Industrial AI is Already on the Shop Floor – Transcript
In this episode of the Future Ready Podcast from Siemens, my colleague Conor Peick sat down with Rainer Brehm of Siemens and David Humphrey of ARC Advisory Group to talk about the future of industrial operations and factory automation.
Nick Finberg
00:10
Hello and welcome to the Future Ready Podcast from Siemens. This year, one of our moderators made the trip over to Germany to sit down with a bunch of great experts within and beyond Siemens at our booth at Hannover Messe 2026. One of these conversations was with our host, Rainer Brehm, the COO of the automation business and CTO of Siemens Digital Industries, as well as with David Humphrey, the Director of Research at ARC Advisory Group. The show floor was full of conversations around industrial AI with co-pilots and intelligent agents, but we wanted to ground these ideas more firmly in what is happening in industry today and what could happen in the next few years. So, let’s jump right into it and give the reins to my colleague Connor Peick to talk with Rainer and David as they explore what happens on the shop floor in the first part of their discussion.
Conor Peick
00:56
So, gentlemen, thank you so much for joining me today. Really looking forward to our discussion. To get us started, I’d like to hear from you both. Maybe David, you can start. If you compare this year’s Hannover Messe to last year. What feels different to you and where have we really taken a step forward with industrial AI and what has impressed you most walking the floor?
David Humphrey
01:13
This year I would say AI is everywhere, but I think we said that last year, and I think we said that the year before. the difference for me is I’m starting to see the stratification of AI functionality in starting to be implemented in its useful areas. if companies were experimenting just even up to last year there are more and more products that are coming out now that have AI enhanced. It’s a built-in tool and companies are starting to use that and they’re starting to achieve real production, productivity gains very quickly That’s the difference to me for to this year.
Rainer Brehm
01:47
Yeah, and I absolutely agree on what add on. like last year we won the Hermes Award for the industrial co-pilot This year we bring it out as an eigen engineering agent. Yeah. So, we basically we are coming from you know having pilot customers not really going into scale. I think that’s a topic. And you see also movement from co-pilots which are more assisting into age and new technologies. And I think we come to that later. Yeah.
Conor Peick
02:15
So then David, to ground this in in data, I would love to know what you’re seeing in terms of adoption rates of industrial AI across regions or different industries and You know, are we already at scale with this or are companies still kind of in the early stages of adopting the these technologies?
David Humphrey
02:31
I would apply the eighty twenty rule here. I would say about eighty percent of companies, according to a multitude of surveys, are dabbling in in AI. Everybody’s trying it, everybody has done something with it. So, they have an initiative, they have a person responsible for it. About 20% are actually implementing tools, either their own tools that they developed themselves or software that has AI. in enhancing it to achieve this is my test to actually modify or improved production outcomes. And that eighty twenty rule like it always does is going to flip. So, I would say in about five years it’ll be maybe twenty percent stragglers that are still dabbling but haven’t really implemented things on a on a plant wide or a company wide basis and the vast majority of users will be enjoying the fruits of AI in implementation.
Rainer Brehm
03:25
I think also this twenty percent rule currently applies when you talk about AI its capability and we have some examples like where customers really use it. And that’s not necessarily the generic AI or generative AI. It’s kind of really the kind of AI which is on the market since a long time with machine learning. But we see a lot of mean a lot of vision applications today are AI driven. Yeah. And that is basically a maturity state. And we have here companies like [indecipherable] which are using our classical PLC, but they do an inspection, high-speed inspection on a on inline quality control using AI. I mean that one example, yeah? Or we have here the booth we have Pringles. And Pringles are using a combination of digital twinned AI-powered predictive maintenance with our sensei portfolio and our industrial AI suite. And they get also 10% higher capacity without adding new light. So, I could go on and on. So, there are companies which are using it already, but not kind of on the eighty percent side, but more on the twenty percent side.
Conor Peick
04:31
Excellent. And you mentioned there how there are companies already using this technology to drive value. I guess Where do you see maybe the biggest potential for them to take this further and to grow with that that adopt?
Rainer Brehm
04:41
We had really impressive numbers on this eigen engineering agent. Yeah. and we did that not on a custom we did a kind of scientific investigation about that. A university was doing that, and they put I think I don’t know how many hundred people which did a certain job alone and the under hundred people which did it in combination with the eigen engineering agent. and they were then really impressive in results. So, it was really a scientific study, and they were coming out that you know the one using the eigen agent were two to five times faster in doing it They had an 80% higher overall solution quality and so in total up to 50% efficiency gains in the automation engineering task. And that is a really great number. So that really shows a fast return of invest. And the good thing is a product. So, you basically you use it, you buy it. it’s also not very expensive. You get into a subscription, And I think that will now really drive adoption on those because the difference on the 20% we just talked about, industrial eye, you need to train models, that’s an effort. It takes time to get the data. You need to have maybe some people which have a knowledge about AI. It’s not kind of you simply use it. Now bringing it really into the world. where our automation engineers live today and it’s simply an add-on. You don’t need to install new environment. You use it in our case it’s a tier portal and you add this on and it’s generating code without you need to be an expert I think this will drive adoption and even if you have that great return of invest.
Conor Peick
06:17
So that’s really, I mean that’s really fascinating, Rainer, what you’re just describing there with this Eigen Engineering Agent. So, you know, we hear that AI is really great at generating insights, but of course, as we sort of mentioned, value is only created when those insights are then executed on the shop floor. So where does Siemens close this loop between the digital and the physical world and is software-defined automation a key enabler in that closing that loop?
