Think AI Podcast - AI Won't Fix Bad Data. It Exposes It. | Ep. 21 with Dave Murray (FXI)

Episode Date: September 22, 2026

🎙️ AI Won't Fix Bad Data. It Exposes It. | Dave MurrayTwo reports. Two different numbers. One very long meeting.Dave Murray has spent 20+ years inside manufacturers fixing the part of AI nobody posts about: the data. He built the data foundation at Bose, then built Welch's AI program from zero, work Bentley University turned into a case study. Now he leads AI enablement at FXI.We talk about why messy data makes AI worse, how a bedtime story exercise turned casual users into power users, and why he never starts a process conversation by talking about AI.In this episode:00:00 Meet Dave Murray01:47 What changed from Bose to Welch's to FXI05:28 Two reports, two numbers: how analytics led him to data09:05 14 revenue metrics and a company that went bankrupt11:46 How to sell data management to executives15:08 The Best Buy deadline that made master data matter18:55 Building Welch's AI program from zero22:40 The bedtime story trick behind the Bentley case study27:31 From RPA to AI: train it like a new employee34:20 The pool ball rule for AI in manufacturing37:30 Why AI should lay out your three outfits42:49 Pilot first, data first, foundation first, then buildGuest views are his own and do not represent FXI.Which part of your business still runs on "my number is right"? Tell me in the comments. Subscribe for more real talk on data and AI.🔗 Links & ResourcesDave Murray on LinkedIn: https://www.linkedin.com/in/dave-w-murray/Dave Goyal on LinkedIn: https://www.linkedin.com/in/davegoyal/#DaveGoyal #ThinkAI #AI #DataGovernance #Manufacturing

Everyone's very excited to, hey, I want to throw Claude at my dataset, right? I want it to sit there and start to pull reports for me. Well, the problem is that now you're going to start to exacerbate all of these potential issues that you have. Welcome to the Think AI podcast. Each week, we talk about the most exciting AI research, tools, case studies and more. I'm your host, Dave Goir, and I've been working behind the scene in data and AI for over 30 years, whether you are an AI expert, skeptic, or something in between, this podcast is for you.

Welcome back to the Think AI podcast. I'm Dave Goyer, and my guest today is Dave Murray. I share the name with Head of AI enablement at FXI. So Dave has spent more than 20 years in the part of AI that doesn't get the headlines. And I know that really well, such as data, RPA, robotic process automation, and girls. governance, almost entirely inside manufacturers and not tech companies. A lot of this has been done in tech companies and it does a different profile. He has built the data foundation at Bowles, that his colleagues credit as the base for everything

that came after. He then went into Welch and built the entire AI program from the ground up. Bentley University actually did the case study on his work at Welch. Now he is at FXI, which is a mandate. And that is refreshingly honest that he's turning AI ambition into results that actually show up in the business. Now, one thing I want to mention that whatever views come from Dave is his own, it's not from FXI in any means. Two experts are talking today here. And that's what we are going to break out.

Welcome to the show, Dave. Thank you very much for having me. I'm excited to talk with you today. Perfect. And I share a lot of similarity with you starting AI in like. like 96, 97, we'll talk about it, and then jumped into the data and solving the data problems all across. And I love what you're doing here. So let me get started on the questions. Is that okay? Yeah. Thank you. So Dave, we have built AI data functions at multiple companies,

mainly enterprise companies, and it's different every time. So what is different? What's the same thing that you have noticed in the past maybe how many number of years? Yeah, I think what stays, Well, I shall start with what's different, right? Is that AI literacy has improved every single iteration. It improves, honestly, daily. So when I was at Bose and we were starting with robotic process automation and some machine learning, that was a very niche area, right? And there was essentially no one that was up to speed on what was happening.

Also, the technology was relatively new at the enterprise level. And we were learning on the job and trying to figure things out. as we went. And then as I moved on from Bose to Welch's, in 2023, when Chad GPT4 came out, right, everyone was really excited about it. They were using it. Their AI literacy went up. And as a result, when they came to work, they had some better working knowledge of it and we're excited to use it in an enterprise setting. Then at FXI, now 2026, AI is really part of many people's lives and those frontier models are causing people to just sort of understand the terms better, understand how AI actually works for them personally and then applying those learnings to their professional life.

