The State of AI and the Prospects for Industrial Automation (MD&M West 2026)

Leonard Lee, Executive Analyst at neXt Curve, was invited by Industry 4.0 Club to present at MD&M West 2026 to talk about the state of agentic AI and the prospects of the technology for industrial automation.  

Agentic AI has been a hot topic since the beginning of 2025 fueled by the promise of autonomous automation of operations and business for all industries. Yet, much like generative AI overall, agentic AI solutions have turned out to more challenging than anticipated and forecasted. What's the deal?

The session will address the following questions:
- Where are we today with agentic AI in manufacturing and industrial applications?
- Where is agentic AI landing?
- What does industrial-grade agentic AI look like?
- What are the leading practices for applying agentic AI in manufacturing?

Please subscribe to our podcast which will be featured on the neXt Curve YouTube Channel. Check out the audio version on BuzzSprout or find us on your favorite Podcast platform.

Check out Industry 4.0 Club at 

Also, subscribe to the neXt Curve research portal at www.next-curve.com and our Substack (https://substack.com/@nextcurve) for the tech and industry insights that matter.

NOTE: The transcript is AI-generated and will contain errors.

DISCLAIMER: This podcast is for informational purposes only.

Transcript

Leonard Lee 0:02

Morning everyone. before I get started, I wanted to thank my and Tim for inviting me here, to MD and M West to present on this topic. And, just before I get started, just a forewarning. If you're expecting a presentation where I'm hyping up AI and gentech ai, you will be sorely disappointed. that is not what we're gonna be talking about today. We're really gonna be talking about it from a grounded perspective, the state of AgTech ai and what really is the prospect, for this technology? This, compute paradigm, for industrial automation. yes, I'm Leonard Lee, executive analyst at Ncur, and I've been in the industry for. across a number of industries, actually through my consulting career that spends over 30 years. formerly a, partner principal at I-B-M-P-W-C, as well as ey. and then, started off my career in, industry research with Gartner. was with them for three years and I started my own firm and, what I focused on. Is basically charting the tech oceans, right? I'm not trying to boil things, I'm just trying to figure out where things are and for those of you who are, big iot folks, I'm just trying to get a sense for how things are shifting and ebbing and flowing, because as everyone knows, technology, especially the stuff that's happening in a AI domain is very, very fluid. And, what I want to do here. For this session really quickly, and I'm gonna try to get through the slides as quickly as possible so that we can open up the discussion to q and a is, to help everyone understand the current state of Ag agentic ai, and the state of the art. Okay? I'm sure everyone's heard of AG agentic ai. Ai, ai, AI for the last, I don't know, decade maybe. and then I want everyone to establish a better understanding of. Agentic ai, what do we mean by that? Right. because the definition has morphed over the past three years, since the advent of, chat GPT and, large language models and then just level set on what do we mean by ai because there's a lot of different varieties of ai and I think that clarification's really important. then we will talk about, the limitations and the challenges of AI and agentic ai. As observed by Next Curve and, finally understand what is the, value, potential and opportunities in manufacturing. Ultimately, what I want to do is arm, whoever has lent their ears to this presentation, the ability to ask better questions, right? Because there are many questions that should be asked that are not being asked, and that's costing a lot of organizations, a lot of money. And a lot of spinning wheels and for vendors out there, hopefully by the end of this presentation, you'll have a better perspective on what the path of value will be for your customers as you, work with these emerging technologies in improving the outcomes that you are able to, present to your customers. So, really quickly, if you follow me on, social, you might think that I'm an AI pessimist. I am not. I call myself a pragmatist, but then also I'm not an evangelist. I don't go overboard with a lot of the prevailing narratives that are out there today that are being incrementally, if not dramatically proven misguided. what I really want to do is help everyone. through all these different cycles of technology to, stay out of POC health, because that seems to be something that we keep falling into, right? investing in technology experiments, finding out what the limitations are, and then, if you're lucky, getting out early, but oftentimes you get stuck in this POC hell. So