital twin, and why physics informed neural networks are winning the trust that black box AI cannot.
And the next frontier. Drilling techniques crossing into mining, continuous extraction, and the bridge that could finally make Western rare earth production viable.
Watch this space.
About the guest: Sami Sultan is Vice President of Oil and Gas and AI at Darcy Partners, where he leads technology scouting and advisory for the world's largest energy operators and utilities. He spent nearly a decade at Shell as a wells engineer, where he helped found Shell Geodesic, an algorithmic well navigation venture that applied AI to the subsurface years before it was mainstream. He holds seven patents, is a BCG alum, and earned his MBA in sustainability from the Yale School of Management.
Creators and Guests
Guest
Sami Sultan
Sami Sultan is Vice President of Oil and Gas and AI at Darcy Partners, where he leads technology scouting and advisory for the world's largest energy operators and utilities. He spent nearly a decade at Shell as a wells engineer, where he helped found Shell Geodesic, an algorithmic well navigation venture that applied AI to the subsurface years before it was mainstream. He holds seven patents, is a BCG alum, and earned his MBA in sustainability from the Yale School of Management.
What is Leadership for the Physical AI Age?
A podcast by Tryfecta Capital exploring the intersection of AI, Robotics, and Leadership.
We deconstruct the business of Physical AI: how to invest in it, how to build it, and how to lead through it. Featuring interviews with Industrialists, deep tech founders, VC insights, and market analysis on the future of automation and embodied intelligence.
And welcome to another episode of Leadership in the Age of Physical AI. I'm pleased to have Sami back. We've done a deep dive into capital markets for oil and gas and a bit of history on Sami. But, Sami, for this one, I really wanna get in. I wanna go all in on the tech side.
Speaker 1:Right? Energy in any form is important. And so so I'm keen to understand what the current market, what the tech in the market looks like. We're talking about intersection between technology and energy. Tell us what's what's what's going on in that field.
Speaker 1:Right? Where's tech at? What what are we what's what are we driving towards?
Speaker 2:Yeah. I think I I think I spoke a little bit about it previously. Right? But within oil and gas, similar to how there is in in other industries, right, there is what I would call, like, your physical AI stack, if you will, or layering of, you know, how you might approach physical AI. And it and it start out as that, like, when people are thinking about it, they're not going like, what is my physical AI stack?
Speaker 2:It's really just how can I improve my operations? And this is just naturally where it might end up. People look at it in, you know, retroactively. However, what I would say is kind of composed of that are similar again, how I touched on before, like similar to what you might see in like levels of autonomy. You're gonna see a lot of things like initially being able to sense what is out there, like through your sensors, right?
Speaker 2:Whether it be a IoT sensor that is, you know, hooked up in your operations and in production space, whether it is a camera on a rig or otherwise. You see cameras now for safety purposes. You see cameras now first, you know, in core efficiency. Some people have it, like I said beforehand, in a rig floor, like making sure people don't come into the red zone. And you have it also in, like, say, a shale shaker where all of your cuttings are coming back from downhole, and you have a camera overhead to be able to understand the cuttings that are coming back because there's just so much it's like a whole swath of information that is coming downhole It's very different than like ones and zeros.
Speaker 2:It's a real rock. And the way that that rock is cut is really important. How it's coming back, the color of the rock, the type of rock, all that information, a mud logger that position on rig side can theoretically do in many ways. But you can also potentially simplify that. Right?
Speaker 2:You can also potentially have an AI component there that is on top of a sensor camera. Companies like DrillDocs are doing that. There is then potentially like a deeper perception layer that is focused with things like say anomaly detection. Like as you might know, you have like these sensors, the cameras that are understanding, like, what is actually happening first, but then there is the actual, like, computer vision approach where you're actually understanding what is happening and then adding that AI layer sort of on top of it as well. Anomaly detection right now is like all the trend.
Speaker 2:Do you see that everywhere? Because it's really important for things like predictive maintenance. It's really important for things like in safety and otherwise. If something isn't what it appears, well, we gotta flag it. And if there's anything that the oil and gas industry or general energy industry has, it's a multitude of data of what it should look like.
