Can AI Upgrade the Factory You Already Have? | SDA × FluxionSystem

Episode Summary

Factories already contain the machines, controllers, and data that AI needs to improve operations. The hard part is reaching the control layer safely and making the economics work. In this episode, Michael Finocchiaro talks with Dr. Josef Waltl of Software Defined Automation and Albert Klein of FluxionSystem about what happens when AI moves beyond dashboards and scheduling into the machinery itself.

Waltl explains how software-defined automation gives engineers modern access to shop-floor control systems, including the software and logic in existing controllers. Klein describes FluxionSystem's approach to using AI to create machine-control systems and find improvements within installed equipment. Their perspectives meet around a practical question: how do manufacturers improve an existing factory without an expensive rebuild or a consulting project that costs more than the opportunity?

The discussion covers factory data ownership, AI agents, controls engineering, machine context, the pace of industrial innovation, and the point where human judgment must remain in the loop. This episode of AI Across the Product Lifecycle is sponsored by Aras.

Guests

• Dr. Josef Waltl — Founder and CEO, Software Defined Automation
• Albert Klein — CEO and CFO, FluxionSystem

Watch the episode

Watch the full conversation on YouTube.

Full Transcript

The transcript below follows the edited episode. Speaker names and technical terms have been lightly corrected; wording remains close to the original conversation.

Michael Finocchiaro (00:00:19)
Hello everybody. This Michael Finocchiaro on the AI Across the Product Lifecycle Podcast. it's been a a while, big long summer break, and then I've been traveling. I've been to Chicago and San Francisco and LA, but I'm back now talking with two of my friends, both German guys, even if one of them doesn't live in Germany anymore. welcome to Josef Waltl and and Al Albert Klein. Josef, you are the head of Software Defined Automation who's working on the left side of the ISA 95 pyramid, I think is the way you describe it. And Albert, you're doing an amazing job upgrading machines in place in order to get to the AI age in manufacturing. So I find these both really fascinating. Josef, you wanna first describe who you are and what you do?

Josef Waltl (00:01:02)
Absolutely. So actually I'm Austrian, but that's a small detail. German speaking.

Michael Finocchiaro (00:01:05)
shit, sorry. German speaking. German speaking.

Josef Waltl (00:01:10)
actually I'm an automation engineer by trade. I learned automation engineering and PLC programming when I was in my teens on B&R controllers, being from Austria. I spent some time at Siemens, I spent some time at Amazon Web Services where I manage the overall industrial software portfolio globally for them. And some time. Good. And then

Michael Finocchiaro (00:01:34)
Sorry, my bad. Start over.

Josef Waltl (00:01:36)
and then I also spent some time as product manager in Azure IoT. And what we are doing is very simple. We make the computers in the shop floor, which are the PLCs, the controllers, the motion controllers, the robot controllers, behave like modern computers which are in the cloud, where you can have an API for every operation that you need. Because ultimately that brings innovation and adaptation. And that's what we do. there is names for it called industrial DevOps. We are supporting OT cybersecurity. And especially I'm excited that now we can do much more now that the AI models have gotten very powerful on the actual control code of the factory, which is the core IP of many machine builders and manufacturers.

Michael Finocchiaro (00:02:30)
Awesome. Thank you. you know, we're already getting comments on the on the chat on LinkedIn, which is super cool. it was my fault, the audio, because I forgot to mute before I opened the tab. So my I'm sorry, Josef. Albert,

Josef Waltl (00:02:42)
Okay.

Michael Finocchiaro (00:02:43)
tell us a little about yourself and your your your your journey at FluxionSystem.

Albert Klein (00:02:48)
So I'm Albert Klein. I'm CEO of a company I founded with some friends not even a year ago. four friends. we've been in manufacturing forever and mostly always advanced manufacturing. So my master my last job was in 3D printing or additive manufacturing. And so we've been grappling with the fact that if you want to get AI into a factory, A little bit beyond the organization of the factory and and you know, scheduling and and logistics and these type of things, the ERP,

Josef Waltl (00:03:23)
Yeah.

Albert Klein (00:03:23)
MES, you've got to get it into the control system of the machine. And that's what we've been we've been addressing with FluxionSystem, developing an AI that creates such a control system. So we don't wanna we don't we think it's gonna be too expensive if a lot of engineers start thinking about how should we design a control system for or a machine that includes AI for the for the machine control, the process control. So we we think an AI itself could be a better way to do this.

Michael Finocchiaro (00:03:59)
I think it's pretty fascinating 'cause it you know, it's obviously a lot easier to change that out than to move a three ton machine from its place on the on the floor, right?

Albert Klein (00:04:08)
Well, it's as I say there is the law of diminishing return of anything you do in a factory. Initially you can f pick the low hanging fruit and improve quality and reduce cost, but it gets harder and harder and then in the end you have to do something big, like buy a new machine or rebuild or change the layout or whatever, introduce an ERP. if you are

Michael Finocchiaro (00:04:30)
Mm.

Albert Klein (00:04:31)
if if you want to go beyond this, we think AI can actually make a difference. A I can find a wealth of improvements inside the existing infrastructure. And that's the job.

Michael Finocchiaro (00:04:44)
So, Josef, in twenty, you know, November twenty twenty two, we had this crazy OpenAI moment, right? Which, you know, everything changed pretty much overnight and it's really hard to imagine even what it was like before sometimes. were you initially very skeptical or very bullish on the open AI moment? Did you think immediately, my god, this is gonna change manufacturing, or were you a bit more let's see how this goes before we before we start implementing it?