Rainer Brehm
06:39
Yeah, I think it’s even improving it. I think it’s not Well we talk about combining the real and digital world since m many years. so, if I would say now hey software defined net automation is the enabler, it it’s wrong. it you can do it today. I think it simply makes it easier. Yeah, because you if you if you’re combining the real and digital world and you come to the digital world and you also have it more software-defined in the real world, I think the connection link is simply easier. Yeah, so that’s the reason I would say yes, it definitely helps. And it basically goes into a direction where in the future I think much more functionalities are decomposite. So basically, of different elements which you want to combine. And that’s much easier on a software-defined world because its simply adding functionality compared if its hardware defined, you need to always add
David Humphrey
07:27
Hardware pieces to make something happen.
Conor Peick
07:37
And does it make it a little bit more agile as well, making it software-defined, it’s easier to update with changes or new functions?
Rainer Brehm
07:46
I don’t know how far our customers on that one. And I still believe in in a lot of customers. You still think about you know never change a running system. It runs. You don’t switch it even off. And if it’s energy consumption, okay. But the risk of switching it off and doesn’t start anymore. So, keep it running. But never change a running system but as you know in the software world y you constantly update and you know, you constantly fix and you know a software product is never really finished. You basically always kind of continue developing and even do a bug fixing because something is wrong or you have a vulnerability on a cybersecurity side. So how we get this this this mindset of constantly updating and constantly no improving it, like Lee Natchild, you won’t constantly get an update into the OT world. I think that that’s something where uh
David Humphrey
08:33
Reiner, that’s a mindset to get people to use to this you have to change the mindset. And y you see this in the software development tools. They talk about using industrial DevOps now, which is a set of best practices to ensure quality and rapid deployment of newly developed software. But if you’re developing software all the time, it’s got it has to be implemented, it has to update. one of the one of the real values I don’t want to get out too far off track right here, but one of the real values for SDA in the long run, in my opinion, will be in the process industries where asset lifetimes are at least twice as long, maybe three times as long as in the discrete world. And with an asset of a piece of process equipment that costs a lot of money if I can continuously update its functionality over time. and eventually migrate to new hardware, the basis maybe 15 or 20 years down the road without disturbing the software at the top, then I get a lot more value out of that asset over the 30 years or more that it’s actually in use.
Rainer Brehm
09:32
Yeah, I and agree and I think this is absolute value for the process world. I would say in a in a more discrete world it’s a topic of flexibility and adaption of your production lines, where SDA will play a major role because you’re not producing the one piece for five years or ten years like in the past. I think you need to be much more flexible of adapting depending on market demands or maybe some upgrades, whatever, on your product and you need to adapt your production, and you cannot say now I rip it out and do it new. How can you get a more flexible adaptive production? And for that you need to constantly also change and update your system and how you do it in in the right way. And I think a lot of, so I like your topic of, you know, industrial DevOps, yeah. And I think it’s very important if you do that because you need to have the confidence. and to get the confidence I think that they talk around testing, about digital twin, about virtual testing. Maybe in the future do a virtual testing not offline but using a virtual testing with the real data from the process. in parallel to the reprocess and then switch over if you’re really confident. I think that will be something we need to work on how we really make that happen that you have the same confidence like you have on the on the classical world that you are doing now this update.
Conor Peick
10:49
Thinking about the industrial stack, all the technologies that that are a part of that, does convergence of those technologies help you build that confidence that you were talking about? help change the mindset within the company?
Rainer Brehm
11:01
Yes, because those technology get more and more the new normal. I think if you if you would go five years, ten years back and you know and everybody had anywhere a different opinion about a digital twin or digital shadow I think it was called that time as well. A lot of topics which have been thought through or even though I don’t know twenty eleven the topic about industry four dot zero was born. it was more the theory. and I think now step by step these new technologies nearly are mature, come into play and they are the basis and the foundation to move in this direction. Yeah. So, I think technologies are there and maybe they even get now boosted because they’re needed to go in this direction and therefore yes, I think there’s a certain conversion.
Conor Peick
11:43
David, then from your perspective, as companies are scaling their AI initiatives and perhaps their SDA initiatives as well, how important is that execution? execution layer become in turning those insights into the reliable and repeatable industrial actions we’ve been talking about.
David Humphrey
11:58
The execution layer is where it happens. It’s where everything is brought together. the newly generated insights that are available from the A model are injected somewhere and have to be acted on or you don’t have to act. It’s it on a small scale; it’s up to you what you want to do. later on, you’re going to want to automate those the inputs with the outputs so you’re gonna want that that the insight to flow into an MES system or flow into maintenance management so that a work order is generated if an anomaly occurs, a person is sent out to take care of it. You have to close the loops. early in the stages your question was about scaling up. So, when I’m small, when I’m just getting used to it, I have to automate that that connection at the beginning. Otherwise, that gap is going to widen as I scale up and I’m going to have More insights, but I’m going to have the same number of actions because I’m going to have a disconnect somewhere there. So, I’m not going to be getting real value out of it One of the important things is that if I have these good closed loops, then my system becomes self-learning. Then the action that that that I took has consequences. did it really solve the problem in the machine three days later or do I have the same problem? Do we have to take a look at it again but if I if I close that the loops as I scale up then I have that feedback automatically coming and my system becomes an automatic learning system. It’s always growing in its knowledge, we’re always adding to the model
Nick Finberg
13:25
Thank you so much, Connor, for moderating this discussion and thanks to the audience. I think we’re gonna take a quick break here, but we’ll be back soon to hear from Reiner and David about how companies are starting to scale these kinds of solutions across their organizations. or how they’re turning that 20% adoption into 80%. In the meantime, we have many other amazing conversations coming out of Hanover on a variety of topics, and we hope to have you back soon. Make sure to subscribe to the Future Ready Podcast so you don’t miss new episodes as they come out.