Speaker 2:And that's where it becomes really, really powerful. You see this also in the utility space with things like power lines. Dash AI is one example that is looking at it from like a vegetation management standpoint where you're like, you might have that that view of what it looks like through satellites and otherwise. But as that that satellite view starts to look a little bit different, whether it's, you know, vegetation growing onto lines or otherwise, then you need to flag it. So, there's that that separate layer there.
Speaker 2:What we're Excuse me. What we're seeing also now is like sort of a decision layer or, and AI agents, I think is a critical component of that, where you then use take in that information and then you need to make a, excuse me, a decision upon it. Well, that's where like the core component of an agent really becomes powerful. In any given agent, you have like the well, it depends on how you view it. If it's like a core AI agent, then there's like the, you know, the reasoning aspect, the plan, the actual action that's happening.
Speaker 2:There's some reflection that's involved there. It's like a loop that you're kinda consistently going through. But that action is where things really start to get interesting. I almost argue that is like a separate layer. Within the loop of an AI agent, there's also like the planning layer, like the sensing of what is the information is being asked of me.
Speaker 2:This is regardless of Physical AI or otherwise. It could be an LLM putting in something, right, some information. Well, you need to understand what it is, and there's like the planning aspect of it before ultimately like acting. With that, you have tooling such as that people are building on, you know, I think I mentioned previously with like geo steering. This isn't what we did because agents just wasn't a thing in 2017.
Speaker 1:There was no memory that could do it. There was no memory that could do it at that time.
Speaker 2:The memory.
Speaker 1:Yeah.
Speaker 2:But you can easily Interestingly.
Speaker 1:Yeah. But interestingly, you we we did have agents in there, but it just just we did not the the hardware caught up with the software.
Speaker 2:That's Exactly.
Speaker 1:What's happened the last couple of years.
Speaker 2:Exactly. The the the the aspect still exists in many ways, but it you know, you can theoretically understand, like you get to a point where similar to like a human in the loop, which is ultimately what we ended up with in that situation because to remove a geo steer in theory, is that worth, like the cost of having that individual, is that worth removing for what in theory is, even if it's a 5% production, right, millions and millions of dollars. Like in theory, that is not worth removing. You'd rather have that human in a loop. And that's a whole another conversation as it relates to like AI adoption and people's comfort levels there.
Speaker 2:But like you can easily see like how you can get to that point to where you want to suggest a decision to be made. After that, like I mentioned in that loop, there's that action. Right? And the action, it varies. If there's, you know, a typical agent that you might see in like a digital sense might be, okay, I'm making an agent to go book me a trip to Hawaii.
Speaker 2:Well, it might make that plan. I might write this all out. It might make that plan. And then they'll tell me like, okay, I found these tickets to Hawaii on United, whatever this price. Do you wanna actually buy it?
Speaker 2:Right? And then there's that action. Too difficult. Like, it's pretty straightforward. A lot more it's involved in the back end, but it's very, you know, digital.
Speaker 2:It's not difficult to create something like that. I shouldn't say that. It obviously is, but it's the but then when you think of the physical AI sense when you think of the physical AI sense, that's where it becomes really, really powerful because I can't do that in my, you know, RTOC or remote control room or whatever you wanna call it, right, real time operations center. I can't do that in my ROC and then be able to access operations purely digitally. There needs to be some kind of physical component typically that's actually making those changes for the most part within a, you know, a rig site, within a production outfit or whatever it may be.
Speaker 2:There needs to some kind of layer there that's involved. You guys think of this from like a directional drilling standpoint. Otherwise, back in again, 2017, well before that, you've got like a centralized directional driller maybe that's communicating to all the other ones on like how they should operate. But they're not the ones that are physically making the adjustments to the rig itself. Because they just didn't really exist back then.
Speaker 2:That closed loop control is a little bit difficult. Now you have things like there's at the edge, you have different types of actuation that's involved, more smart types of injection, auto frac and things of that nature that get really, really interesting that people are starting to adapt. And you see a lot more of these closed loop controls that are really becoming impactful here. And all of this to say is what's ultimately being injected into, like I mentioned, those remote operation centers. That's what's all the hype right now.