Josef Waltl (00:05:13)
First of all, we always had auto improvement in the foundation of the idea of software-defined automation because as Albert said, there is not enough engineers in the Western world. There is not enough engineers and talent to cope with the increasing challenges and volatility a manufacturing system sees, be it energy costs, be it production schedules and so on. what what really has changed is this now after the ChatGPT moment, we do not rely anymore on purely machine learning. We have systems that understand systems automatically. And I give you a concrete example also related to what Albert said. Well three weeks ago we had a conference where we showed on live something rema remarkable. We had a packaging machine with two conveyors, two motors. We did a backup of the software, the AI agents learned what the application is. So they came back from the software, hey, that seems to a packaging solution where we have a Siemens controller with two SEW motors and they seem to hand a load between the two packages completely alone. And I think this is really what

Michael Finocchiaro (00:06:30)
Mm.

Josef Waltl (00:06:30)
changed because as Albert said, if I have a use case for ten K but it costs me a hundred K to have consultants. Come in and understand everything and then do the tag labeling and so on, then it's tough. What we also showed on stage is that EI came up with an auto proposal on how to reduce cost for this application. Very simple example, just let the second conveyor not run so long after the package physically has been handed over. What we still do is We say everything that AI proposes, an engineer needs to approve, you have approval workflows and so on. And I think for the time being that's the right thing. What I can see in the mid-term future is that we get to a engineer gives certain levels of you can optimize within those levels, and then you can outdo do out to iterations. Like giving a junior engineer or an engineer generate a task and say, Okay, you can Change everything, optimize it within these boundary conditions. What I cannot see right now is that we have completely out to optimizing factors. I think there are too many questions, are unsolved. And frankly, also we are dealing with physics. Physics have real consequences. So we really want to, at least in everything we build, keep the engineer always in the center of the decision cycle.

Michael Finocchiaro (00:07:57)
Until the s things are deterministic enough to to remove it, yeah. Albert, what about y Go ahead. For the people.

Josef Waltl (00:08:02)
I think this will come this will come from the people. We would not force it. People will at some point in time decide am I confident enough? And our job as technologists is to provide the systems. But people who are in charge of production lines will at some point decide, Yes, I do. As we decide at some point in time full self driving is good enough, I I I basically take this option.

Michael Finocchiaro (00:08:32)
I like that example 'cause I was just in San Francisco and LA and and it was just incredible to see five or six Guaymos all one behind the other at a tr red light, you know. how about in your experience, Albert? Were you really enthusiastic about open AI or maybe you were a little skeptical?

Albert Klein (00:08:49)
Well, the first thing I was, I was depressed because I I I realized that one of the defining traits of humankind, the mastery of language, was now passing on to machines. And I realized quickly that machines would be able to manipulate language better than us. And that is depressing. but then I overcome came my deep depression and said, Well, how am I gonna use this for How is it going to improve my capability of doing something? And I think that's generally what we have to ask ourselves using AI is what am I good for now? That some of my talent is now assumed by a machine. And there is plenty left, I'm sure. And so but I and and and I fully agree with with Josef that human in the loop is one of the fundamental principles of automation and factory. There are The the other one is as as as you have said, is that AI, because of its language capabilities, understanding systems when they are documented in language, and software is nothing other than language. And so but I we think that there is another dimension that we need to understand in in manufacturing automation is the process, the physics process. we we are we are amazed and encouraged by lots of people collecting lots of data to understand the world around them. But it just describes the world around them. It doesn't understand how it really functions. So in machinery control, we need to sometimes go beyond the process that we that we are currently implementing in a machine. We need to deal with new materials, new assist gases, extreme temperatures as we've seen this summer and this type thing. And your existing data does not allow you to do that. You can interpolate, but you can't extrapolate. So physics are a good map to guide you into the functioning of a machine. Of course, the physics are happening inside the material, not inside the machine. And software, when you get it from a PC, doesn't contain information on what's happening in a material. It contains information on what the machine does to it. But those two things are giving you the complete picture. So that's what Fluction System has done. It's built a garganton physics model that understands every Behavior of material in machines. It understands the melting of metal. It doesn't matter whether you weld or whether you laser cut melts. So and it combines this digital twin of the material with a digital twin of the material with a machine, which comes from the software mostly, and it creates a new control system. And of course, it uses the open and i and the clouds, etc. to To develop new code for the same, even though I'm s there is a problem with that, you you you can't use it to create code inside the factory to render the product. So because you don't know its cost. We do not know the price of tokens in the future, and we don't know how many tokens we use for a task. If you've benchmarked your your your claude, you know that it for the exact same problem. It delivers a different answer, different token usage every time it does solve it. So that's a big, big if we have with the great new world of LLMs. You don't know their cost. If you don't know what something's cost, you can't employ it in manufacturing.

Michael Finocchiaro (00:12:25)
That's a really cool perspective. I hadn't heard that one before. I like that. how how has it changed? Because you guys both manage teams of developers and of course coding has changed forever since about what eight eighteen months or so of Cloud Code and anti well, I think it was anti gravity probably gets credit for the first one and then Cloud Code kind of rolled everybody over. So how has that changed for you guys both running companies with lots of developers? How has that change the way you l you develop code and and the way you structure the the development organization, the R and D.