And so what it does for me is it allows me to spend less time on some of the fundamental education and start to get right into some of the core of what people want to do and start to build those solutions. So that's what's changed over time, right? People understand it better. The things that have stayed the same is really being excited and being wowed by new capabilities, right, myself included. So if we look at like, you know, what Claude is doing on basically a regular basis, oh, it can access your emails or you can start to build artifacts and you can vibe code, you know, software solutions for maybe just yourself or for your team, right?

even if you're knowledgeable about large language models and maybe you use even Claude at home, the application of these tools to your work is what can kind of blow your mind on, for me, like a weekly basis. So when I'm going through my trainings with the team members, at no matter what level, there always seems to be something new that people's eyes kind of go wide and they get excited about what they can do next. So that thing that stays the same is the excitement and the, the wow factor of what they're being exposed to.

That's amazingly well put. And that's what we also notice with our customers too, which is you forget about how to get things done anymore. You really focus about what needs to be done. Now, one thing I always mention, and that's why I'm excited to connect with you, which is data before AI, because if you don't build the data in the right order,

guess what, AI will expose your problems much faster way. and it's not going to go anywhere unless you fix your data issues. Back then, it was very difficult in terms of life cycle to show the data quality. Few years ago, I read on CIO magazine that every other data project fails because of data quality. And a lot of people don't believe in it because, you know, they'll say, oh, yeah, I use ARP, especially manufacturing. I use MES. I use, you know, supply chain management software, PLM, whatever. But still, it's the interpretation of the implementer.

not thinking about how to read the data, but really how to write the data back to the system, isn't it? What's your experience looks like with data before AI? And what changed now that AI is doing things much faster? Yeah, I think, so I spent the majority of my career, right, was not in AI or even robotic process automation, right? It was in analytics. So I started in analytics.

And one of the things that I disliked most about my job, was having to justify my numbers, right? Because anybody that's worked in this field understands that you produce a report, and then somebody else at a different part of the company produces a report, and you have two different numbers, and they don't match. So then my job becomes defending my numbers, right? Why is mine correct? Or it is actually wrong.

Why is it wrong? So I kind of backed into data management because all I was trying to do was not have to have those conversations anymore about, you know, why is this report incorrect. And I think people that were on a similar career trajectory as me kind of have that similar path where they're looking at data, understanding what's wrong with it, focusing on the data management, looking at data management tools, right?

There's tools out there, Calibra, Informatica, a number of others that help with that process. And it's great that those companies exist because 15 years ago, there was some stuff. but master data management wasn't what it is today. And because it's getting such a spotlight put on it by AI, everyone's very excited to, hey, I want to throw Claude at my dataset, right? I want it to sit there and start to pull reports for me. Well, the problem is that now you're going to start to exacerbate

all of these potential issues that you have. And one of the things that is the most concerning, right, is synthetic data, right? The data that AI produces and then potentially will then use for future AI implementations or, you know, prompts or whatever. So then you kind of have this problem that can start to grow exponentially if you don't have a good foundation in where you start. So for me, it's the data management piece was such a key component.

And one of the things that I was helping with at Welch's was a new ERP implementation. right and that's a fantastic opportunity to say okay here's our old system we're going to migrate the data from the old system to the new system it can be a smaller subset maybe we only want to look back two or three years that's a manageable amount of data to cleanse put rules in place right put that data management software in there so that we protect the core know we have golden records and then put those into the ERP system once that's there now we can set AI on that data set and say, okay, now we're comfortable. We know what we're getting. And the defects that we get drive different conversations. It's not, is the data clean or dirty? It's, we know the data's

clean. Why are we seeing the results that we're seeing? What else in the process aren't we, you know, aware of that's causing these, these anomalies? So it really changes the conversation and it also builds more trust with your executive team because now they're not questioning the data set. They're putting the right questions on the results. Why is this happening? What can we do to change this, X, Y, and Z? So that's why it's such a fundamental piece and why I think a lot of people in my position have similar histories is because it's the natural progression of data or analytics, data,

data management, governance, and then AI. It's sort of like the next level. This is so soothing to hear all the components on data. I've been teaching and preaching even to my team. One funny thing just came to me just now. So I was working for a very large enterprise telecom organization, and they were creating these turnkey systems, which is like multi-million dollar size products.