Speaker 2:That's what we're kinda really exploring a lot as well in Darcy because you have these Imagine like Mission Control at NASA. Right? You have these control centers that have all of these screens, all of this information and data flowing to them. And it's very difficult when you have, you know, 20 rigs or all these different types of, you know, operations that are running in front of you and you're working a twelve hour shift potentially trying to understand like, what do I really need to have my eyes on? And some operators are, you know, outsourcing a lot of this to a company like Neighbors, like Precision Drilling, otherwise.
Speaker 2:Again, that's in like the drilling domain. There's other aspects as well, but you can outsource a lot of this, but a lot are also trying to build out these centers internally within their own organizations. And those are real differentiators there. I don't have any specific operators here, but those are, these are the ones that are really trying to understand what is the extent that we can bring in all this information from like that sensing in a layer, right? All this different data and how am I ultimately then able to make it easy for me to understand when I need to move my eyes to certain operations, whether it's from an anomaly detection standpoint or some kind of alarm or something that's coming out.
Speaker 2:Is it suggesting a decision for me to make? And then me as a human in the loop, am I able to then say like, hey, go ahead and make that, that makes sense or no, this does not look right. We need to address that or adjust it. More often than not, you would hope that ultimately it would be 99 times out of a 100 that it's making the right call. And there's just that one where it might be a bad call, but like you need to investigate a lot further potentially.
Speaker 2:Right? That's like the goal where all a lot of these operators are really going towards, but it relies on a lot of critical components, I would say. If you don't have any layers, it fails. Right?
Speaker 1:It's interesting to say that, and I really wanna get to the automation bit and actually leave a fair bit of time to talk about automation, which I know you have a lot to talk about. But just before that, it's it's quite interesting because you brought up the whole RTOC and remote control centers. And that's quite interesting because, really, the space industry started the whole concept of of mission control and and remote operation centers, and they built that from the ground up. We've never had a precedent of actually incorporating these mission controls into existing operations, different departments. And I think I think what sparked it all was the was a botched mission in Iran back in the back in the eighties.
Speaker 1:And what actually happened was the US military just found out that there were all these disparate teams not really coordinated, and they they came up with a joint control. And our industry has just picked up from there. We've used that concept to build the whole concept of RTOC, which has now gone into mining, which has gone into, you know, now everywhere. It's a standard across manufacturing now. But where do you see that?
Speaker 1:Where do you see the evolution of that RTOC going? Right? And and remember, this is the segue into automation. So we we have a pretty defined industry now with remote control centers, and that was all about getting all these teams working, these these humans in the loop and basically based off of humans. Now you have a combination of agents and humans.
Speaker 1:Where do you see that RTOC going? And remember, we we're gonna jump straight into automation after.
Speaker 2:Yeah. Look, I mean, these there's a lot of challenges and hurdles that I think need to be resolved before you really start to see that next evolution. All very, you know, very, very much hurdles that you can overcome. But I think whether it's RTOCs or otherwise, there is a component of, I would say trust that is critical to that. And that trust a lot of times is built off of like what is being inputted into any said system that is feeding our TOC.
Speaker 2:So like the data management layer is a critical component of that. Are we denoising all this data effectively? Are we only putting the right types of data in any, you know, any kind of operations? You might have like thousands of different channels of data that are coming to you, but what are the ones that actually move the needle? Right?
Speaker 2:And how can I make sure there's no like anomalies and areas that I need to really recover from there? So, once you have all of that and you can build out like the explainable piece, that's where you start to really trust the AI. And then that's where like the human in the loop becomes really, really prominent. In our conversations with the industry, this doesn't matter what industry it is, but human in the loop is very ubiquitous across this group, especially given the high risk nature of an operation, both from a safety standpoint and just a capital standpoint. And based off of those two major factors, it almost similar to what I said or shared before, a geo steer, you almost need to have a human in the loop.
Speaker 2:And I suspect that will be the case for quite some time. I do think that over time for the non safety critical type of work, we gain more and more trust. And I think we start offloading a lot of these things to an agent per se, and maybe there's other layer other layers of work that are involved there, but maybe you don't need a supervisor that is managing all of this. And as you may have heard obviously before, like a lot of people think that, you know, everybody is gonna become a supervisor of agents and all this. Right?