Albert Klein (00:12:59)
Well, let me start

Josef Waltl (00:12:59)
So

Albert Klein (00:13:00)
as a so a little bit easier for FluxionSystem. We never had developers. So we're only eleven month old. So we ran straight into this into this enormous increase in productivity in coding. So we're coding ourselves now. Well what coding means has also completely changed. It's everything is kind of like the old high level programming languages now. basically it's having the idea for an application is what counts now. So your your knowledge of non-coding stuff, of the use cases and this kind of thing become much more important. So, for example, we realized very soon very soon that we couldn't develop a control system without really creating a digital work instruction software, which we hadn't planned on doing. So we we looked at tulip and safety culture and all of these packages and said, hey. If somebody wants to use a machine

Josef Waltl (00:13:55)
yeah.

Albert Klein (00:13:55)
with a physics control system has to enter values, we need to tell them how to do this. And if we really want people who are not super experts to use machines, they need step by step instructions how to use a machine. They need training before they are allowed to use a machine. They need to know how to use safety instructions and all of these things. Imagine today you're still going to a human, to another human to learn how to use a machine.

Michael Finocchiaro (00:14:17)
human the other human to learn how to use the machine.

Albert Klein (00:14:21)
That's not necessary. The other human doesn't know as well as the machine what it needs. So we have developed a s a full suit of work digital work instruction software, including training, compliance, quality management, all of these things entirely ourselves without any developers at all. So and that integrated into physical control system

Michael Finocchiaro (00:14:21)
And that's not necessary. The other human doesn't know as well as the machine what it means. So we have developed a full suit of well which is what subject of city training, compliance, ISO quality management, of these things entirely are in development control. So that would be based into the physical control system.

Albert Klein (00:14:45)
makes quite a powerful you tool out of a machine, any machine.

Michael Finocchiaro (00:14:50)
Pretty radical approach. Josef, you had something to say too on this one.

Josef Waltl (00:14:55)
Absolutely. So, first of all, where we are. We are the first software layer on top of existing industrial controllers of the actual machines. And what we also see, no one rips out complete machinery for AI. First of all, you need to really work with the existing equipment. Sometimes this is also weird, sometimes even decades old. So what we see is we need the developers because we have the iteration with the customer. You don't know how a 20 year old PLC behaves in every situation unless you have these edge cases. What we do see is that our developers almost doubled their productivity in the last 18 months. Also in this very hardware adjacent topic, so massive improvement of productivity. But there's one additional thing that I want to mention is our customers are also developers. Our customers are the automation engineers. We had a customer on stage from a big US drama company that said, Since I have tools like SCA, I'm thinking of buying a dog. Because ultimately this job of him playing around with the laptop, then not having the right engineering software at hand, completely goes away. I think there's something that is really important for us We've been fed collectively to the most valuable programmers in a factory because we gave them tools that were ancient. We sent them around away from their families around the world for a very simple process. What we see now emerging is a new craft, which is the industrial physical AI engineer. This is, as Albert said, a person that understands the process deeply. Provides ideas what can be improved, and then has systems that ideally immediately do these iterations and then move on. And what you need is you need both. You need an instruction to really read and write back to actual machines. And I agree, you all that we can only see what the source code knows and what the telemetry gives us. If you combine this as a physical system on top, we have a complete picture. But I think Yeah, AI is fantastic because it ultimately elevates the computer programmers in the factory to what engineers at Amazon or Google always had. And I think the positive story is the value of those engineers will increase. And I'm sure also the salary, because they can do more that they generated value is more that I think is the positive story. These are extremely intelligent people. I think now is the time that they also earn what they deserved all along.

Michael Finocchiaro (00:17:57)
Good good point there too. so the the next question is more on where AI sits in your stack. I think you guys have already started talking about that. Like lar Albert, you're actually building more or less a large model for all these machines. And Josef, you're doing something similar to look at all the different controller systems. so would you say is there and it doesn't sound like there's much of a user experience level on either of these because they're much more closer to the hardware, but Where where are you seeing the most traction in terms of where you're adding AI to the stack and trying to improve the process?

Josef Waltl (00:18:34)
L let me first disagree with you.

Michael Finocchiaro (00:18:38)
Okay.

Josef Waltl (00:18:38)
Our software is all about user experience. Our software

Michael Finocchiaro (00:18:41)
Okay.

Josef Waltl (00:18:41)
is all about doing something in four minutes that needed forty-five minutes in the past. and the applications for frigidity AI or for real AI in the shop floor are sometimes very modest. I give you a concrete example. We have a customer in Germany, they have machines for packaging, and they need to mo And all the source code is in Italian. So they came to us and said, Can you help us? Because in the middle of the night they invisible it's up to understand Italian variable means. So what we used LLMs for first is explain it. And explaining actually complicated code is tougher but then generating good new clean code. That's a specific use case. And this is actually helping him. Then we built on top of it. Now we have code generators, now we have what we call agents. You can give people agents tasked to do certain repetitive things on your control code. But it really developed from a very simple use case. And I think there's also one thing that I want to mention is the cost. I think this is something where we would see much more specific modes. We do not need an LLM that can recite all German philosophers. But we need a model that is really, really good as a control pro code programmer for Rockwell, Siemens, Beckhoff, Mitsubishi. And this

Michael Finocchiaro (00:20:11)
Mm-hmm.

Josef Waltl (00:20:11)
is where where the world will go, because then it will also be deterministic. And it will be deterministic in cost. And that's important. You need to understand what is the X on my input factor as you have in any factory and the same is true for every tool which is ultimate an input factor to a manufacturing system.

