“We were talking about AI assistants a year ago. Now we are deploying advanced AI agents on the factory floor, and that shift is transforming how manufacturers approach automation, data, and solving some of their biggest challenges.” 

In this episode of the Manufacturing Matters podcast, Ozgur Tohumcu, General Manager for Automotive and Manufacturing at Amazon Web Services (AWS), joins TECH B2B Marketing’s Winn Hardin and Jimmy Carroll to discuss how AI (agentic AI in particular), cloud computing, and digital transformation solutions are reshaping manufacturing. Tohumcu discusses the rapid evolution from AI assistants to agents and how these technologies are helping manufacturers conquer legacy infrastructure challenges, address labor shortages, and rethink their operations overall. 

Additional topics include real-life examples of how AI agents have helped major corporations, strategies for building a strong data foundation, the role of generative AI in accelerating product design, and emerging trends like software-defined factories and humanoid robots on the factory floor. 

Jimmy Carroll: [00:00:02] Hello, everybody, and welcome to this episode of the “Manufacturing Matters” podcast, where we aim to discuss the trends and technologies reshaping manufacturing today. I have the great pleasure of once again being joined by my friend and colleague Winn Harden. Winn, .Always a pleasure.

 

Winn Hardin: [00:00:16] Good to see you, Jimmy.

 

Jimmy Carroll: [00:00:17] And we are joined by Ozgur Tohumcu, who is a general manager for automotive and manufacturing at Amazon Web Services. It is a tremendous pleasure to have you, Ozgur. I really appreciate having you on here and thanks for taking the time.

 

Ozgur Tohumcu: [00:00:29] Great to be here. Thanks for making the time. And nice to see you, Jimmy and Winn.

 

Jimmy Carroll: [00:00:33] Of course. Yeah. So let’s let’s jump right in there then. And I’d like to know from your perspective what’s most exciting for you all at AWS today?

 

Ozgur Tohumcu: [00:00:42] Quite a few things, but I think probably what I would say is in the company I’m at right now, I feel like we have a front seat row to everything that’s happening in the tech space across the manufacturing industry. And I mean, we’re going to talk about it throughout today, what’s happening with AI, how it’s impacting the industry as a whole, but ultimately, it’s such a fast-changing world. I’ve been in technology space for the last 25-plus years. I haven’t seen this much unprecedented change. We were talking about AI assistants a year ago. Now we’re already talking about agents. And I think being at a company like AWS, you really get a front seat row. The other thing is, of course, in that front row, we’re not just watching, but we’re also working with some really large key customers who are thinking through their business challenges, thinking through what’s next for them, talking about, like, thinking about speed, how they can accelerate their own transformation. What’s super exciting for me is  how you partner with them. Because if you look at the space, tech is such a big part of the company strategy. Now, it’s not just sitting in the IT department, it is part of the . . . It’s on the board agenda. It’s on the COO agenda. And we feel like being at AWS, we’re part of that, thinking and future planning. So that’s super exciting.

 

Winn Hardin: [00:02:11] You know, Amazon Web Services, which is the correct acronym, correct of AWS.

 

Ozgur Tohumcu: [00:02:15] Yes.

 

Winn Hardin: [00:02:15] Fantastic. And with your perspective, you know, in the center of automotive and manufacturing, I just wanted to share how excited I am about this particular episode because as you mentioned, I mean, we are going through this point of data transformation, right? And manufacturing has traditionally been very, very, I don’t want to say sparse isn’t the correct word, but protective of their data for lots of good reasons–liability, risk. And finally, the marketplace has realized the benefits that come from cloud services, right? So that’s very exciting. And then AWS being one of the largest data providers in the world, you guys are really at the center of how manufacturing is leveraging cloud services across the board, AI and many other applications that we’ll talk about today. So super stoked you can make it, Ozgur, thank you so much. So we’ve mentioned some of the major trends in AI and other locations. What are some of the major challenges that manufacturers, I mean, is it still about data acquisition for them, how to get the data into the cloud, or is it visualizing how to best use it? What are the big challenges they’re dealing with on a daily?

