 Welcome back to theCUBE's coverage here on location, AWS re-invent 2023. I'm John Furrier, your host. This is our 11th year covering AWS on the ground, on the floor. Start out as a small set, pulling people in from the hallways. Now it's packed house, 50 plus thousand, I think 60,000 people. Again, continuing to just thunder along, a lot of great announcements. This year, more than ever, has been the, I think the inflection point of generative AI and has bring that next gen cloud, next level legit, has been the conversation here. We've got a great guest here, Doug Bellin, worldwide head of smart manufacturing. AI's coming to theCUBE, theCUBEAI.com. Check it out. We've got the smart cube, we've got smart manufacturing. Doug, welcome to theCUBE. Thank you, thanks for the time and for inviting me over here. I did put a little plug in for theCUBEAI.com, which is our own little language model. We got 13 years of transcripts. We put vectors in and it gets us answers. But I mean, we're a small little use case, but the points to some of the things we were talking about before we came on camera, which is the big takeaway from this re-invent is the accumulation of all those years we've been talking about cloud and then you guys always had an industry focus, but this vertical industry focus around like say, manufacturing, healthcare, health sciences, fintech, they're the real winners in this generative AI boom because the gift of generative AI is the role of data becomes really more valuable and the more specialized the data, the better it can work across other data sets to actually produce better results. It's kind of nuanced, but it's the big theme here at re-invent, what's your takeaway right now? As we get down to things like smart manufacturing, is that inflection point here? Did we have it already? Is it coming? Where do you see this? It's coming. I don't think it's here yet. We haven't had it yet, especially manufacturing. I think manufacturing's an incredibly interesting market in here and I think a couple of reasons. One, manufacturing is the largest producer of data and most people don't think about that. They think, no, it's financial service, it's media and other things from that standpoint but it's an incredibly large amount of data but it's locked down, it's disparate, it doesn't talk to each other, it spews off numbers that a human being has no clue what it means from that standpoint and while we're the largest consumer or the producer of that data, we're the smallest consumer of that data. So there's a huge opportunity for us to actually start using that data, creating value in this market. And I've been on the company for six years, six and a half years and it has been that evolution of starting with first cloud and manufacturing being miles away from each other and every manufacturer going, I'm never going to do that. To now they're going, why am I not doing that? Help me accelerate and get into here faster as we go forward. It's interesting, and I was at a CUBE event when we were talking about regulation, regulated industries. Actually, Generative AI is perfectly set up for regulatory industry because the data's been so managed well. It's kind of labeled, it's got compliance. So a lot of low hanging fruit opportunities are in these industries that have had compliance and operating technology of the old school and IT kind of came together. We saw that collision, that kind of help happen. Now you've got cloud and data bringing together on manufacturing. How is that going to actually change the developer experience? Because, okay, now you have new ways to do things. What is, I guess I'll back up and say, what do you see as the key low hanging fruit for manufacturing to kind of go that next level? It's got a lot of data laying around. You got developer apps being built. What's the low hanging fruit? I think there's a couple things. One, we're seeing the evolution of the citizen developer. In the past, it was always a data scientist or something from that standpoint that first had to do data wrangling. How do I get to data? How do I make sense of that data? How do I understand what's going on? Second was, okay, now that I have that, how does it affect the business? Because a data scientist might understand the data, but the business is where the value starts to go. So how do you blend together that data scientist with the business person or make it easier for the end user to become a data scientist light like type of person from that standpoint? And if you look at some of the partner offerings out there, we've got a great partnership with a Siemens division called Mendix, which is low code, no code. And what they're able to do is allow a person on the shop floor to make an application. Now in the past, guys on the shop floor would be like going, I just need to use the application, but now they're not waiting for the person to bring it to them. So they're looking and going, let me just build this right now because I know where the data is. I know what I need to do from that standpoint. Yeah, and I think that's a great example of kind of this next generation that I call the genu. You're in charge. It's almost like this whole enablement of, democratization, I love that word, it's been overused, but it truly is democratizing in the sense of if someone could build their own app. But now, if you're back and say, okay, how do you get there? Like, if you look at manufacturing FinTech and just any kind of incumbent legacy environment, it's got a lot of components. And you're moving from that components and engineering. I got to send a request in. I got to get a developer to a more that an environment is driving productivity, okay, with simplicity and composability. Yep. Okay, that is an architectural shift. So okay, so assume the end state is low code, no code, build an app. I want to be more productive. It's got to be simple and composable. Yep. Okay, that's the here. What's behind that? What has to happen? Because it's not easy when you think, okay, where's the data coming from? Can I get accounting data? Where is the data? And that's the hard part a lot of times is understanding