And this company, and I'm not exaggerating at all, 14 different revenue matrices. And everybody is tracking their own. And their board meetings and or even a meeting, and or even other meetings too. They're always fighting. My number is right. And they come back with a lot of spreadsheets.

So when I went in as a data architect with Microsoft consulting from back then, about 16, 18 years ago, I think, that's the first thing to bring on table. Like, okay, let's just start whiteboarding. Forget, you know, data, master data management reporting. Just figure this out. Who will define the definitions? Eventually, we had to create a steering committee who can actually mark it down correctly. In any case, so then we started seeing the back orders when we start

building the first system, you know, for sales and inventory and quality and things like that, a few subject areas. And we started seeing, you know, they're losing because of open orders, taking so much time because it takes a lot of time to do it and they were advanced booking. And CFO loved me and I kept showing him the number and he liked it. Then he goes back to everybody in R&D and everywhere and everybody said, this is bullshit, my number says something else. Guess what? In about eight or ten months, they got bankrupt. That story. But the power of data is where I'm going with it, that if you do it right, and it's never about the technology, right? It's always about people and process. Well, technology too a lot of times. I should not

say that. But most of the time, you can control technology really, really well if you know what you're doing. And with AI, that makes it much easier. And so that was the story. I wanted to mention. And then you mentioned a lot of tools like informatica. I worked on Ebnacio, Informatica, Kognos, Microstatology. So that brings some warm, fuzzy memory. So since we are talking about data and you have a hero story to tell at polls, your colleagues credit you to build and do a data work, what you did there.

What it actually meant on day one, just like this story? What was the challenge you were facing? You don't have to go into the details or specifics of those, But really from that angle where people would learn why data, solving data problems, data system problems, building process and strategy is so important. That's where I'm going with it. Yeah. I think the most challenging piece, right, is that when you're trying to sell a data management program to an executive team, right, the challenge is, this is not necessarily out of the gate a revenue generating type of exercise, right? it's something that's going to cost money and you're not going to see a benefit from it right away.

So we had an opportunity where we had a product that didn't have the right, basically master data associated to it. I'm not going to the details, but basically it caused a problem at a retailer and it caused them to reduce some of their order. And what I did with that is I use that as the opportunity to say, listen, If we had proper data management, we would be able to, you know, catch this earlier and be able to not have this problem in the future. And that was the kind of the crux, the beginning, to say, all right, we do want to build out data management. We want to have data stewards in place. We want to have data governance councils.

We want to have all of this stuff, but we need to have a proof of concept, right? So going in the life cycle of a product, for example, and having proper data management with it and showing the reduction in defects, showing the speed in which you can get something from point A to point B, right? That's sort of the first seed that gets planted to say, okay, now we can do that with product master data. Let's try and do it now with customer master data, vendor master data. All of those things kind of grow upon themselves and you start to see the benefit. And then ultimately, you do see benefits. You see that we're going faster. We're doing things less costly.

Things are less costly to get across the finish line. We don't have to go back as often. There's a proper workflow so that things are happening in the right order. All of those things, while individually are very small, when you put it all together, has a really wide and large impact. And the work of doing all of that, again, back to data management. management is that you now have a system that is ready to receive AI. Before we even knew AI was going to be a thing, we just had the foundation built and ready

to go to say, okay, now we're clean, we're ready, let's go ahead and do it. So I think that was probably the thing that got us on the right track. We weren't on the wrong track. It was just sort of got us onto the track of being ready for future state, whatever that happened to be. it was AI, but it could have been any number of things that came through. That's really nice and amazing. And you were talking about, you know, data governance and also master data management.

Did you end up creating any specialized form of master data management like PIMP, product information management or customer, sim? You had to do that. Did that came through and having a data steward to manage that and have the committee or someone who can steer towards really focusing and paying attention to the master data because transaction data is transactional data. It will be captured as it will be. But master data is where all the analytics lives, right, whether it's AI or data.