 

Ozgur Tohumcu: [00:03:18] I mean, there’s always a technology challenge, but if I can start maybe with the business challenges. First of all, and I think it’s going to lead to how data and AI is actually helping with that as well. I think one of the biggest challenges, especially in the US, is the labor shortage. I mean, you continuously have a shortage of qualified factory workers. I think right after Covid it was 1.4 million estimated shortage is estimated to be almost up to 2 million by 2030. There’s a labor shortage. And there’s I mean, people are retiring. I think one of the challenges is going to be, how do you actually retain that experience? How do you retain that knowledge? And we can discuss that throughout the day. The other big challenge is again around supply chain. We’re reminded every two years how vulnerable our supply chains are. It was about during Covid, we felt like we solved the whole optimization, linear programming, inventory planning, challenges of the world during Covid. We were reminded that was not the case. Then a lot of the automotive industry went through a big microchip shortage. Models were delayed. So, I think we were reminded once again how vulnerable it is. Now we see impact coming from tariffs, for example. We see again how our supply chains are super vulnerable. I think that continues to be an issue. Inflation, I think in the last few years, I think about 10, 20 years ago, like growing up, I actually lived in a world where inflation was not a concern.

 

Ozgur Tohumcu: [00:04:50] Just in the last few years. It is now a pretty big thing. I think it has a big impact on costs. I think those are like the macro-level challenges for the manufacturers. Then coming to your point, I think then it translates into a lot of technology challenges OT/IT integration is still a challenge, like how do you actually capture the data? To your point, I think manufacturers are suspicious about their data. They’re also super paranoid about security. And when you have so much machinery out in the field, sitting in so many different locations, how do you actually maintain the security of your data where a single machine could actually allow a point where people can hack into your whole network and data repository? So, there are a lot of data challenges, but I would say the major challenges have been mostly on the business side. However, maybe I’ll lead you to your upcoming questions. But however, we’re seeing a very fast adoption when it comes to some of the AI-related technologies in manufacturing because it’s actually creating a way to almost jump ahead and skip a few technology shifts. When it comes to the existing infrastructure, how they can leverage technology using the latest AI toolsets. And then we can talk about that later in the show.

 

Winn Hardin: [00:06:15] I know, can you give us a couple examples about how we’re able to leap, how AI is helping us leapfrog? And I also was wondering if, I mean, when we talk about data acquisition and existing infrastructure, there’s a lot of legacy networks out there, not just legacy equipment. You know, there’s been a lot of effort to improve the security of the individual at the edge of the network and everything. So, we’d love to hear about how AI is leapfrogging and does leveraging a powerful cloud partner help you to increase your security, especially when you’re still dealing with maybe very widespread, large, geographically spread legacy networks?

 

Ozgur Tohumcu: [00:06:50] Sure. Look, I mean, I think maybe I’ll give an example. One of the things that we’re seeing is, especially with AI agents becoming popular, we’re seeing a super fast adoption across the manufacturing industry. Why is that? I mean, ultimately, agents are the next layer of the next level of automation, and the whole concept of automation is not new to the manufacturing industry. As a matter of fact, when the assistants, the AI assistants came about, the conversations I had with many of the manufacturing customers we had was, okay we understand that the assistant is going to make a recommendation, but that’s not enough for us because we want full automation, we want execution. So, when we moved into not just the recommendation, but based on an algorithm and through learning how the agents are able to also execute on some of the actions, such as, I mean, that could be doing your inventory planning and not necessarily just giving you the plan but actually executing on doing the actual planning. When you think about, for example, use of vision-based systems on the factory floor, it’s not only showing you where the defects are, but it’s also then giving the data back into the manufacturing line to actually correct the process so the defects are being addressed. I think being able to build the execution part into the overall manufacturing process has been quite widely adopted across the industry. That’s like one example I was mentioning. When a lot of our other customers are looking to work with assistants, I see a lot of the manufacturing customers are already moving into agents because automation based on complex algorithms based on learning really matters to them.

 

Jimmy Carroll: [00:08:57] Yeah, there’s a lot of different ways that manufacturers are using AI today, and you’ve touched on a number of them, right? And agentic AI is obviously a really hot topic right now. But, you know, in the space of AI, generative AI is something that’s very popular beyond just the factory floor, even just for fun, right? But, you know, when it first came out, it was maybe almost dismissed by some as being something that would not be potentially all that useful, or at least it wasn’t there yet. But in the, in the last however long, six months, year, whatever, it really seems like people are taking on generative AI and finding value. I know I have personally for my job. What are your thoughts there? What’s your take on generative AI, and how it can add value or how you’ve seen it add value?