let's talk about what is Agen ai. What is an agent? And for the industrial sectors, what basically means is taking your Six Sigma operations, taking your six Sigma processes, the six nines of reliability, and then introducing less than one, nine, right? That's what this stuff is about. And this is one of the reasons why, we're seeing so many challenges adopting, Gentech ai. So, what I wanna do here really quickly is talk about the genealogy of a Gentech AI and automation because, This stuff is not new. Many of you in the audience have already seen this stuff before. there's these things called chatbots. They used to be called agents. early on about a year and a half, if not two years ago, when this whole notion of agents started to emerge off the back of, this excitement about large language models. When people were talking about agents, they were talking about chatbots that were being used in customer service, right? And so a lot of industries like telco, retail, they're experimenting with chatbots with, very varying success. so you have this, agent layer that's evolving. Okay. And then below that you have the models themselves. The model architecture is evolving. And then below that you have the different modalities of what you have to do in order to, build and operationalize these, AI technologies evolving as well. Right? And so there are these three swim lanes as you look at AgTech, that you really need to consider as you evolve your own thinking about where we're going with this Ag agent stuff. And so going back up to the top, you guys are probably familiar with RPA, totally not foreign workflow management. These are things that you guys are familiar with. These are now being married into what are called agentic frameworks, where now there are these plugins to large language models, reasoning models, et cetera. And so below that, with the models, you're seeing architectural changes, right? We started off with large language models. They weren't good enough. They weren't reliable enough, what did the AI folks do? They pivoted to mixture experts, which was represented by GPT-4. that was the first largely and widely used, mixture expert model. Okay. And what that basically is, is the number of smaller domain expert model stitched together under a. Let's call it an orchestrator model. Okay. And without getting into too much detail there, what you then saw were these multimodal models that combined vision, with text as well as audio. Right? So there was that new, generation of excitement everyone had. And that includes applications such as, SOA that came out that wowed everyone. This is LLMs being mixed together with diffusion models. which generate images, sound, and so forth. And then last year you started to hear a lot about reasoning models, chain of thought, chain of action. And so this new class of, models that, and architecture that enables long thinking as they call it, those emerged. And then now you have a Gentech ai, right? And so you have these. Automation frameworks, from the chat bot, continuum to the AgTech frameworks come together with what's happening with all these model technologies and architectures. And now you're seeing a lot of the ISVs like Microsoft and others, build out all these frameworks help organizations build these agents, and below that, we've moved from post training. Now to, rag. Rag was actually something that became big a few years ago, actually three years ago, because there's a thing called context, right? These models didn't have context. They weren't up to date, and so oftentimes the results were irrelevant. Inaccurate, what have you? And so the idea was, hey, let's plug these things to our data stores and our corporate data and knowledge bases, and then improve the outputs. And in some cases that did happen, but they proved problematic. And we'll get into that in a little bit. And then now we have to deal with this thing called MTP, which is the mocon protocol, which I maybe many of you have already heard of or have heard of. This is now a way of providing context to agents. And so this is a, I put this together just so that everyone can get a feel for how all these different threads are coming together to arrive at a agentic ai. And so a lot of you have probably heard or read or, experience what is implied in the, MIT Nanda study, right? That came out. I think people were freaking out about this thing about six months ago, if I'm not mistaken. But the question is what is the state of ai? The suggestion by this research study by MIT was that the success and the outcomes expected of the technologies falling short, right? And we're seeing this, across a wide range of industries. And then, at next Curve, if you go and check out. My website, follow my research, you'll realize I cover a lot of stuff. So I see a lot of stuff, including what's happening in manufacturing, industrial, and adoption is