So yeah, like a PIM system, right, product information management is so important. But it also gives us a standard, right? So when you put your data set, your master data for a product into the The name's escaping me right now. I can't remember the name. But basically the flow of master data that exists for retailers that they're able to then pull your data. Well, you have a standard. You have to have certain key elements.

So if you're selling to, let's say, Best Buy, for example, right? They need to know a certain time in advance about the weight of the product, the size of the box, you know, some costing information if it's available at that time, color options that would be within there, how many boxes you can fit on a palette, right? This is not glamorous stuff, but it's the information that needs to be there. Well, that gave us a standard to say, okay, we need to have all of this information ready by this time. Well, that kind of puts a line in the, you know, on the calendar and you're like, okay, now we have a date we have to hit. And you can kind of start to build out your process, your master data process based on, okay, Best Buy needs it by this date. We work backwards.

what's the absolute latest we can kick off this process. And then if we say, no, we can't kick off the process until two weeks later. Well, that means Best Buy can't get their data for two weeks. Are you willing to make that tradeoff? By having that foundation, you're able to then kind of put some firepower behind your lines in the sand. You're saying, okay, we don't do this. We can't work faster. We've brought this process down as much as we can.

So if you say, no, we can't have it until two weeks later, that means it gets pushed out two weeks later. Are we willing to do that? And the answer is often no. So then you have to have a different conversation about how we can start earlier. So that's really kind of all the data stuff and stuff that's really not that exciting. And maybe it's not as exciting to your listeners about an AI podcast about the data side of this. But it's all of those enablers that make all of this other stuff possible within an organization. No, this is perfect.

And I think people do need to listen to that because AI looks really fancy. And with all frontier models, it feels like you can do things. While you're sleeping, you know, you send a prompt or a goal or a loop in an agentic harness and then you go away. But it's going to build the bad stuff just because you don't have the right data. So focusing there makes perfect sense. And I appreciate that you're detailing that out and not leaving any bit out of it. So thank you so much for doing that. But since you mentioned the podcast and my company name Think AI too, let's jump there.

There's the most exciting questions I think listeners would love to hear, which is you build the Welch's AI program from zero. And a lot of companies today doing AI just because they have a formal. And most companies and people, you must have noticed, do not understand the levels of AI or what AI means. It's just a former AI where you really need to have something. And I see a lot of people. people still doing legacy.

And I've done that in my past. Machine learning, I have a patent in that too. So they're only doing machine learning and doing those kind of things. And that's not bad at all. But at the same time, you can speed that up quite a lot, as long as you understand all the elements correctly. And there are like four or five levels of AI. But before we go into those levels,

how did you really convince when there is no program that this is what we need for which this has the value? Because there's an investment attached to it. there's a foundation attached to it. How does your first 90 days split? What was the real first decision that you made? I'm pretty curious to know that. Yeah.

So when we were, we were obviously all companies were having the conversation, right? Should we enable AI here? What can we do with it? Is it safe to do it? So the first thing we did was a pilot, of course, right? And we ran a pilot with a couple LLMs, actually three, to determine what was the right solution for us. And the way that we did that, right, is I wanted to get the executive team on board.

So I made sure they all had pilot licenses to be able to try them out. And the way we did it is we had some people would have clogged, for example. Other people would have, you know, Gemini or Chad GPT. I think everyone had Microsoft co-pilot, so I think it was just part of the suite. And then we said, okay, great. Everyone for now, it's a sandbox. Go ahead, do your thing. It's all protected.

We kind of gave some policy. on what, you know, you can and can't do, certain things like that. And then we said, all right, now I want each executive, I want you to name three individuals within your organization that you think would benefit the most or would provide the most value if they had access to this tool, right? So we expand the pilot out to those folks. From that point forward, I then had weekly luncheon learns

where I was just going over basically the fundamentals. I don't make the assumption that anyone knows anything, just because they have access to the tool, education is absolutely number one. And then I would record those, right? So you could go back and say, okay, what's, you know, chat GPT 101 or whatever model we were focused on at that time. And we would then start to grow, right? And then what would happen is generally they were leaders within the organization that were picked by the executive team. And they were showing their teams.