 

Ozgur Tohumcu: [00:09:45] I think in two ways, really. One is maybe I’ll start withthe people who are actually using it, and what that means for the overall enterprise, for the organization. Like, I mean, AI and AI-related technologies have been around for a number of years. I think what made generative AI become like so hot and like everyone using it, all of a sudden, even my mom, is like, since in the last 2 or 3 years, is the whole interface between the human and the machine became conversational. All of a sudden, we felt like we can talk as if we’re talking to a human, like we’re doing right now, and you’re actually talking to a machine. And that barrier, which requires a bunch of software programmers, IT engineers, was dissolving. So, I think one of the biggest benefits of generative AI is actually giving the power of creativity in the hands of people. And people with very little AI skill sets can actually harness the technology and can create things. We see different implementations across different enterprises. Some very much believe in the democratization of this and just giving it in the hands of all their workers. I mean, maybe not all of them, but most. But then putting the right governance around it. Then we see some other, more reluctant manufacturing customers saying, look, I mean, I’m going to give it to this subset of the users where I can quote unquote, control how they’re using it.

 

Ozgur Tohumcu: [00:11:20] I mean, we believe on the first matter, I think the first group, because you can easily implement governance with some of the use of some of the cloud technologies and centralized governance methods. The other point that I’m going to make is essentially, you touched on that Jimmy as well, in my opinion, one of the biggest differentiators in manufacturing today, how fast you’re able to bring innovation to market. You have an idea? How long does it take for that idea to actually make it into a product, into a market? And I think generative AI has really shrunk the product design and development life cycle there. I mean, the design that you were able to do over weeks now you’re able to do within days, even within a day because you’re able to iterate so much faster. You’re just basically having a conversation interface with the machine and asking how you want it to be done a little bit differently. The development cycle has shrunk significantly as well because a lot of the generative AI tools are actually shrinking there as well. Testing validation has been so much faster now because we’re able to generate using AI test cases automatically that people were writing by hand, putting in Word docs and all that. So I think shrinking the product design and development life cycle and being able to bring innovation to market much faster has been the big thing that generative AI I think has brought into the manufacturing industry.

 

Jimmy Carroll: [00:12:54] Even in the wider space, it’s just incredibly useful for even small tasks. I saw I, I hate to admit it, but I was on a plane just a couple of days ago, and I saw a guy who I believe was an executive or, you know, president type, and he was putting his emails into ChatGPT. I don’t know if he was asking it to make him sound friendlier or to summarize or what, but it’s just like easy little tasks like that incredibly useful.

 

Ozgur Tohumcu: [00:13:21] Absolutely.

 

Winn Hardin: [00:13:23] So, we didn’t really define this. So, it’s for the users who may not be familiar with intelligent agent versus assistant. Ozgur, can you kind of separate those?

 

Ozgur Tohumcu: [00:13:31] Absolutely. I think that’s that’s a great call out. And we should have done that.Thanks for that. So, the way we see it, I mean, assistant is basically looking at a set of data and is making a recommendation around what you should be doing. An agent is actually looking at that same set of data and taking an action on that. It’s actually doing executing something. And in some cases you may have a human in the loop. In some cases you don’t have a human in the loop. But the agent is actually . . . The additional part is around the execution part.  And then maybe we haven’t talked about it, but you also have the next evolution, which is agentic systems. It’s essentially multiple agents talking to each other and therefore being able to execute tasks autonomously but in coordination with each other. And once again, I think this is super useful for the manufacturing industry because even if your data is sitting in silos, you can actually create one agent taking one action on your set of data. And you could have another agent taking another action on another set of data. But those agents can actually interact and talk to each other.

 

Ozgur Tohumcu: [00:14:52] You don’t have to bring all the data together, but the agents can actually interact with each other. I mean, I’ll give you an example to illustrate the point. I hope I can. I’m really, definitely going to now try to put this in. I’m thinking out loud now, so you need to bear with me. But an assistant would be essentially like you’re booking a flight, you’re searching for flights, you’re finding the most available flight, the time, date, whatever the price. The agent would be actually, you’re booking your whole trip with the car rentals and the hotel and everything. The agentic system would be more like you’re going to Spain for vacation for a week, and you ask the question to the agent or the agentic system, can you please plan a week-long vacation for me in Spain? And then the whole thing is actually designed and executed for you, including all the tours that you would be taking to the historical sites or to the beaches. So, I think it’s like different levels of complexity and levels of execution.