challenging. one of the bright spots though, before we get into all the realism here, is that, ML is trying to gain some traction is benefiting from a lot of the. Innovations that are happening with large language models, especially from the compute perspective. so just keep that in mind. not everything is doom and gloom, but the things that, most industries have been working with that used to be called ai, which is really machine learning, is starting to find its home, which I think is a bright spot. But let's talk about LMS really quickly. a lot of people probably won't remember this presentation, but. I would suggest that you take note of this slide and go back and listen to Andre. He is a open AI researcher who did a presentation at Microsoft Bill 2020, three, and his presentation was a state of GPT. And if you look here, the use cases. He listed out certain things, that are challenges so he has stuff like, models may be a bias. Models, may fabricate models may have reasoning errors. Models may struggle in classes of applications such as spelling, et cetera, maybe even math. models have knowledge cutoffs to a certain extent. models are susceptible to prompt injections. So these were early. recognitions of the limitations of the technology. And guess what? A lot of them haven't been solved because they're fundamentally rooted in the limitations of the technology. And this is something a lot of industries ignored for the last three years. So, bias, hallucinations, we've now heard from, companies like philanthropic that, they can also be, prone to deceit their alignment issues. And most recently, I think a lot of you might have heard of what else, I think called Bolt hook, right? With open crab or whatever this is a perfect example of how alignment, whether it's intrinsically generated by the AI or human induced causes problems with the application, especially if you start to create these high intelligences, if you will. There's also drift collapse, what I call AI dementia. These things decline over time. They have to be maintained. Consistency, reliability issues, and of course reasoning errors. Going back to that one, chart that I showed you with the genealogy, that's why you're seeing this fast evolution of model architectures. Let's talk about AG agentic AI for a moment, because AG agentic AI depends on AI or gen ai, there's a lot of good ML stuff, so all of you guys out there doing the good ML work, I'm not hating on you. I love you guys. The thing is AG agentic AI pose, this is now another series of issues, number one, conflict and collision. how do you orchestrate these things? How do you. Ensure that the policies that are instilled across a corpus or a collection or a hive of agents don't create collision and conflict with each other. testing and validation. How do you test these things? This is still a really big problem. Uh, observe observability. This is another big gap, chain of confusion. You probably haven't heard that before, so if we go back to reasoning models, chain of thought, there's a phenomenon that I've observed several times. I call it a chain of confusion because these things can confuse themselves and get into an infinite loop cycle of, long thinking. And guess what? That costs money. Then there's the security identity issues that everyone's trying to figure out. there's the oversharing problems, okay? And then there's a cost variability based on tasks. So if you try to cost out how much AI compute you need to complete, let's say a specific task, you can't baseline that. 'cause you don't know how many cycles that agent actually going through to create that output or decision if that's what your expectation is. So the bottom line is this. Gene AI enabled AgTech AI for the industry, including manufacturing. most folks are in the process of experimentation, right? There's many mistakes and failures. That's all part of the POC process that I was telling you about earlier. this is why there are increasingly, there's increasingly more talk about how a lot of these early POCs have FA either failed. Or continue to be problematic in terms of, delivering on expected outcomes. And then, there's been learning, but then there's also a need, and it's something that I'm telling a lot of my clients. There's a need to unlearn a tremendous amount of, stuff. What you know about AI today, especially generative AI is wrong. And so now you have to go through the process of figuring out. What did I believe about AI before that is no longer true. And so that's a really difficult exercise that a lot of organizations are starting to go through. And that realization or that initiation of that process really happening through these POCs. And of course there's a big write off of, you know, these experiments that are going on. But the biggest question for manufacturers out there as you're looking at industrial automation is, Should we be