And the team's like, oh, wait, you can do this. Wait, I want access to that. How do I do that? Okay. So then the groundswell of interest grew, right? But we don't just want to blanket the company with access to the tools. So then we also made sure we had kind of enterprise level activities that we wanted to do. So it's sort of that top down, you know, all the projects we want to have.

So we've got the top down to the projects the executives want to see. Then we have a groundswell of individual contributors, for example, that have a lot of use cases that they want to try and build and do. then in the middle, bottom up, top down, you have the layer of teams that have some actions that they want to do that will help their functions. Okay, so now I've got the groundwork together. So then we make the decision on what LLM is going to be the right solution for us. We then say, okay, we're moving forward. We're going to do an enterprise license. We're going to monitor the usage, right?

We're going to go low. We're not going to go with the everyone gets it. We're going to say, okay, we're going to have a certain amount. We're going to fill to that quota. We're going to then monitor the usage, see who's using it, see who's not. And then after that, continue on with the learnings that we're going to have. And, of course, the entire time we're developing new policy and making sure that we have the right solutions in place. The thing that really unlocked everything, and you mentioned the Bentley.

Bentley College, doing their course-long study on my program, was because we did something a little different, which was we weren't getting as much adoption when we were, we started to push it down. Of course, you have your early adopters who are very excited and they're going to be working and they're going to be advocates for the program. But then we wanted to also expand it out to teams that would benefit from it, but maybe they don't have any necessary, they don't have any interest in AI or there's negative connotations about AI, right? You fight those battles as well, people that aren't interested in it. So the reason that I was selected by Bentley was because the program that I devised actually had everyone go home and use it.

Try out some things. Everyone does the meal, make a recipe or something like that. The things that I was trying were like, at bedtime, I did this with my kids, right? Come up with the craziest story.

Let's make crazy characters and let's have crazy situations and then let's give it to chat GPT and have it create a bedtime story for us or turn into a song or a rhyme, right? Those types of things started to permeate within individuals. and we found that the people who were using it personally were graduating to high-level,

high-consumption users within the organization, right? So we found a strong correlation of high-impact users who also used it for their personal use. And it's that sort of duality of using it in both worlds that really started to unlock the potential of the tool. So that's really how we kind of got it going. And then once we have everybody up and running, we're doing regular trainings, and then we're starting to prioritize projects and things like that. That's really how things got going. Well, that's pretty amazing.

And, you know, I talk about three A's awareness, acknowledgement, and adoption. And I think you kind of lived on to that. Making it aware is the biggest thing. Because most, if you remember past in our days, data projects used to take so much time that the only end result was reports. You unfortunately cannot show how that ETA transformation, how all that thing is happening. I mean, you could, but they were not so much interested. With AI, it is much easier to start showcasing fail early and see and adopt incrementally what is going on.

And you kind of did it. So congratulations on that and also for that case study. One of the interesting use case I personally had, so I'm a disabled entrepreneur. I'm dyslexic too. So a lot of things I forget. My AI knows that. And whenever I make that mistake, my mistakes are on number. So actually build a product with my friend called readon.com for learning disabilities, for that matter.

So it doesn't fix mine. I put it with a lot of willpower, but it does fix the kids between the age of 6 and 16. So that was a good example. But we did it when this like 2015, 2016, 2016 traditional AI, but we are still able to solve the problem. But today, you know, I'm also a musician. I play guitar but don't get time. I would like to compose, but learning, oh yeah, I saw that. That's why I was excited to say that.

So I wanted to learn Logic Pro to start doing recording in production, but it takes a lot of time and a lot of discipline. So with AI, what I do, I just did it like a week ago on Cloud Co-work. I threw a chord sequence to it, a small write-up that I returned to compose, and gave some court sequences like this is how I wanted for my main. and chorus and things. And it created an amazing composition. So composition is still mine, by the way.

You know, I don't want to give air credit. But where it took it, because it started thinking like what I was asking for it, you know, creating a particular rhythm section. It's quite difficult for me, what I'm imagining and putting it on to the logic pro. So those are the possibilities I see. Like my son has a harness from me. We got a Google funded startup for him. He's 15 right now.