where the data is, understanding, like I said earlier, what the data means. You know, if you think about a manufacturing facility, there's 10, 15 vendors in there. You've got your Siemens, your Rockwells, your Honeywells, your Emersons. All of them have different languages. But on a single line, you may have a machine by vendor A, B, C, D, E, and F. You may have a machine that's out there, it's 35 years old. Hopefully it's no longer running DOS, but some of them are, which is also pretty scary. But now you've got, each one also probably has its own database. And what manufacturing has been good at is looking at each individual cell and fine tuning it to beyond Mth degree. Now, how do you look across the line? How do you look across the facility and then extend that to your supply chain? And you've got these disparate data sets. Great, okay, how do I get the data off of there? So with- You have to unify them. You have to unify them. You have to make sure that the same languages are there. You have to make sure it's human readable. The context is there. So in April, we launched a program called the Industrial Data Fabric, which is both AWS and partners. And it does, how do I collect the data? How do I contextualize the data? Normalize the data? And then even start to build a data model. Now, that is going to allow you then to have that citizen developer or gen-U, I love that comment, the gen-U of you make it now from that sample. You are in charge. As long as they have the security to get to that data. We're copyrighting that, by the way. Gen-U is the cube word. That's not what you can do for your cloud. But what it can do for you is what it will be what it is. And it is going to be down that level where they can get to that data. Again, security is going to be paramount. Do you have rights to get to the data? Rights to see what's going on with it? And then start to ask the different questions. And it's always been the problem many times is you get to a certain point in a question. You can't ask the next question. Now you have, not going to say you have access to everything, but you're getting better access. But no, but you're going to have, you're bringing up a good point. We brought this up in the cube again earlier on Monday. The words are changing with Gen-U AI. And it's almost interesting. It's almost words that we used to know and love. Words like memory, latency, actually have meaning on the data side. Latency to getting the right answer faster or that next progression, whether it's a words being spit out from a language model or a creative action on the move the next step on a solution that's being generated. Memory, retrieval, augmentation generation. Did I get the right response? Does it have hallucinations? Not physical memory, but that's good for storage too. There's even latency in the process. So think about a person standing there and if you've in your auto ever get an error message up and it is like a 12 digit number that means nothing to the human being. Is that like a hex message? What is this, a core dump? Right now. Wouldn't it be awesome that instead of that the car actually says or the machine that you're standing in front of says, hey Doug, you know what? I just saw these three things happen. That means in the near future this is going to be an issue. Wait a minute. One, the machine's talking to me. That's pretty cool. Two, it's not giving me a hexadecimal message. That means a zero to me. And historically what you did was you basically went off to the air conditioned room. You found the right file, which was 500 pages long. You had to go through that for about an hour and a half and find the one paragraph that you were looking for. Now that's right like that. So if that's a quality issue, or if that's the machine that's going to break in an hour or six hours, you just stop that. Yeah, and I think that you're pointing to exactly why generous, so powerful why it's going to change the expectation of what happens at that end point of the user experience is going to be changed. For example, that outcome is saving time, reducing the step takes to do something, and it's easy. These are major simple concepts that are winning for most for success. Okay, so let's take unpack that. The data's got to be there, right? So okay, we're seeing you guys do things like Health Lake, Security Lake. Are you going to have a manufacturing lake? So I'm a big fan of Data Lake. Never really liked the word Data Lake. I like the word Data Ocean because a lot of things are moving around, and currents and things are more different, but Data Lakes can be good if they're done properly. So what does the Data Lake concept look like, whether it's centralized to enable the intelligent fabric or whatever it would just, industrial data fabric. The industrial data fabric. What does that concept look like? What should be the principles of a good hygiene oriented Data Lake? The end point or the target for industrial data fabric is a Data Lake. Now, do we need to build a custom Data Lake for manufacturing potentially with you, but it's more use case driven? It's customer driven. It's customer driven. So instead of saying, okay, now how am I going to format this Data Lake for use particular because every business, so am I first, am I discreet manufacturing, process manufacturing? Am I regulated or unregulated? And all of the different permutations. And, or can I actually make it to a point where because we're doing with industrial data fabric, we're building that contextualization. We're building the data models into it. We're building the capability now that I could say, what application are you looking for? Oh, you're looking for a quality application to understand what is happening on the product. Guess what? That now attaches right to that data fabric, sitting on a Data Lake, and we can make it happen very quickly. You know, I get, I get nerded out when you hear things like, you know, context. And one of the things I like about general AI and the models, large language models and the concept of a context window. Very token oriented, very in the weeds, but like the point is, what's the window