Speaker 2:I don't know what's true, what isn't true. There's a lot of factors that are involved there. That's probably gonna be the case in some form of way. I think it sounds a lot more glamorous than it actually is. Like, everybody becoming a manager.
Speaker 2:Side of the point.
Speaker 1:That's why the podcast is called Leadership and Age of so we're basically redefining leadership. Yeah. We don't know whether are we gonna be do we have agents or do we have people in in the next four or five years? Right?
Speaker 2:Well, you you know, a lot of this also is like the Jevons Paradox situation. Right? The more that, you know, you use these these tools, the more of like the demand that kinda comes from it. And I think I do believe that will be the case. You know, what these roles ultimately end up becoming, well, I don't want to derail your podcast, but I think there's a lot to be said there.
Speaker 2:I think it will evolve, you know, that is for certain. I think, know, how we manage this stuff becomes really, really impactful. What is actually ultimately shared to said supervisor does change dramatically. So as I was sharing before, they may not be a human in a loop for every one of those tasks that they may have had. Maybe instead of like, you know, ABC tasks, it's starting to evolve into only like a and b that are really critical.
Speaker 2:And slowly that kind of changes and take changes. I either will always be for certain very safety critical tasks that are done. You'll almost always have a human in loop because potentially lives are at stake or other things of that nature. You may lose the well if you make a dramatic change. We're just not at that level of comfort in the industry, but it will evolve.
Speaker 2:So I do see an ROC or RTOC or whatever it may be evolving in many ways as it pertains to that, but it'll take some time.
Speaker 1:So you're you're sort of saying I mean, I think I think where you're going with that is we're basically seeing the evolution of the RTOC becomes a key part of how we manage heavy industries, and that overlaps a point where automation naturally evolves Yeah. From an RTOC setup.
Speaker 2:I think so. Right? Especially as you get more and more of these sensors, of these data points that are coming to you, part of the automate before I touch on the auto we're gonna touch on the automation part here in in a little bit. But you start to see like if I need to, you know, going back to what I was sharing before in that little framework beforehand, right, if at the very end of it is or middle of it, forgot exactly where I was at with it. If I need to act, right, similar to like what an agent is doing, if I'm gonna ultimately act, well, there's some things that like I can flip a switch and like there's an actuator that turns off or whatever it may be.
Speaker 2:There's other things like maybe now I'm working with a different type of innovator, like a pipe crawling robot, like with Eric's technologies that is now involved and I'm able to get that information back to Ascend for like asset inspection and like pipe integrity. And now I'm able to get that information and then I can make a decision based off of it. It may not be an ROC, but you know, it very well could. It can be from an engineer or whoever it may be. But now like I have all of this disparate forms of data and information.
Speaker 2:You could think of like a ROC, it's just a great example for it as like this brain behind everything that is ultimately kind of feeding and then going out and making these decisions and actions. So it does ultimately evolve, you know, like I said, the automation part does change quite a bit. There are just like, you know, in any kind of, well, in any kind of like digital wave, I say that's kind of come through over time. People want to experiment and try things and you see a lot of, you know, this stuff grow. And we saw this very clearly with like generative AI.
Speaker 2:You start to see though, the governance slowly come back, right? That slowly will eventually come. And then when it does, then that's where people are really starting to look at some of the commonalities with any of these digital waves, data, the trust aspect, like how do I feel about from an adoption standpoint? All of those start to really start to rise and it's gonna affect every aspect that, you know, AI might touch. Whether it's, you know, a robotic crawler or whether it's like a rig floor operation type of work or otherwise.
Speaker 2:So, all will change, I think, over time. But we've seen a lot of these trends beforehand. I think it's gonna continue to rise and grow though going forward.
Speaker 1:Now, I think it's worth noting also that we we come from a similar project. We're working on automation for drilling rigs. And we know that some of the challenges where we could not really, in all good faith, could not commercialize some of this tech was is unreliable. Right? And so we're basically looking at a, like, sort of, I would say, post AI world where we've got some of these tools to really clean the data that we really could not do ten years ago when we were working on these models.
Speaker 1:Talk us through this whole your automation journey, and where do you think that's going? Right? And what what are the technologies that need to evolve? What what needs to happen for for automation to become a thing?