 

Winn Hardin: [00:16:01] I’m sure the agent would do a better job than me. For example, the last one of the last places I went down in the islands didn’t have air conditioning. And so, I mean, the agent would probably like, Winn, I know that’s historical and cool, but we know how you feel about air conditioning. So, they could have saved me a lot of trouble. You know,  at our other job over at Tech B2B Marketing we work with a lot of manufacturers. And in one case, we’ve been developing, working with the team to develop an intelligent agent. You’ll have manufacturers that have half a million SKUs, right? Across lots of different product lines, all kinds of different customizable parameters for each item. And, you know, they’ve had engineers that for years have been pricing those out as necessary, you know, developing CAD files and pricing out. But we’re finding that we can write intelligent agents that will reverse engineer these different pricing calculations and then help us and then apply that to all. And maybe they’ve only got one  or 200,000 SKUs that they’ve actually produced and have clear BOM on there and pricing structures. And then an intelligent agent that can fill out the other 350,000 is a massive time-saving capability. And then, you know, engineers can go back and ground truth a couple of batches at a time in each category to make sure that the agent is doing what they want it to do, make small adjustments as necessary. Can you give us any other examples, maybe without sharing the end-user customer, about those intelligent agents or even how manufacturers specifically are using those or some good use cases for folks?

 

Ozgur Tohumcu: [00:17:34] We have . . . I can actually share one with an example that we’ve been working with, the Kone escalator company for a number of years, and they’ve been a big customer of ours. I mean, these guys move probably, I don’t know, hundreds of millions of people every day and then they’ve been around for so many years. So, one of the things that they’ve done is, because and if you think about it, like how annoying it is when an escalator or an elevator is not working, it’s a pretty big nuisance for everyone.

 

Winn Hardin: [00:18:05] It’s been in the news recently. As a matter of fact.

 

Ozgur Tohumcu: [00:18:07] Yes, it has been. Yes, that is very true. That is very true. And I use the Tube, living in London, as you sometimes go to Tube station and the escalator is not working. I mean, it’s quite a big nuisance, and I think it’s something that has to be fixed pretty fast. So, what happens is when something breaks down, you call the company and then it needs to be troubleshooted. It needs to be fixed. One of the things we have done is working with them is we put in years and years of troubleshooting data, all of their operators manuals that are relevant for all different types and models and years of escalators and elevators these guys have built, all the customer calls, how things were sold or not sold into essentially our AI models. And then we created this assistant where when somebody calls with a problem, you’re having a conversation interface with an agent, and then, at some point, it’s handed off to the customer service representative, of course. But by the time you get there, 90%, 95% probability, the problem already is known. It has been detected somewhere else. Okay, I’m calling about an escalator problem in London, but it’s probably the same problem that was fixed in San Francisco, like five years ago.

 

Winn Hardin: [00:19:31] Right.

 

Ozgur Tohumcu: [00:19:31] So, I think this type of consolidation of enterprise knowledge coming from completely different sources, including the tribal knowledge that your workers might have, into a repository, being able to tap into that is outstanding. And this has been a huge success, I think, in terms of how they’re able to respond to customer calls, how fast they’re able to actually solve the problems. It’s been a big success working with the Kone guys.

 

Winn Hardin: [00:20:02] Outstanding.

 

Jimmy Carroll: [00:20:04] You know, it veers a little bit off of our core topic of manufacturing, but I’m curious from your perspective, Ozgur, like a lot of different technologies have made their way beyond the factory floor and into people’s houses, right? And a lot of people are using AI already, and they have been for years, and they don’t realize it. I’m just wondering, like we had a conversation with a company that does agentic AI a while ago and, and we were talking about its potential usefulness in the household. Have you seen that at all? Or could you see it? And, you know, and I made the joke before, but wouldn’t it be useful to know if my nine-year-old did a good job brushing his teeth, or how many cookies my six-year-old stole, or things like that? Like, will the technology ever be approachable enough for, you know, normal, everyday people to use?