that you can ingest in to the AI? Now that's increasing from a token perspective. I think Anthropics is bigger than OpenAI at this point. But that's the constraint there. Okay, so, okay, now from the customer standpoint, that means what data is being fed into the fabric? Into the system. That's context. The behavior is the manufacturing. So we're kind of back down to the contextual, behavioral paradigm of a search, I guess. I mean, like, this is a data search retrieval problem or is that the architecture? Part of it is a little bit with that. I mean, but also part of it just has to do with making sure that the languages are the same. Like I said, every vendor has their own language. And, you know, we all use Babelfish or something like that to do translations when we're on a phone call trying to talk to somebody. We need that type of thing. GenAI is going to allow us to very quickly build that. There's, you know, do we need an LLM for industrial? Not sure yet, but I think there's going to be some segments of that. Take PLC ladder logic. Who codes ladder logic nowadays? But wouldn't it be awesome to be able to say, make that robot do these three things, and you kind of walk away and you come back and you look, well, wait, that'll work. Let's test it. I mean, we have an opinion on this. We think, and we published this research on theCUBE research. We have published the power law of AI, which is just a simple power law. It's going to have a big fat neck and tail, and there's going to be a long tail of specialty, custom, because it's all, beauty's in the eye of the beholder. If the customer sees data that's valuable, they're going to want to build some sort of model around that and leverage it as either as context input or into the process. So again, back full circle to why you're here is, people are really trying to figure out, Doug, how do I start re-architecting my system, whether it's rewriting code or throwing stuff away or rebuilding to enable, like the security business enabled, developers to be in line shifting left and doing it properly with governance built in and security built in. There's one thing that I would probably say, which is different in my mind and different in how I talk to customers is a lot of these statements or records, these source systems that are out there, that machine that is sitting there with his proprietary database, I don't want to rip that out because if I touch that, it's going to break everything. But, and it's so expensive, I mean, you look at a facility, it's how many billions of dollars to build a factory. They're not going to just go, hey, Doug, you know, that's a pretty cool idea, give me another billion and a half to redo it. So how do we take those 10 and 20 year old data sources, which are statements of record, but then you're building a layer on top and within it from there. But we're also looking at it in the long term where it's not just what I would call northbound. How do I get the data into the data lake, but is there a prescriptive way that the data lake can actually start to write back as well. So now you've got, you know, this is the mega version of that autonomous factory. You know, the autonomous factory is self-healing, self-generating, self-normalizing as you start to move forward. And to be able to get to that standpoint, you have to have those insights, you have to understand what's going, you have to bring that back, and then the machine starts to. That's the data unification piece. You're essentially not going to throw away the old leverage it and have mechanisms to figure out how to get that into the model to compose new stuff. That's the new, that's the first phase. That's the models. How do I first get the data? Again, that's where industrial data fabric is helping with us with that standpoint. Getting it so a human being can understand it, right? Part of it is that. I go back to that all the time from there. But part of it as well is, you know, right now there's the great surveys from a lot of our analysts talking about pilot purgatory. I go out there and I can do predictive maintenance on one machine really well. End of sentence, right? I stop there on purpose, right? But a factory has hundreds of machines. Well, the theme of this show has been end to end, right? So that's the ultimate end to end is factory, right? So I guess the final question is, what is a smart factory in your mind today? And how does that evolve? Okay, assuming that the architecture people start rethinking, reimagining, reinventing, the data piece of it, which you just laid out, what is smart manufacturing today and how does that evolve? It's more towards the self-healing, self-normalizing from that standpoint. So it is not a person sitting there in front of a human machine interface on that machine, hitting the keyboard every five minutes, hitting enter. It is the machine doing it from that standpoint. It's the human being, though, giving that input of the, because the cognizant standpoint of a human being is completely different and AI can't do a lot of that and probably won't be able to, at least in my lifetime. So I don't like the personless facility, but the autonomous is much better because the person is still in the loop. The person is still in the capability to be able to have that capable offering as we go forward. Human capital, Swami said that on his keynote, is a huge part of the scaling, the data in our heads, by the way, we have data there, too. Gotta get that in the fabric. Yep, thanks for coming on, great conversation. All righty, thank you. Great, smart manufacturing is just another example of how Generative AI is going to move closer to the edge and the customer, where the data is different and important, but super valuable for that use case and application while leveraging the cloud scale. This is kind of the paradigm. We've been talking about in theCUBE for years, horizontally scalable, vertically specialized, kind of all working together. This is what Generative is going to enable. And we'll be back with more CUBE as we get content coming to you later today from that location. We'll be right back, back to the studio.