Michael Finocchiaro (00:20:35)
can I throw that one to you, Albert? Same question.

Albert Klein (00:20:37)
well yeah, I also have to disagree slightly that it's not in

Michael Finocchiaro (00:20:41)
Ha ha.

Albert Klein (00:20:41)
a user interface as a matter of fact. AI is absolutely everywhere. Yeah, it's in the in the basic technology stack, it's in the physics model that drive the insight and how the machine should act. It's in the in the simulation, the digital twin of the machine down to its functional simulation based on data. All of that is done with AI. Then the reprogramming is done with AI, obviously with LLMs, etc. Of course, not not these super Frontier models is i I agree is not needed. There are small models which are quite quite capable of understanding code, in particular PLC code, which is not very complicated. So the AI is in the user interface. The software continuously has to adapt to the its use cases. So in our system, people are able to specify use cases and they're automatically deployed in in in version changes. No programmer can

Michael Finocchiaro (00:21:33)
In our system people are able to specify cases and they're automatically deployed and and that you change no code.

Albert Klein (00:21:42)
do this today fast enough and cost-effective enough. So And the AI is there, always on the side, a little chat saying, I have a question, I want to know something. You have need to fill in a form for a for something, for for a BMW manufacturing requirement, or or safety documentation, or quality documentation, you scan the PDF into the AI, translates it into instructions to the users

Michael Finocchiaro (00:21:59)
documentation or quality documentation to scan the PLF into the AR and loads it into facts.

Albert Klein (00:22:06)
of the machine, how to comply with BMW requirements or anybody or any OEM requirement or thirteen four eight five medical directive requirements all of this stuff. AI, AI, AI will be absolutely everywhere. And it's it's important to to to to know and understand where it can be used, how it can be used. we have a deterministic world, as Josef said in in manufacturing, five is five and not sometimes four point nine and sometimes five point one. Just not possible. so we need different AI systems, L L Ms,

Michael Finocchiaro (00:22:22)
it's it's important to to to to know and understand where it can be used out and because we have a deterministic world as Josef said in manufacturing size is five and not sometimes four point nine times five point one just not possible. And so we need different AI systems and

Albert Klein (00:22:46)
foundation models, time time sequence foundation models, multivariate, single variate models. We have a a very, very complex AI stack in manufacturing. And I I venture to say that

Michael Finocchiaro (00:22:49)
foundation models and popular simulator modules we have a very, very complex AI that we manufacturing. And I I venture to say that

Albert Klein (00:23:02)
yes, we will sometimes need new hardware stacks as well. The old PLCs simply have some restrictions

Michael Finocchiaro (00:23:03)
Yes, we will sometimes need new hardware systems as well because DLCs simply have some restrictions. the metric management and all of this is just not good enough if you go to much more discrete sensors, much more continuous and big complex mix of analog signals. All of this stuff will require zero to test people who really understand what's going on. But the implementation of

Albert Klein (00:23:11)
the memory management and all of this is just not good enough if you go to much more discrete sensors, much more continuous and a big complex signal of a mix of digital and an analog signals. All of this stuff will require, as Josef says, people who really understand what's going on. But the implementation of such a system

Michael Finocchiaro (00:23:33)
system usually automated. We always add sensors to systems and these when we look at

Albert Klein (00:23:33)
can be hugely more automated. We always add sensors to a system when we when we look at a machine.

Michael Finocchiaro (00:23:40)
the machine system just not enough sensors to understand what's going on in the material. That requires new UCs, new additional IO things and sometimes faster systems have other seasons, how I

Albert Klein (00:23:40)
Because there's just not enough sensors to understand what's going on in the material. That requires new PLCs, new additional IO thing. Sometimes bus systems have latency is not in in sync. So I think we will be replacing hardware a lot. That's my prediction. We did this when CNC came. We had the old machines, they had manual actuators, they were replaced with with NC and CNC systems massive. Over decades, most of Siemens 300 and 500 systems were used in existing machines, not in new machines. And because the the value of the hardware

Michael Finocchiaro (00:24:10)
Over decades most of CNC hundred and five hundred systems were using existing machines not in the machine. And because

Josef Waltl (00:24:18)
Right.

Albert Klein (00:24:19)
is great. I mean, most of the hardware is fantastic. The the mechanical side, it's just the control system that might not keep up with the AI. Is the cost going to be worth it? Why can't we we can definitely program IPCs with with C using Cloud.

Michael Finocchiaro (00:24:26)
the control system that might not keep up with the AI. Is the cost going to be worth it? Why not we can definitely program I P Cs with with C plus plus using Clore. There's no problem there. so can we find the P

Albert Klein (00:24:39)
There's no problem there. so can we find the people who know how to do this and how to implement such a thing? That's the challenge. And this is where Josef and his companies play a huge

Michael Finocchiaro (00:24:46)
think that's the challenge. And this is where this gap. Maybe a

Josef Waltl (00:24:52)
So maybe a short addition to this, and I hope it's okay. I agree there will be changes once people see the value. And what we see is people first need to see the value on existing equipment. We also do analytics of PLC code for simple errors. Give you an example. We had a customer with two packaging machines and two robots at the shrink wrapping. And When there was a problem, both machines stopped. And they couldn't find it for one and a half minutes. We upload it into the backup of the C code. We by the way also decompose it and make it also better to be processed by AI. We found in a couple of minutes that there was a copy and paste error. The programmer just changed seven out of eight variables on the second robot. So there was reflective very simple thing. So I think the precondition

Michael Finocchiaro (00:25:47)
So he could republish.