So we built a math quiz builder for him. And the harness is completely controlled. So he can't cheat. And that's his account. So what he does, whatever he studies, he spoke it out to Guy on Gemini. And then based on that, the AI can figure out his own gaps. And his quiz is completely tailored towards him. So he was scared about certain things like chemistry.

Math is really good at. But when he started talking, his grade improved literally. and he would hate me saying this. So literally from that C grade on the first one to A plus in the literally on the second one. Wow, that's great. And it's a guided education where you can. And I'm saying this story because of a lot of AI skeptic thing that is going to harm. And, you know, there's a lot of talk.

Every technology will harm. If you have a faster car, it can kill people, right? If you have autopilot cars, it's going to do its own thing. So everything comes with a risk. But when you live in a controlled environment and see the goodness of what it can bring, I think the new edge world, I'm loving it, especially for innovations and things of that nature. And your stories are connecting to that. That brings me to another question, which is, you know, RPA.

And RPA was all about discipline, how you build things, versus AI is all about flexibility. How do you see that real evolution from RPA to this world? Yeah. I think, so going back, right, the mechanics of it have really changed entirely, right? So we were using a tool called UiPath, right? And I think a lot of people are familiar with that. I remember. I know it too, yeah.

Yeah, yeah. And it was phenomenal, right, at the time. And it's, I'm not familiar with it now. It might still be unbelievable. I've not looked at it in a number of years. But the way that one worked, right, is it was UI stands for user interface, as you would imagine, right? And it was really kind of mirroring the work that you were doing and trying to understand the screens and start to make inferences from the work that you were seeing, right? And that was groundbreaking work and it had a lot of potential, right?

And you would have an orchestration layer and there was a lot of pieces. It was, it was a GUI, right? It was, you could see what everything was doing, but it was still somewhat labor intensive, right? You were still very much along for the ride for a lot of it. And I think what's happened over time, right, is that things have gotten far easier, right? And I don't think we saw things like MCP, right? For example, model context protocol to give us those gateways into other softwares that allowed us to start to build out some solutions. And the ability for ourselves now to take datasets and maybe you want to build a rag architecture,

vector databases and kind of build on your own data sets, or if you just want to set up something, you know, even easier than that, which is just sort of dumping files into clod or creating a custom GPT within OpenAI, you're kind of training, like I had a conversation recently with someone who was trying to have AI count boxes, right? That's what it was doing.

And what they were doing was they were loading the images and saying, hey, count the boxes, and then they were coming back and the number was wrong. And they were telling them like, no, it's actually 14 boxes. And the conversation I had is actually, you know what, do this. Actually go into Microsoft paint, draw a grid, put the boxes around the eye and put one, two, three, four, five, six, that. Do that for a lot of them so they can start to understand where the edges of the boxes are, start to do that. And then have it, you're training it on your own and then have that, you know, as a skill.

and that is so different than it was 10 years ago. And it's honestly like training a new employee. Like, this is what the conversations I have is like, well, how would you train a new person who's, imagine there was a person who was incorrectly counting the boxes from images. What would you tell them, right? Tell them the same thing. And we're kind of starting to blur the line of how you would teach a person

and how you would teach a machine how to do things. And that advancement and also that change in our own minds of how to do that, I think is what's going to unlock the next level of functionality within these tools. So to me, it's just the evolution and not knowing where we or what we would have available in the future. And I think we're in the same point. What's the next MCP, right? What's the, we had APIs. We had MCPs. What's the next thing?

I don't think we necessarily know yet that natural progression, which is also why, even though it can be tired. to staying on top of the AI news and listening to podcasts like yours and keeping up on the things that are happening so you can start to get those early indicators of, oh, hey, this is the next thing that's happening. Let me learn about that. See if that's valuable for me or my company. And let's dabble. Let's try it out and see if that's the right solution and see if there's benefit there.

Because unfortunately, the cutting edge is changing on a weekly, if not daily basis. And, you know, you want to be there. If you're excited about it, like I think we are, right, and other people that may be listening to this, it's also fun, right? So that's the other part. It's the most exciting time in my career because there's so many new advancements and interesting things happening. You correctly articulated it really well that this is that exciting time. And, you know, you can't really bet on, you can't fight a battle that plot is better or OpenEye is better or Krogpot is better or one of the Chinese model is better. everybody is fighting this battle.