Speaker 2:Yeah. Automation is an is an interesting one because I don't I I'd argue that, like, a lot of people see automation as this grand final destination that we're all kind of going forward with. There's a lot that's
Speaker 1:about Final destination.
Speaker 2:Maybe maybe those are poor words. Maybe final destination isn't really where we want
Speaker 1:to go.
Speaker 2:But you know, maybe that's the case. No. But you do see nowadays that is like the perceived goal in theory. Right? There's a lot of ways to slice that.
Speaker 2:There's a lot of different paths to get there. It may not actually be like that final goal because a lot that's involved still with it. And automation, the definition of it, it it could just be repeating the status quo just autonomously, you know, if you will. Right? Use the same word.
Speaker 2:It's just doing that on repeat. The innovation factor where, like, AI really starts to become, you know, why that's so important is because if I can innovate and I can see patterns evolve and then corrects my patterns, some kind of like reinforcement learning or otherwise, right? That's where it starts to really get interesting. Nonetheless, not to get into like any of that area. As I mentioned, like when you think about like the overall like AI stack that is needed to achieve, you know, an overall comprehensive like AI goal, I would say, there's new technologies that are coming out across every layer.
Speaker 2:It's not like there's no one layer I would say that is more important than the other because we need all of these different layers to improve over time. The automation ones are where that gets like probably a lot of the, you know, attention grabbing headlines because I can now do this automatically. We've seen for, I won't call out any companies, but like we've seen these rigs of the future, these maybe I'm calling out companies by using marketing taglines, but like we see all these, you know, what does the, what are these futuristic rigs really look like where everything is fully automated and we've seen this for decades. Unfortunately, we've also seen a lot of companies try to do the full stack of like automation and then go bankrupt. That happens quite a lot, especially with the complexities are involved in the rigs.
Speaker 2:So now you're seeing these individual components that are being automated and you've got like interesting companies like Neighbors that are putting together all these interesting components. And like, you know, I mentioned things like pipe crawlers and all that, but you also have things like, again, I'll use drilling as an example, just for my background. Like if I have something from that's bringing up pipe automatically and then I'm stacking it into my finger boards up top, and then I'm able to actually, you know, make those connections using an automatic ST 80 or whatever it may be. And then all these other specific aspects that are kind of involved, you're starting to connect them. It's very modular in many ways.
Speaker 2:But like I mentioned before, to enable these modular components of automation across any kind of operation, there's the other previous layers that need to be done right first. Whether it's from a sensing standpoint, like we see the evolution of these sensors improve quite a bit. Everything from like the edge computing part, like I think I mentioned Armada beforehand, who's going quite a bit in prominence, all the way to, you know, like acoustic vibration sensing of like compressors and things like that where I can then understand like, are there any anomalies that are coming out from this specific piece of equipment? And if it is, then I need to flag it down to another layer. Right?
Speaker 2:And then they're gonna make a decision or it's a full integrated stack and I can do it all in one. All of these kind of come together, but they're improving every aspect, every layer to ultimately then prop up and make automation theoretically possible, I would say.
Speaker 1:Now that's that that's that's amazing. Right? Like, because you have I I don't think a lot of people understand how a small you you need one failure for the whole automation stack to fit, and I think you learned that the hard way. So you could build your whole house of cards on automation and have one competent, one sensor fail, and the whole thing falls apart. And and you see that, say, you know, some of the pioneers of, you know, automation and its, you know, level one, level two automation, Boeing, Airbus, and the whole aerospace industry, and what they have done.
Speaker 1:And they've went through years and years of failures before they sort of refined and managed to build a whole management system around their automation, which people don't take into account. People think it's just purely a technical task, but it's not, is it?
Speaker 2:No. Yeah. It really isn't. To kinda bring this full circle, you know, one of the core components that we see nowadays is, like, this digital twin in many ways that allows you to be able to, like, do a lot of this stuff. And anybody that really knows me knows that I really don't like the term digital twin, and it's evolved quite a bit.
Speaker 2:But in theory, like, that is similar to like that metaverse layer that is is critical. Like, it's always there. It's more of a marketing term more than anything, but you need all that aspect, feeding all that data. And it's similar to what we did back in my work at Shell. You need all that information to be able to then understand like, is this something I can automate?