deterministic or non-deterministic? how should we place ourselves in this continuum? And it's not a black and white answer. It is really, you have to look deeper into your operations, your systems, architectures, and figure out, okay, what are those things that we know are deterministic that support the quality of production that we have? Today and then figure out where can we introduce, more of a probabilistic but augmenting capability, leveraging large language models or agentic ai. within the greater context of, the system, because oftentimes we think about the task or that specific technology, but at the end of the day. Reliability, throughput, quality, all these things are an outcome of the entire system. And, there's a lot of folks that tend to forget that for some reason. So should you go AgTech or algorithmic, programmatic? And quite frankly, I don't think there's anything wrong with the old stuff. And eventually each of you who are, looking at, continuously improving your operations or manufacturing. your business are going to be, trying to figure out what is that best blend, and so that's gonna be the discovery process that comes next after you've unlearned a lot of stuff, and so the holy grail of Ag agentic AI is really autonomous, autonomous control loops, either down at the sensor level, your perception layer all the way up to, management orchestration, maybe even arguably into your ERP. So the goals here, ultimately with what I call the autonomous aspiration is self-healing, self-learning, self optimization, intent driven or intent based. operations prescriptive, zero touch operations, right? We wanna take people out of the loop, but of course, that's. Really challenging, but hey, let's talk about where is AgTech AI being applied in manufacturing today? And so I did an interesting exercise. You know, some people tell me, Hey, you don't use AI enough? I said, okay, maybe fair enough. I don't know. So I went ahead and asked Google, Gemini, and perplexity, well, where is it being used? It spit out these areas, right? predictive maintenance, quality control, digital twin integration, inventory management, supply chain optimization, autonomous production, scheduling, energy and sustainability. so that's what these two models presented to me. And so I went back and looked at the sources, and guess what? It's all marketing material. So, why did I go through this exercise? Because I wanted to demonstrate and also confirm for myself how detached now these models are from telling you what the real answer is. Right now, the adoption of AgTech AI is, the maturity of implementation is very low. It's barely out there. if you are. Deploying AG agent AI out in your operations today, I would tell your ciso, go and check out what they're doing and get it validated. so everything that I've presented thus far, these are hard realities. But for the vendors out there, the solution architecture, text within the organization, these are things you need to know about. 'cause going back to the POCs, this is the reason why it's not change management and all this other stuff. These are the reasons why POCs fail and it's rooted deeply in the technology. let's take the example of telecom industry. They've been looking at trying to leverage large language models as well as EnTec AI for years now. And some of the prominent use cases they are still working on is autonomous network automation fault and root cause analysis, RAN or radio access network, and network management orchestration ran and, network optimization service operations for, fault to fix. And after all these years, there's still. Working on it. These are hard, hard problems that they're dealing with the technology to realize some of these aspirational use cases. I'm not saying there isn't progress, there are patterns for safe, reliable uses of agen ai, but there's also, certain dynamics that you need to be aware of that we, I'll cover maybe, briefly. Toward the end of the presentation. Now let's talk about the semiconductor industry. So one of the industries that I cover is the semiconductor industry. And as many of you, probably know, the semiconductor industry is probably one of the most intense, manufacturing environments out there, especially nowadays. Given, semiconductor manufacturing is getting such a spotlight some of the prominent agent AI use cases they've been exploring as generative design and prototyping. Validation and, test automation simulation and synthetic data generation. diagnosis and root cause analysis and software and firmware coding and testing. So, there are some interesting things going on, Especially on the coding front, where they're trying to now take PLM. Processes as well as SDCs and try to reinvent those, leveraging agentic frameworks. you still have to have the human in the loop from what I've been able to observe. one of the things that I'm seeing with