And they are coming up with something amazing every time. And like you said, the new concept in 2026. So it started with prompt engineering, which is not a thing anymore. It went into context engineering because the size was pretty small. I think initially 4K and then 200K and then 250K. And now a million and people are talking about a billion already. They are already testing on it. So that's a new and exciting thing, right?

So then came, okay, so you can build the context, you know, and start summarizing everything, but the window is still the same. So how much can you repeat and it will still start to forget? So then people started to put together harnesses. And harnesses like Claude is one of the best harness I've ever seen. And there are some others out there. And I've plugged in different models to Claude also, not just Claude models. And I use all kinds of models, including local.

But the harness was amazing because of hooks and things like that. a little bit technically here. But the thing is to learn that as a concept, which is agentic engineering in general, all three things. I think that's where the career is going. People talk about FDE forward deployment engineer. People also talk about builder versus so many other things. But I think one change I see, AI says four, not AI,

but most of the magazines are saying that four out of five jobs will be AI jobs. And that's kind of true because a lot of those jobs are repeated. So either you are a hyper-specialist of what you do or you are extra ultra-journalists, I should say, you know, and then you use AI for it. So either you will be an AI expert or you will be an expert of something, but you can't stay in between where you do a repeatable job. And that's the change I'm waiting to see how that will play in the real life. It's not there yet. Right now we are getting excited with the models. But then when the vertical models will come, like K.

when you really build a proper CFA who knows everything about accounting and who can, or you can build a paralegal person who can do everything for you. We are not there yet. We all are building today, but we are not there. I'm pretty excited to see that. And I have that question to you, but let me wait for it. Before we go there, what one shift you see in manufacturing organizations that should run for the next three years?

What's your wish list there? I would say that there's so when we talk about I'm going to go into the process a little bit right is sure of course when when we look at a process I refer to it as like playing like pool right billiards where I'm pretty confident that if I have a cue ball I can take a pool stick and I can I can hit it into a pocket that's like one task think of it that way when I have two pool balls I have to hit one into to another into the pocket. That's like two tasks. That's a little more complicated, but I still think I can do it.

If I have three gets more, it gets harder. Four, five, as you can see, right, exponential. That's sort of what we talk about right now when we talk about these tool AIs, these frontier models, right? They are good at accomplishing a task. You start to tie multiple tasks together to have a process. It starts to get more complicated and there's more things that can go wrong and you kind of deviate, right? So the conversation that I'm having is really around how do we identify an

entire process map? And I don't have a conversation about AI. I don't want to talk about AI in the process map. I want to say, okay, good. This is the process. What are the tasks within that process and who's the owner of those tasks? Great. Now I have kind of a bite-sized piece that we can look at. And then I say, all right, in your task, what are the steps that you do within your task? And are there spots within that task where AI can assist? I never go into a conversation saying, AI is going to help with this task or this process. Because then we're forcing things down. We're forcing something that may not fit.

And we're also not looking at it from the lens of how are we enabling our employees to do their best work? And that's what we want to do. We want people to, have the less, you know, the tasks that require maybe more repetitive, like you're saying, that we don't necessarily need the human in the loop for they can do that piece. We want to have it so that those employees are working with AI as a thought partner, right? The human is the thought leader. And they're working together to solve their task.

Now, as we accomplish tasks, which are in part or are either not included, including AI, somewhat including AI, or maybe can be automated through AI. We can then link those tasks together and say, okay, we have one accomplished task. Then that kicks off the next part of the accomplished task. It's not, hey, do this entire process. It's not, maybe it will be at some point. I still don't, I'm still a very, human in the loop is the most important part to me because there are just so many variables that can go awry.