 

Ozgur Tohumcu: [00:20:50] I think it’s happening now. I feel like it’s a little bit like, now you’re going to be able to tell my age, but like in, I don’t know exactly what year was that, but I remember the time when AOL was actually sending these CDs to everyone so they could download the CDs.

 

Winn Hardin: [00:21:11] Me too. We’re all of that gen.

 

Ozgur Tohumcu: [00:21:16] Exactly. We’re all from the same era. And so when we use,when we downloaded the software and were able to get on the  internet, I mean, it was a pretty magical moment, like something that was deemed as this internet that was sitting out there. All of a sudden you’re able to access it sitting on your personal computer or Mac. I think we’re at a similar moment when it comes to AI. You’re going to see it, like, right now we talk about this thing as if it’s like this other thing. I think you’re going to see it as getting into pretty much every gadget, everything we use at home, starting probably with the more smart equipment you have, your smartphone, your vehicle. But eventually, I mean, it’s going to make it down to your vacuum cleaner, your coffee maker. I think it’s going to become more and more embedded into the software that actually comes with . . . even if it’s like a firmware, it’s going to come in pretty much anything and everything you do, even your electric toothbrush at some point, is definitely going to have some level of AI in it. I think at the edge. And I think  that’s the way it’s going to become so pervasive, and it’s going to become part of our lives being embedded like into anything and everything we know. And at some point, five years later, maybe we’re not going to be asking the question because we’re going to know it’s already there. Just like you assume many of your things at home are connected. You’re going to assume there is some level of AI application or a toolset sitting in the equipment that you’re using at home.

 

Winn Hardin: [00:22:55] Yeah, it’s not hard to imagine especially large volume, like appliances, you know, a refrigerator . . . Using it for preventive maintenance, monitoring its operation not just so that you can avoid your refrigerator breaking on you but so that you can know when a compressor is running too high and it’s sucking up too much power. And, you know, optimizing your power bill, which would be a very real, very possible scenario.

 

Jimmy Carroll: [00:23:19] So, your milk is out of code. You could easily do that.

 

Winn Hardin: [00:23:22] I already know that. I, you know, date codes are optional. All right. I’m just saying, that’s just the way it is.

 

Ozgur Tohumcu: [00:23:28] To your point, I think the things in terms of . . . I think things that are easy to measure and show a lot of fluctuation and based on that fluctuation. There’s also I mean, again, I think you need to think about if you were to ask the question to yourself, where would be the first applications? Then I think the answer is going to be, the answer is going to be in those areas where there is enough benefit, like your electricity bill is not small. There is some fluctuation, there’s some level of unpredictability, there’s some level of pricing change based on your the way you use it or your working habits. So, that’s like a prime area to your, to your point, that you could see agents coming in and actually organizing things for you,

 

Winn Hardin: [00:24:12] 100%. It’s interesting because cost savings would be a big reason why manufacturers . . . I mean, ultimately, they always say it’s the power of the pocketbook that generally drives us, right? And of course, safety and other benefits too. But the pocketbook drives a lot of decisions in manufacturing. Can you . . . Let’s say that we’ve got some customers. We’ve got some folks out there in manufacturing who are considering how can I best use AI to make my enterprise more efficient? Is there an initial checklist of three or four things that they should consider before moving forward? Getting their ducks in a row before they come speak to AWS or or whatever?

 

Ozgur Tohumcu: [00:24:49] Sure. I think I’ll give two perspectives. One is a little bit around the mindset. The other is a little bit around where we see early applications being used already. So, I’m a bit more practical. On the mindset side, the first questions I would ask, I think all manufacturers should ask, is where is the bigger pain point in my operations? About two or three years ago when we were talking about, like when AI and generative AI became super hot, we saw a lot of CEO mandates essentially calling the CIO and saying, okay, what are we doing on generative AI? Please do something. So we saw all these like POCs pop up, which 90% of them didn’t make it to production because it was actually working on things that were not solving a real business problem or addressing a pain point. So, I think starting with the question, what is my pain point? Where do I want? Where do I think in my operations there’s a higher cost or there’s a lack of predictability? I think that’s where I would first advise, recommend for manufacturers to start looking into. The other thing I would say is that the tools are so easy to use.