Speaker 2:What happens if I automate this? What happens I make these certain moves? And then looking at like say a digital twin or otherwise or some kind of scenario analysis or modeling. And then say like, okay, I do this, then what happens? It's the, if this, then that, right?
Speaker 2:Type of approach in many ways. That's kind of what's happening in a lot of these systems today. That's making things like automation potentially possible. Isn't always gonna be as complex as something like that, or you're seeing a lot of the complexity start to arise as innovation starts to grow in this industry.
Speaker 1:No. Absolutely. Because I I love the work, especially what NVIDIA is doing with their sort of simulation layer or their digital twin, which is which is, I think, yeah, which is a silly term. It was a but but, you know, a good marketing term. And I think people understood, and it's become, you know, a a category in the market.
Speaker 1:But but I remember a conversation I had with a couple of for all, you know, Norwegian Norwegian scientists back in 2008, 2009. And we're basically talking about digital twin wasn't a thing at that time. Right? And it was we're basically talking about how sort of sublevel physics is going to help change. If we just had the compute facilities within these within these industries, we could actually fundamentally change the way we make decisions, the way we construct these things.
Speaker 1:So really, when we're talking about rigs and to people who are sort of from the mining world who listen to our show and they use rigs in mining as well, but it's fundamentally construction equipment and heavy heavy heavy hauling, heavy moving. That's all that is. But sequencing it to get your desired results is quite a complex thing, which traditional industrial control systems just could not do. Or if you did it, it's just prohibitively costly. It's so expensive that it the economics of a project just does not make sense.
Speaker 1:But now with sort of so what we're talking about is this layer where we can predict things. It lowers the cost of automation and brings automation more mainstream and commoditized automation. And some of that work that NVIDIA is doing is actually I think it's quite fascinating, especially the Isaac Sims and the Omniverse that they've they've built, which I think is a better physics model than, say, what, you know, Meta was doing with their Metaverse.
Speaker 2:Yeah. I'm glad you mentioned, like, the physics aspect of it. I I I do agree. I think they're making a lot of strides in that way, and I think improving a lot of what's happening. Any kinda, like, heavy industry or whether it's, you know, in the mining space or only gas space, there's a lot of moving parts that happens.
Speaker 2:And especially in oil and gas, anything downhole, there's a lot of physics that happens as well. A way to get a lot of trust and adoption comes with things like fusing physics and AI. And what you're seeing now are like these physics informed neural networks or these pins, right, types of approaches that are really starting to take center stage because people understand that, okay, if I can model the real physics that are happening here, forget any of the AI stuff. Like, I understand the physics as an engineer. You know, I don't understand the potential black box that's happening.
Speaker 2:But if I kinda fuse the two, it'd be something really special happening there. That's what we're starting to see in in this space. We're starting to see also in a lot of other industries as well. So it's certainly an area to watch out for, especially as we start to see the fusing of digital twins, automation, different hardware technologies kind of fusing with the the software space, and this whole space kind of just evolving over time.
Speaker 1:Well, I think we'll have to leave it there. We're past our time for this one, but thank you so much for coming aboard. It was absolutely brilliant. But I think you've got to hold that thought. I think we need to get you back at some stage and talk to you a little bit more about automation.
Speaker 1:And the major reason why I talk about automation there is drilling is actually becoming a thing in mining. So we've been working with a couple of or on a couple of projects where they're trying to adopt the whole continuous mining techniques and integrating it with with what oil and gas did over the last sort of hundred years with their drilling techniques and get a continuous flow. And it becomes all the more interesting for things like rare earth metals, where processing isn't prohibitively expensive. It's incredibly complex. You know, you there's railroads everywhere, but it's just the fact that trying to extract it and refine it in the Western world is still not a thing.
Speaker 1:And and and drilling is kind of seen as that bridge to actually bridge the gap and make it possible. So watch the space. This is gonna become this is potentially gonna become much, much, much bigger in the next coming years if the technology evolves and then automation becomes a critical part of the whole stack. So we definitely want you back talking about some of these things. But for now, this is all we have time for.
Speaker 1:Thank you so much for your time, Sami. Really appreciate you coming on board.
Speaker 2:Yeah. Tito, pleasure. As always, thank you again.
Speaker 1:Take care.