And we need to have that human expertise in that step to make sure that we're accomplishing it. So from a manufacturing perspective, I think that is exactly the same, right? We want to make sure the person is there making those important decisions. Another analogy that I use, right, is when you get ready for work in the morning, right, you look at your closet and you have all of your clothes. and you make a decision. You do it very quickly, probably. You just decide what you want to wear. But if I wanted to make my life even easier, right,

I would have my three outfits laid out, and I just had to make a decision of between one and three instead of I've got five shirts, four pair of pants, six different T-shirts, and how do I want to combine them, right? When we lay out the options, which AI can do for us, right, and say, okay, now you just have to make a decision. and you use your decision capabilities to make the right decision on what you're going to do. Well, that takes less time.

You're making a more informed decision. You're including your own expertise in it. And you're also helping AI understand what good decisions look like. So for me, in manufacturing space, it's really about having people there to make the best decisions they possibly can and have the kind of grunt work be done by AI so that the person can really focus and start to elevate and go to the next level of what they're trying to accomplish. Yeah, and you said it well,

AI, human in the loop, and I even say human in control, is highly, highly needed if we are getting scared or skeptic, that's one area we need to focus. So the answer cannot be, don't use AI at all, or use it at a level that then it screws things up. I've seen horror stories about it deleting the whole hard drive for you

or, you know, doing certain things that you never wanted to be done. And this is where you're losing control. I used to be a white hat hacker. So I always look at it with that skepticism that, you know, anything that can go wrong will go wrong. So what are you going to do about it? How do you really salvage out of it? What's your backup policies there?

business person or anything and that's the story you are mentioning and I'm loving it one last question this is more on a personal level so like I said disabled entrepreneur going into the second phase of my life where I want to give back to disabled kids across the globe that's why I started building the podcast I was scared to talk because I was you know on the lap of my father and used to put me in different seats and places so just speaking was a challenge then going across the world became a big thing then leading nine companies five failures and building one company which is three times INC 5,000 winner was a story that I love or personal level not to brag but is the story I should be telling so and the reason why I'm saying that and I always said that to motivate kids so leadership

motivation and building solutions personal or business using technology and now with AI also is my three things that I would love to do what's yours what warms you and what you love to do. And it could be technology too, but what's your things there? Yeah. Well, I mean, so for me, one, personally, I just, I like woodworking and I like building things and, you know, using, using my hands. And then I also have a big 3D printing, I love for 3D print thing. So I've got a little workshop where I'm doing stuff. Never, it's not like a side business or anything. But one of the things that I do is, I also help local businesses in building like websites, or coming up with some AI solutions.

I do some work in adult education as well. Because for me, the most rewarding thing is opening people's eyes up to this technology, right? And I think there's obviously there's a huge amount in the media about AI. And it's all over the place, right? Some good stuff, some bad stuff, some scary stuff. And I think that the more informed you are, right, you can make a better determination on how you feel about it, right? And to me, I'm not trying to drive people's opinion. You have your own opinions about how this plays out.

And I also think there's a lot of unknowns, which make it very difficult, which can also make it scary. And to me, it's about getting people to just understand it more. Because I think the more that you understand, the less imposing it is, the less daunting it is. And you can kind of decide for yourself, is this something that I want to continue to learn more about? But at a minimum, I think small businesses can really benefit from just having a more improved website, adding a chatbot to your website to help with some of the overflow calls or scheduling, things like that. And also I dabble a little bit in app development and stuff like that. So it's all, I'm always trying to learn more and to teach more.

And that is what kind of continues to excite me. And I'm looking forward to doing more with more people and doing stuff like this. This was really exciting. This is actually my first podcast I've ever been on. So I really appreciate you inviting me on to do this. It's been really fun. No problem. And we are happy to have you.

And few things stuck with me after talking to you. was a great natural conversation, I must say, and I love that. And I picked up on, I'm always about three things. I don't know why. So three things I picked up. Data first, foundation first, and then build, right? So pilot first, data first, foundation first, then build something. And I think that's the right way to do it and build it incrementally, not create something humongous, which will fail, and it's too big to fail, and then you spend your life on it. So love that conversation. Happy to stay in touch with you. You are always welcome.

Any way I can help, I would love to help. And good to have a friend on the other side now. Thank you, David. Absolutely. I appreciate it, Dave. Thank you so much for your time. Thanks for being on the show. All right.

Talk to later. Bye. Thank you. You have been listening to Think I podcast with Dave. Take one idea from this episode and turn it into action.