the semiconductor industry is a lot of the generative functions that seem to have promise. Are Design, and then of course marketing. But you guys don't care about marketing. You want to know how can we leverage this technology to improve our, operations as well as, enhance our industrial automation capabilities. It is a slow, they're going through a discovery process. There are patterns coming out that, look promising. now the challenges of LMS and agent for, industrial, number one, latency. LMS are not magic. They're really tough to work with. Even SLMs latency is a big challenge, oftentimes brought up quite a bit when you look at, at manufacturing solutions. The cost of intelligence, which tends to be pretty invisible until you get the bill. reliability and consistency is still a challenge. And so a lot of the scaffolding and, framing that you have to do to have safe and reliable agent and, generative AI applications, it's evolving. Those frameworks are not baked completely yet. But they're coming. Even though I'm citing as a challenge, the complexity of lifecycle management, this is not fire and forget, okay, you have to maintain these things. And oftentimes that maintenance cost is not factored into total cost or ownership, calculation. Then there's the knowledge and domain specific and specificity challenge. We're finding out that more, industries cannot use the generic, models, they need to have domain specific models that they develop. And then I put this last, but this should be First Security. biggest, biggest, biggest, biggest, biggest, biggest, biggest problem. And then the potential though for Elon's and AgTech ai, sweet spots for industrial, I think Diagnostic and troubleshooting support, there's benefit there. just have to be really careful and test a lot. non-deterministic automation, if there's any aspect of your process or operations where determinism is not as critical, there's a lot of potential there. It's just you have to figure out what those aspects are. contextual, sensing and understanding. This is an emerging area. This is where AgTech, as well as small language models, especially at the edge, are going to help improve the perception and sensing capabilities in your industrial environments. intent driven orchestration is still aspirational, but people are starting to realize these in more of like a function task level, right? And that, that could still be very, very helpful. It's just that you're not going to see this big magic, transformation in all likelihood, Knowledge and domain specificity. the reason why this is important is that, there are ways to get these things more tuned into your, your business. You just have to be cognizant of the cost, and it's also being used in, data analysis and augmentation. So not necessarily processing your data, but helping you figure out your data. And also your knowledge base. And so, um, recommendations through your Gentech journey. focus on building your network graph for your ops and assets, So if you're a digital twin fan, this is probably one of the first things that you really need to focus on because Gentech stuff is really dependent on. A really solid foundation. Otherwise, it's just not going to have that scope of, reliable impact in whatever kind of transformation you're doing. Investing your knowledge bases and knowledge management. This is really, really tough for a lot of organizations, but it's essential because the AI's AI is not the problem. Ais react to data. They need to be trained with knowledge so they can make decisions off of data, right? And so that's something really important to understand 'cause everyone keeps saying, oh, well, you know, it's a data problem. No, it's not. It's mostly a knowledge problem and most organizations haven't harvester knowledge. And then understand the full lifecycle of agentic management. establish AI and agentic ai finops, observability. And analysis. This is very immature area, but this is something you guys should really focus on because it's going to be essential in, help you realize ROI and also, AgTech ai, solution design and be wary of gen ai, you know, Gentech AI evangelists kick the tires on them. Okay. and I'm really serious about this. I've seen some scary, scary stuff being presented on stage. I mean, absolutely scary. And then learn from others' mistakes. 'cause there are many, you don't have to make the mistakes. Just observe. Being first hasn't proven to be good for anyone in this race. if you're interested in my research, just, uh, check out Next Curve. I, a substack. Also my site, which is www.next curve.com. I also have a podcast that I do on a weekly basis where this is like, Wayne's World and SportsCenter, IOT and tech broadly. It's called IOT Coffee Talk. check that out. Now we're gonna move into q and a, but, follow next curve for the tech and industry insights that matter. Thank you very much.