 

Ozgur Tohumcu: [00:26:07] I would really recommend a lot of experimentation. And in a way that, I mean, if you’re not experimenting fast enough, the tech is changing so fast that you are going to be always working on something that’s already, I don’t want to say obsolete, but not the latest. So, I think fast experimentation and iteration is super important because you’re learning. As you’re learning, you’re basically applying your learnings. And then maybe the third part is think about scaling. One thing that we have seen is, when manufacturers or other customers are doing the experimentation, it looks fantastic on a few use cases, but all of a sudden, when you’re scaling it across your global operations or hundreds of machinery, you look at the cost and you’re like, wow, this doesn’t make sense. A prime example of that is around what LLMs or what large language models to use in the back end. What foundation models to tap into. If you have a very basic question, maybe you don’t need the most expensive model. I mean, you’re going to use the most expensive model, and when you’re experimenting on a few cases, you’re going to get a fantastically accurate response.

 

Ozgur Tohumcu: [00:27:23] But do you really need it? You probably could get the same accurate answer from a much less expensive model. That’s why, for example, one of the things that we talk about, if I can do a little bit like two minutes of AWS promotion,  we have our bedrock service, and one of the beauties of this service is it actually gives you a choice to all foundation models out there. It gives you price benchmarking. So, when you have a simple question, you use the cheapest model. When you have a very complex question, you go and use a more expensive, more elaborate model. So, I think these are the things like think about the pain point you want to solve, experiment, and fast iterate. And then always think about scaling, because when you scale, you’re going to see a different economics that you may then need to rethink your overall design. On the more practical side, like inventory management, supply chain, predictive maintenance, these are all areas that we see most manufacturers have started experimenting already because it really impacts quality, recalls, after sales type of issues. It has a direct impact on cost.

 

Winn Hardin: [00:28:41] It’s funny because when you were talking about, you know, first analyze the benefits and find that good use case. I mean, to me, I think to translate that to manufacturers is basically look for waste. Look for areas of waste and inefficiency, right? The ones that are costing you a lot of money. And that’s a good place to help identify places where it might be able to improve. And then I love your point and your second point, which is basically experiment, which means iterate. And you know, when DL, you know, early AI or deep learning was first coming out several years ago, a lot of people, a lot of the CEOs called the CIOs, they got integrators and everything else, and a lot of people got burned by the integration, the NRE, nonrecurring engineering costs to iterate. But now, with the advent and the prevalence of LLMs, to be able to translate an application need directly into execution, leveraging models that have already been well trained, because that was usually the huge component, right, was how do I acquire enough data to optimize my model so that I’m going to have enough accuracy? Is it correct to say that the prevalence of available models today, the growth of LLMs, have made it infinitely cheaper for people to experiment and iterate?

 

Ozgur Tohumcu: [00:29:50] Absolutely. I mean, I’m not sure if it’s infinite, but yeah.

 

Winn Hardin: [00:29:55] I’m in marketing. I can overstate, you know.

 

Ozgur Tohumcu: [00:29:59] But it’s it’s, absolutely. And I think it’s, of course, a very competitive space as well. You have new model providers coming in focusing on different parts of the problem or the business challenge. So, absolutely, I think the choice really matters. And then that gives you a lot of optionality in terms of what type of model you want to work with. And that optionality, I think, gives you a lot of flexibility in terms of how you want to design your operation. Maybe one thing that I, because what you said gave me another idea, one thing that we underestimate sometimes is, of course, in the opening you mentioned around data and cloud as well. I wanted to touch on that a little bit because it’s something we’re seeing in the last six to nine months a lot. AS our manufacturing customers are looking to leverage AI and generative AI, one of the things they’re realizing, oh, wow. You know, I can see how I can do all these things, but my underlying data foundation is not entirely ready yet. Or I designed it, or my architecture is not suitable to do this. Even some advanced manufacturers out there who have been using cloud technologies, sometimes even multiple cloud players, they’re realizing, look, I mean, I’m using these  multiple cloud players and I have some on prem, but now this doesn’t make sense because the AI use cases I want to run, it gets one data from here, it gets one data from there. It’s creating a lot of cross-cloud computing costs. So, we see a lot of the AI and what you could do with AI has led to rethinking of data architectures. How should I think about my data strategy? How do I think about the enterprise data? What is important for me? What should reside where? So, we see a lot of focus on data as well as a result of what AI has been providing as options for manufacturers to work with.