Josef Waltl (00:25:52)
To making this investment step to say, hey, we need more compute power is first of all showing it on existing equipment, and then I absolutely agree. We are really here to support customers, to have a meaningful PL impact in their operations, on existing

Michael Finocchiaro (00:26:06)
needed for key and out of the head.

Josef Waltl (00:26:10)
equipment, and guiding them to make decisions as Albert laid out.

Michael Finocchiaro (00:26:17)
And thanks for that example. That's really helpful. so what do you guys think it's when do you do you think we're gonna have an open EA moment in manufacturing? Will there be a moment where suddenly everything changes? It sounds like it'd be kind of hard to do because of the the variety of of machines and all the techniques and just the complexity of of manufacturing in general. And then is that question apply to discrete or to continuous or to batch or all three? I mean it's how do you think that w What do you guys think is gonna be the next step on on the AI journey for for manufacturers?

Albert Klein (00:26:54)
Well, when you say manufacturers, I think that there are different types of manufacturers involved in the manufacturing world. There are the makers of machines, that's one one side of it. And then there are the

Michael Finocchiaro (00:27:05)
Yeah, industrial equipment.

Albert Klein (00:27:06)
users of machines who make a product. And there are all sorts of other intermediaries and users there. There are this SCADA systems guys and then there are all sorts of other and there is these these the these general system environment people who keep factories going, Then there is the MES and ERP world that has a huge impact on what's happening in in factories. So all of these are going to have their different open AI moment, but they are going to have it or they have already had it. And I just some haven't realized what what what's going on. So the the there is there is no doubt that that we the the The open AI moment will look a little bit different. I mean, we saw we saw what happened with OMI because it touched something that was so close to us, language. Now, buttons on a machine are not quite the same thing. But for the people who use the machine, yes. For those individuals who are in factories and have to fight with machines every day and with manufacturing problems, quality problems, delays, scheduling, material problems, etc., these people, every one of them. Will have their open AI moment over the next two, three, four, five years. They will see something which they've always been dreaming of. And they will a solution, they will see something which solves their daily nightmare. That's where we are addressing the manufacturing problems with with a with the open AI moment in our dreams and our nightmares. And it's coming quicker than what you think.

Michael Finocchiaro (00:28:24)
for five years. They will see something which they've always been dreaming of. And with a solution, they will see something that solves their daily night. That's where we are addressing the kind of action problems with with with the open AI moment. Our dreams and our nightmare. And it's coming quicker than quick. Josef

Josef Waltl (00:28:49)
Let me add a a contrarian perspective to it. Not in the sense of this will not happen in existing factories, but there is a second thing happening. We have now companies, especially in the US, that take manufacturing back as their core competency. What I mean with this is for decades manufacturing has been treated as a financial problem. I have input factors, I buy machines, I buy a factory and then I've machine builders, I pay them and then I basically check what the output is. And then the mask companies cave. And then Hadrian K and under K. And they build their own factories, they build their own software stack on top of it. They buy shipyards and say we make them software to find. So this change should also come by these companies just being much more effective. And then winning market share and I think the indeed it's a good example. Completely contrarian view on how to build cars, what automation is, what the what the amount of automation is in manufacturing. So I think we will have these two forces. And I'm very happy to see, honestly, also, especially in the US, there's more and more companies also coming out of those ex-engineers from these companies that say we need to master manufacturing, we need to increase the innovation speed in manufacturing. And then automatically

Michael Finocchiaro (00:30:17)
And then

Josef Waltl (00:30:18)
they do stuff differently. But most important, they invest in talent and tools in software in manufacturing. I think that's the big change. And that will happen

Michael Finocchiaro (00:30:22)
the invest balance controls in water in manufacturing. That's a big change.

Josef Waltl (00:30:32)
outside of existing large scale manufacturers, will always have a legacy, which is good, but sometimes it's also a burden.

Michael Finocchiaro (00:30:42)
Another question I'd like to inject at this point is s for the younger people in the audience, because I do tend to have some people earlier in their career that are probably very worried about being AI'd out of a job before they even start their real career. What kinds of skills or what kinds of things you think well, you guys are both hiring people, I I assume because you're both growing. So what are those skills, those those intangibles that you're looking for in those candidates that the AI will never be able to replace?

Albert Klein (00:31:15)
Well, if if I get starts I think the most important skill that you have nowadays in a job is how to use AI. So whatever you've learned, wherever you are in your life, young, old, whether you've gone from

Josef Waltl (00:31:35)
Right.

Albert Klein (00:31:35)
university, come from a university, you're looking for a job, or you're thinking about going to university, you need to start using AI all the time to do the things you want to do. And there are two things you need to consider there is you have to look far beyond the scope of a problem to solve it today. And that is you need to look at it in a broad sense. So you need to understand what's what's the environment that the problem exists in. And then you have to look at it in a deep end and saying, How do I really solve this? So it's it's been astonishing for me, and this is where the LLMs really have Have done a major difference, is they are like an incredible knowledge search engine. And if you don't use them, you can't compete with somebody who does. It's simply it's it's as simple as that. You don't understand what's the context of a problem that you need to solve, and you don't understand how deep you must go to solve it. When do you have to stop understanding it? And it all happens in a matter of, let's say. minutes or hours. So that's that's if that's what we're looking for. We're looking at people who can really solve a problem and as a matter of fact, recognize what kind of what class of problem is it and how do I solve all of the problems within that class.