Mike Ungar 26:50

next letter. moving to the first question for Leonard. All right. We'll start here. Thank you.

Leonard Lee 26:58

I know you.

Audience 26:58

Yeah. Thank you so much. When you talk about ai, the gen models are only one aspect of it. Machine learning.

Leonard Lee 27:09

Yeah.

Audience 27:10

And, how do you relate these different AI is in an approach to a production line in the industry.

Leonard Lee 27:22

Wow. That's a big question. the first step is really having a good sense of the taxonomy of what is ai, right? I think that's one of the biggest problems right now. It's been a highly appropriated term. So before, early on we used to call it gen ai, this thing that's related to diffusion and LM models. Now everything has been packed into AI and there's I think, less of an appreciation of how diverse. AI really is. And so one of the first steps is to kind of break that out and be a little bit more specific. Don't generalize how you're talking about ai. Be specific. because down at the, lower levels of your, tech stack closer to the sensor, you have one variety of ai, mostly machine learning that's happening. And then as you move your way up, you might start to see more of these robust. AI models that are leveraging generative AI technologies start to make their way in. Now, there is a trend, that I'm seeing where, small language models are starting to make their way much closer to the sensor. This is still early days, but the silicon is starting to be there. it's emerging. And so the placement of a AI. whether it's machine learning or generative ai, it's gonna be shifting. I guess the most important thing here is don't generalize ai. Hopefully that addresses your question. At least. That was my reaction to your question.

Audience 28:59

It's a high old.

Leonard Lee 29:02

Oh Yeah. And then the age agentic thing is almost separate. it's on those two tracks. Right.

Audience 29:09

Have question. You mentioned something your other about how you've seen some scary things presented onstage, also have a lot of anxiety related to like, what's going on with ai. Could you go more in depth into what you've seen that. You feel it's like kind of scary.

Leonard Lee 29:27

Oh yeah. hey, how do we create an agent and give it our credentials for our bank account so that it can ally go out and buy stuff for us? I don't know, that kind of scared me if we have millions of people, looking to do that without any idea of what a agentic. AI really is, or what AI is. Isn't that scary? I think that's pretty scary. and then the entire room went, yeah, yeah, that's a great idea. How do we do that? Are you gonna talk about that? end of story. this is what's happening outside of the tech bubble, which is really frightening. And there are companies that are doing that too. But again, going back to the POCs, this is what they're discovering. Security is the number one issue. Privacy, confidentiality. These are the number one issues.

Audience 30:21

Alright, thank you for the top. my question is around, what you mentioned the, it's not a data problem, it's a knowledge problem. Can you go into a little bit more specifics around how. Companies or organizations can, either extract or get the

Leonard Lee 30:41

Yeah, I mean, it's kind of a age old problem. I used to do a lot of knowledge management back in the day for industrial as well as oil and gas. it's a practice that's, not mature in a lot of organization. It boils down to being able to support and, hydrate an ontology, right. so you hear about these graphs and, you know, they've been trying to use vector databases. a combination of these two types of, uh, quote unquote data store technologies, to support large language models. The thing is, to be able to hydrate these things well with domain specific or company specific information, you have to have a good knowledge base. And so it really requires that you do the really boring, mundane task of giving that stuff together. Otherwise, these models will not even know what to do with the telemetry and that's coming in from your sensors. Right? and it's as simple as that. You know, I think here's the thing. We know what the problems and the solutions are. We're being distracted from focusing on those and then looking a little bit too far down the road. And not focusing on the foundation that you have to build. So that's really what I was trying to articulate there. And then, you know, data has its own issues, right? And a lot of that has to do with the fact that, we're still trying to instrument things. We're still trying to hydrate that digital twin, which is really gonna define the footprint for what you can do with agen ai.

Audience 32:13

So there's a ton of, hype out there about, autonomous robots in manufacturing and in listening to what you're saying, it seems like you're still a long way away and the genetic AI for being able to allow those robots to make decisions as to what they're gonna be doing on the manufacturing store.

Leonard Lee 32:32

Well, it depends on what the. The A GV or, you know, the robot does, right? If it's more task specific, it's not general, then there's plenty of our, there's a bunch of companies out here, right? Doing that. Now, if you try to make a general purpose in humanoid, which is probably the suboptimal form factor actually for manufacturing, you're gonna run into a lot of problems. That's not focusing on the problems to be solved. That's. Putting science fiction in the top of your technology agenda. and, and I talk to a lot of real pro, robotics guys and they will paint a very different sobering picture than what you're hearing from the hype narratives. So that that's why, you know, I told you when I get up on stage, this is no nonsense, right? I'm I, my company, I do not deal BS is shown the door. Because, you know, I'm about helping you guys avoid the POC but I also don't want to discount the good work that a lot of people have done where there are solutions that are being disregarded or ignored at the moment because of the hype narrative.