 

Jimmy Carroll: [00:32:16] I’m curious, you know, we’ve talked about a lot of different technologies today. Excuse me. And somebody with your experience and exposure to larger manufacturers, are there any other trends or technologies that we haven’t touched on that are notable? For example, I would think of cloud native infrastructure or software-defined systems. Anything like this that stands out to you that you’ve seen recently?

 

Ozgur Tohumcu: [00:32:41] Sure. I mean, quite a few. Let me keep it brief. One is definitely around digital twins. That’s a big thing. Ultimately, it goes back to a little bit to the concept of speed as well as quality. How can you actually create a digital twin of your operations so you’re able to decouple the hardware and the software so you can experiment on the software much faster, iterate, see where the problems are, see how you can accelerate. Let’s say product product development was one example we discussed. And then how you then put it back onto the physical systems and replicate that. I think digital twins is a big one. The other is, again, I mean, we talked about robotics a little bit, but I expect a huge increase of use of humanoids and more and more robots on the factory floor, especially with the coming with agentic systems that we discussed. I picture in a few years, you’re going to have humanoids walking around in the factories, handing off tasks to humans, and humans handing out tasks to humanoids, and they’re actually talking in whatever language they choose — English, Korean, German, wherever they are. I think the working environment is going to become truly hybrid. For some reason, we as humans, seem to feel more comfortable when the robot looks like a human. So, I think we’re going to see more and more humanoids show up on the factory floor. And to your point, I think the whole concept of what we refer to as, Jimmy, is a software-defined factory. How are you able to rethink the design of your factory or manufacturing operations so it is software defined, meaning you’re able to make changes much easier to it, in a way that creates an abstraction or a separation between the software layer and the physical machinery that’s in the factory.

 

Ozgur Tohumcu: [00:34:46] Because to the extent that you can create that abstraction, it’s going to make your operations a lot more flexible, a lot more agile, and you’re not stuck to the machinery that’s on the floor. I’ll give you an example from Amazon, not Amazon Web Services, but from Amazon. I recently visited a fulfillment center. And in the fulfillment center, one of our . . . These are huge warehouses that are essentially where the Amazon.com packages get shipped around and sorted and all that. As you can imagine, we use a lot of robotics, vision-based systems and, and these things. But one thing that I learned in one of my visits was there are quite a few humans in these fulfillment centers. And the reason being is humans provide an incredible level of flexibility in terms of doing one task today and doing another task tomorrow and being able to move around freely. And I was told by the fulfillment designer, center designer, is that they actually don’t like too many machines in the fulfillment center because when you have too many big physical machines in the fulfillment center, it creates a level of operational gravity and lack of flexibility. I think operations are going to become much . . . That’s why this whole concept of software defined, factory, software-defined operations is so crucial for the future.

 

Winn Hardin: [00:36:17] That is super cool. And I hope that we can get together at some future state and talk about, maybe get some guidance from you about how manufacturers would evaluate their existing data infrastructure. Because I just and then, you know, whether it’s in terms of load sharing, maybe a couple more specific examples in that area, but I know we’ve already kept you pretty long today. So, I guess that’s a little seed. I’m hoping we can get you back on the show in a few months.

 

Ozgur Tohumcu: [00:36:40] Awesome. I would be happy to.

 

Winn Hardin: [00:36:41] That would be super cool. Ozgur, thank you so much for joining us today. It’s been a real pleasure. For our audience, I know you guys have all really enjoyed this. If you have any questions for AWS, Ozgur, please post them below. We’ll make sure they get to them. Ozgur, where should they go to get more information about your section of AWS?

 

Ozgur Tohumcu: [00:36:59] Well, I mean, probably the best place to go is, find me on LinkedIn. Or you could do a quick search on AWS and manufacturing in any search engine, and it would lead to our web page.

 

Winn Hardin: [00:37:14] Awesome. Awesome. All right. So fire away any questions guys. Hope you liked this. Be sure to like and subscribe. You can find past episodes at Manufacturing-Matters.com. If you’d like to be on a future episode, let us know. And until the next time we all come together, have a great week!

 

Jimmy Carroll: [00:37:28] Thanks, everyone!

 

Ozgur Tohumcu: [00:37:29] Thank you, everyone!