Michael Finocchiaro (00:33:07)
Good answer. I like that one. How about you, Josef?

Josef Waltl (00:33:11)
I think very similar. Let me let me put it down. I think people that are builders people that are curious on problems are have the best time in your and their life in front of them. People that are builders go down to to details and AI by the way will help you go down to the details. AI is not a competition. I think AI is your best teacher. Meaning it's like For an engineer this will be like flying. I have an idea. I do a first iteration. I test it out. I understand there's a problem. I go down to the deep end of the problem. I go up again. I go to the next iteration of the problem. So I think our world is getting faster and even more complex. And there is an I think unlimited need for people that are really interested to build. So I would flip it around. Instead of going back and saying, Hey, I have no chance, why should I do it? Embrace it. You will have much better tools than anyone else in mankind and have fun

Michael Finocchiaro (00:34:19)
Mm.

Josef Waltl (00:34:19)
with it. NAI will help you learn and solve problems and go to the next element. And I don't see this going away.

Michael Finocchiaro (00:34:29)
I I love that. yeah, it's funny being

Albert Klein (00:34:32)
And the other thing the other thing that doesn't go away. It will be a long time before an AI really solves a problem on its own. You know, we are we are still steeped in the physical world and in there's no place like in a factory where you see this. You try and solve a problem only with AI without understanding what's going on in your factory. You'll fail. And so what you and the the thing you need to do is you need to have some some kind of understanding what's going on in the real world. And the other thing you need to do is you need to cooperate with other humans to solve a problem. And AI can't communicate very well. You have to tell it, you know, you we all know what AI, how it writes and how silly its powerpoints are, etc. So if you want to communic cooperate, you need to either lead or be able to follow. Those two things are human. And that doesn't go away with w for a long time. Eventually it might change the game a little bit. But that's what I believe is is confronting young and old as we speak.

Josef Waltl (00:35:39)
let me put it differently. My engineers, including myself, have hard toed boots under our desk because we go in night shifts into bottling plants and talk with the engineers. And unless AI can do this, I'm not afraid that there will be no place for guys like us in the industry.

Michael Finocchiaro (00:36:01)
let's just to the last w thing I like to talk about on these calls is digital transformation. And these are it's probably a great example because you guys are both in places where that's where the the digital maturity of the companies we're working with is probably relatively low. I usually think of scale of one to five, like one is the digital maturity is they're still using Excel and and email to to to collaborate and to do engineering change, whereas a five would be fully agentic, autonomous, adaptive digital twins, which nobody is at five, right? so first of all, how do you get to look at your customers? You guys are both looking at a manufacturing customers and and aging factories. Do you see them already reaching before they start using fluxin systems or or SDA, do you start do you see them already, where do you see them in terms of maturity today?

Albert Klein (00:36:57)
Else that place

Josef Waltl (00:37:00)
I think first of all, we start with the existing machinery from the hardware on top of it. So it's a bit unfair to say on majority. But what I see is we we see one, two or four. We also have extremely innovative customers that immediately take something up, have no challenge to to to to test out new technology in small small elements of their manufacturing. But the bulk definitely has very simple problems. But I think there is also one thing that we need to understand. In manufacturing, all systems are built on top of each other. If you're not solving the simple problems, like I said, understanding the source code of your machine. Getting it, yeah, it's as simple as that. Getting a global repository of your source code of your machines, of all changes, of all implicit learning that comes out of it, you never get there. So I don't think it's bad. that a lot of customers still have basic problems. I think this is a a joint need that we as technologists need to do and continue to support. And this is also where I see where some IT companies always struggle. If you're an IT company, it is very hard to understand what it really takes to control all robots in a factory. It's like looking from Manhattan from the top and say, it's a bunch of squares. You don't see that some buildings have like 200 stories.

Michael Finocchiaro (00:38:28)
Right. You're you're missing some dimensions to the problem, exactly. How

Josef Waltl (00:38:32)
Yeah.

Michael Finocchiaro (00:38:33)
about you, Albert?

Albert Klein (00:38:36)
Well, when I was in the tender age of my teeny years, we we suddenly were confronted with a personal computer. So and since then I think we've go undergone a continuous digital transformation and this is just another phase in it. I think that I would I would like to call it differently. I think it's the it's really is the intelligence transformation. How much is a company able and willing to combine the intelligence of its human side with the intelligence of the machine side? And where that intelligence on the machine side comes from is there many different things. It's LLMs, it's it depends on the function. And and and computers are getting better and also conventional software is getting immensely better, it incorporates AI, et cetera.

Michael Finocchiaro (00:39:26)
getting better and and also conventional software conventional and corporate AI et cetera. When you want to trans what when you want to do what you call digital transformation and happen you need to still comp on popular that's in your people, not in your company. So that is what the transformation is really about. And I think if you give a part from one to five for that

Albert Klein (00:39:33)
When you wanna transform when you wanna do what you call digital transformation as a company, you need to instill a sense of of openness and readiness for that in your people, not in your software. So that is what the transformation is really about. And I think it if you give a mark from one to five for that, then it's

Michael Finocchiaro (00:39:56)
the average and the average and department.

Albert Klein (00:39:56)
it's the average is more towards the f one than the five. So

Michael Finocchiaro (00:40:02)
So and and therefore I think I think it's important to look at what are the people going to do with

Albert Klein (00:40:02)
And and therefore I think I think it's it's important to look at what are the people going to do with AI

Michael Finocchiaro (00:40:09)
AI.