Audience 33:47

I really enjoyed your talk. so there's a lot of things you will fail at as you go on this journey.

Leonard Lee 33:53

Mm-hmm.

Audience 33:54

But I do think that we will get there eventually. What are some of the good bets that you're saying? Because, you know, I've been on this journey for a period of time.

Leonard Lee 34:02

Yeah.

Audience 34:03

And then there's a lot of positive returns and ROI outside. I'd love to hear some of those examples from you.

Leonard Lee 34:09

Yeah, I mean, it's, right now we're on a journey, as I alluded to before. I think there's a lot of great stuff happening with ml, the old school stuff that really struggled in the previous, phase of this Uber AI hype that are starting to exhibit, some success. A lot of it actually is augmented by. generative ai and going back to what I was saying before, using generative AI to contextualize all that data that is non-con contextualize, right? And then being able to work on those basics and using it as a tool to do that. But then you're gonna be feeding in all likelihood, deterministic system, and so, you know, you use these tools for what they're good at. You don't try to position them for something that they're gonna fail at. And then to get a sense of how they're gonna fail and how you're going to waste your time and your company's money is, by focusing on and understanding the limitations of the technology at any given point in time. But then having that visibility to the roadmap. And that's what I do. I'm looking at the stuff all the way down from silicon manufacturing, all the way up to a number of different end markets. So that's how I get my land. So

Audience 35:26

excellent. Thank you Leonard. question, when you talk about the AI journey to the network graph, can you explain a little bit about the network graph and like, landing with business outcomes or whether it's on the operation shop floor or,

Leonard Lee 35:38

yeah, sure. Sure. Um, yeah, let me see if I can generalize a response that probably shouldn't be generalized. It's, all about, dealing with the new, like, it's not necessarily new because graphs have been around for a long time. it's really understanding, what the knowledge architecture to support AI looks like. more often than not, you're not gonna be using RD BMSs and structured data. You're gonna use some kind of network representation of things. this is the education maybe the organization has to make. Now, the digital twin folks have been looking at this for a long time because their thesis was, if we can create an ontology for digital twins, then it can be the basis for scaling and enabling ai. And I don't think that that was wrong. For a lot of organizations it's foreign because most organizations are used to RDBMSs, right? And, all the goodness that's associated with that technology for use in industry. But these graphs are different. and this, again, going back to security, it ends up being different because the graph architecture, the vector data, vector architectures present. A huge, security issue and limitation. So you have to be very careful how you architect your system. And what I mean by system is systems of systems.

Mike Ungar 37:08

We've

Audience 37:08

got

Mike Ungar 37:08

time for one

37:09

last

Audience 37:09

question.

Leonard Lee 37:10

Okay.

Audience 37:12

Thank you, Larry.

Leonard Lee 37:13

Oh yeah.

Audience 37:13

I picked up on the word they mentioned about maintenance costs. let's say 10 years down the line from now, some companies are gonna be operating. Agent AI system to the manufacturing process, what kind of, unknown manufacturing or maintenance costs do you think would come up from Agen AI that doesn't exist today?

Leonard Lee 37:34

I think, my problem is that I already see this stuff. for people who didn't just listen to the last 30 or 40 minutes. they're not gonna know, they're gonna discover all of these things, right? I mean, that's a simple answer. The level of awareness of what I just presented to you guys is very low. After three years, and I've been working with some industrial big industrial, brands and names on technology roadmap and strategy, this gets ignored. This has been ignored for three years. I'm going back to unlearning. There has to be an unlearning. You will unlearn whether you like it or not. Depends on how big a price you want to pay.

Mike Ungar 38:16

Well, on behalf of the four, I want thank Leonard again for this wonderful talk. We, we given big him.

Leonard Lee 38:22

Thank you. I would.