Albert Klein (00:40:09)
than what is AI capable of doing. And here I have some warnings. I think I think people have have adequate and and valid reservations about using AI in their companies in general and in manufacturing in s especially.

Michael Finocchiaro (00:40:10)
and what is AI capable of doing. Here I have some warnings. I think I think people have have adequate and and valid reservations of using AI companies don't and in manufacturing

Albert Klein (00:40:30)
being out of control in using data. If this is your manufacturing data, you have valid concerns about this data being processed and and used by people who use AI systems to do so. You need to be assured that this data doesn't get I mean, it's not even a matter of of it getting into other hands or being lost or so. It's a matter of it being used by by by by your contractor.

Michael Finocchiaro (00:40:43)
AI systems. I mean, it's not even a matter of of it getting into other hands or being lost or so. I don't know if you're doomed by it by by your contractor,

Albert Klein (00:40:59)
By your trusted AI provider, the way you agreed with him.

Michael Finocchiaro (00:41:00)
by your trusted AI provider, the way you would read something.

Albert Klein (00:41:05)
And this is something which we are going to face more and more discussions about this. And unless we resolve this, what I call trust issue, we are not going to see widespread adoption of AI in manufacturing because it's too important. It's the core IP of somebody who is in manufacturing is inside that machine and inside the heads of the people. And if you ask him to give him your data to improve your factory, he's gonna say, I need to be sure that you just do that and you don't do anything that I didn't agree. So I think that we are going to see debates about the AI transformation, the intelligence transformation. And those are healthy debates and we need to answer

Michael Finocchiaro (00:41:48)
I'll get together.

Albert Klein (00:41:49)
them. Just saying and saying, don't don't worry, trust us, is is not it. That has never happened.

Michael Finocchiaro (00:41:51)
Saying you say, I don't know why you trust us. Please do we didn't trust any technology before.

Albert Klein (00:41:57)
We didn't trust any technology before we really could trust it. So and and and that that was always the case. I'm not critical about it. I'm a a fantastic supporter of AI and and the intelligence transformation. I think it's going to be of immense societal and and

Michael Finocchiaro (00:42:09)
AI and the intelligence transformation I think is going to be public defense societal.

Albert Klein (00:42:15)
and and and economic benefit to all of us, not only few rich people. But I think we need to be very careful not to neglect this side of the intelligence

Michael Finocchiaro (00:42:20)
I think we need to be very careful not to neglect this side of the

Albert Klein (00:42:25)
transformation.

Michael Finocchiaro (00:42:28)
Yeah, and one of the panels I was in in and last week and the week before, I was mentioning that I think the the the moat really when everybody's using AI and and especially if everybody's using foundation, you know, the the frontier models, the real competitive mode is gonna be trust, right? Who do you actually trust with your data as opposed to what cool things can they do? what about you? So the the other side of that question is my The premise of the reason why I built Threadmoat in the first place is I feel like there's a lot more innovation happening outside the the big incumbents, the the the major companies that are providing in this space. And that's where companies like software defined automation or flexian can step in and and up the game. Have in your experience when the customers see what you guys can do and the transformations you can make, is there sort of an aha moment? Like, my God, if I if I took digital govern data governance seriously and I took di you know transformation seriously, I could get so much more out of all of the my equipment and out of my factory and out of my people because I wasn't wasting time on the stupid mundane tasks that I would have to do if I was just using traditional enterprise automation software. So do you guys any have any comments on that? Like how th those kind of transformations and that kind of aha moment that they might have your customer.

Josef Waltl (00:43:52)
Yeah,

Albert Klein (00:43:52)
Well, that that's a open AI moment that you were talking about. You suddenly you put some sensors on a on a machine and it and you have a mobile phone in your hand, or you put in an AR glass and it tells you what to do, and suddenly your machine runs more than twice the speed that it did before. You had no idea that it could do that. And well that that's definitely what we see happening. And from then on they're sold. They don't go back to running the machine half the speed. Just because it was like that before.

Michael Finocchiaro (00:44:23)
Mm.

Albert Klein (00:44:24)
That's unimaginable.

Josef Waltl (00:44:26)
Yeah. So in our software we also face the challenge of securing core IP. We are not using it. We give people a car and the key. Meaning they have a core IP via the tool and they down they upload basically into their tenant, but it's com encrypted, you all have all types of security certifications. That's immensely important because other than that they will not use it. And I think there is a difference in especially how IT and some companies see it, but we can see in manufacturing that doesn't work. If you make ball bearings for forty years, the core IP is in the grinding software of those ball bearings. And it's it's the almost a whole company value. But once you have the tools, and with our customers, we went through IT certification. Some people want to do a specific tests with us and We I think we have quite impressive customers. Then they see their value. And then they see their value in addition to what Albert said, not only on machines, but across fleets. We are cloud first. That means it doesn't matter if you have one factory or fifty factories, if you have one machine or two thousand machines. And then you see things that go beyond the single machine. You see that problems occur in all of your whatever eighty bottling lines in parallel. And it's always the same problem. And then you can see it if some changes happen, it's the same problem and there is a solution that already developed three times over. But agree, first of all, we really need to build the trust in our user base. We have no observation on AI because the problems are so big. We need to help the users, the automation engineers, to get the allowance Value is there and once the security and the trust is there, we see exponential adoption.

Michael Finocchiaro (00:46:33)
thank you so much for the Albert, did you have something to add? You said you you looked like you're about to say something.

Albert Klein (00:46:38)
You were your your question was about where is the innovation happening. I think it's happening everywhere. However, I can say one thing. The speed is is a key element. and and and

Josef Waltl (00:46:50)
Yeah.

Albert Klein (00:46:51)
big companies they there well, you know, a lot of people are talking about bureaucracies now, et cetera. And you feel this a lot when you use AI, is that and and somebody once said big companies can only survive if they become small companies. And I think there's some truth there. The the the innovation and the creativity is a is is is a function of delegating responsibility and tasks into where they can be solved. And I think companies need to get better at that because and it's very important they get because we it rely on their technological prowess and their capital and and and the plethora of populations of of of control systems which we have in factories. And and We rely also on on solution for some fundamental problems. Josef, you know this one. I mean that machines are have never been designed to be safe computer systems that are hooked up with the world. They are completely open in man in some cases, in particular older control systems. So it's going to be really difficult to find the technology to connect machines to intelligent systems, and big companies will play a major role in this. So at the moment I'm hoping that that that they are that are are they are understanding and and and providing the technology which we all need to prosper in in manufacturing.

Michael Finocchiaro (00:48:19)
well thank you guys. That was really fascinating and and enlightening as well. so I where are you where can people come and meet you guys? Are you guys doing any any trade shows between now and the end of the year or beginning of next year?

Albert Klein (00:48:34)
Well, we're at the Asia Berlin Summit next time. And then in the new year. We don't know. We are now in Doha in the in the creative destruction lab, but nobody can go there. So

Michael Finocchiaro (00:48:45)
Yeah.

Albert Klein (00:48:47)
we're looking forward to talking a little bit about about how this will also lead to an internationalization of manufacturing know how. Because it definitely leads to what Josef said, to an onshoring and an insourcing of manufacturing. But that effect of upgrading your manufacturing and prowess is taking place everywhere in the world. And we have not made enough products yet in the world despite people saying we need to s scale down and slow down and do whatever. I don't believe that. I want more nice cars and better planes and all sorts of things.

Michael Finocchiaro (00:49:23)
Well you you would you would hope too that here in Europe that there would be more of this passion for reindustrialization because that's obviously an America a massive thing. And I mean it's I'd spent the last month there and it's just really the obsession of all the manufacturing companies is reindustrialize and I I don't think Europe has caught that bug yet. I think we s we still have a lot of catching up to do and you know, we spent decades

Josef Waltl (00:49:47)
Yeah.

Michael Finocchiaro (00:49:47)
getting rid of it and now we're really and tr we're really behind, you know.

Albert Klein (00:49:51)
we haven't gone out

Josef Waltl (00:49:51)
actually we

Albert Klein (00:49:53)
we have not de industrialized. There's a lot about pre preventing that from happening and but we we in Europe we didn't de industrialize.

Michael Finocchiaro (00:50:02)
Well, UK and

Josef Waltl (00:50:03)
Actually there is there's a

Michael Finocchiaro (00:50:04)
France did though, right? I mean a lot of the manufacturing is done outside.

Albert Klein (00:50:07)
The day.

Michael Finocchiaro (00:50:09)
Well the UK nothing is manufactured there anymore. Everything is done outside. I mean almost everything's contract manufactur well there's there's quite a lot of contract manufactured there's still like mom and pop shops, but the bigger pieces of Airbus or whatever, a lot of those were done outside.

Albert Klein (00:50:21)
Really.

Josef Waltl (00:50:21)
Let me again disagree here. also in Germany, we had three weeks ago 130 people from all types of manufacturing, pharma and automotive and parts manufacturing and machine builders in the BMW world in Munich. And people were on stage reporting how they use technology to re-industrialize. It's not at the scale yet, I agree, but I think the interest is there. Also, in terms of where you meet us, just ping us on our website. We give you a demo tenant and you can play around on the same day for fourteen days. This is you do not need to wait until we are on a trade show. So that's the

Michael Finocchiaro (00:51:07)
Mm-hmm.

Josef Waltl (00:51:07)
message that I would give because it's

Albert Klein (00:51:08)
just so.

Josef Waltl (00:51:09)
completely it's completely s self explanation explanatory. And people get their machines and first demo environments, NPLCs online the first day. There's no local hardware required at all. Of course we are also at Trade Jos, we are at SPS Fair in Nuremberg and we'll be on Prove It in February. I think Fina where we also will meet. Yeah.

Michael Finocchiaro (00:51:32)
Yeah.

Josef Waltl (00:51:34)
Fantastic. We are also part of NIH, the Neo Industrials Alliance in America because we're American and European company. so if you're interested in this type of innovation, don't wait for Trade show, reach out to us right now or ping me on LinkedIn.

Michael Finocchiaro (00:51:54)
I just wanna so say thank you to Ares who's sponsoring this podcast. thank you once again to Albert and to Josef. I looking forward to seeing you in Dallas, of course, Josef. And Albert, I hope that we'll we'll catch up sometime near Munich or or wherever.

Albert Klein (00:52:09)
Yeah.

Michael Finocchiaro (00:52:11)
I hope so i I'll be back in I think two weeks. I should have Neuromil and and pla and paperless parts. And well just me signing off. Thanks everybody, and we'll see you the next podcast.

Albert Klein (00:52:23)
Thank you, thank you for having

Josef Waltl (00:52:24)
Thank you.

Albert Klein (00:52:24)
us. Bye.