Shawn Cordner (00:26.68)

Welcome to the Amplitude of Tech podcast. I’m your host, Shawn Cordner, Chief Marketing Officer at Amplix. And today we were joined by Scott Weiner. Scott is CTO and Chief AI Officer at NuVion. He’s a seasoned expert in the technology industry with a deep focus on artificial intelligence and its transformative impact on business and regulations. In this episode, we dive into the evolution of AI, its implications for business, and the delicate balance between fostering innovation and ensuring safety through thoughtful regulation.

Scott shares insights on educating policymakers about technology and navigating the complex patchwork of state-level AI laws. We also discuss how organizations can balance purchasing AI solutions versus building custom applications and the critical role of data management in driving AI success. Scott introduces the emerging importance of prompt engineering skills and outlines a robust framework for implementing AI strategies effectively. We wrap up with a look ahead to the future of AI, exploring the rise of agentic AI and the integration of robotics into enterprise.

operations. Whether you’re just starting your AI journey or looking to refine your approach, this conversation is packed with actionable insights and forward-looking predictions for you that you won’t want to miss.

Shawn Cordner (01:46.798)

Scott Weiner, good to have you on the podcast. How are you today? I’m doing great. You are the first guest of the first episode of the Amplitude of Tech podcast. So thank you for being our guinea pig. Appreciate it. For the audience, we’ve had some technical difficulties. This is our third go around. So that’s why we’re smiling a little bit. But we’re going to make it happen this time. I feel good about it. So.

Good, how are you doing, sir?

Scott Weiner (02:08.778)

We’re a little giddy on it.

Shawn Cordner (02:14.74)

Scott for the audience. Why don’t you just give yourself a little introduction? Who you are, and you know how you got to where you’re at today in the technology industry.

Sure. My name’s Scott Weiner. I’m a technology strategist at Nuion. We help companies navigate their technology journeys, and I’m very focused these days on AI in particular, which is why we’re talking today. I work with companies that assess their AI readiness and develop governance strategies, and define their ROI for their AI initiatives. For instance, I recently was helping a manufacturing company with a legacy system, trying to determine how AI might help them automate and support migration to a more…

modern ERP and also enhance their customer and sales insights. That’s a typical type of project we’ll get involved with. I also work with leadership on developing high-performing teams within those initiatives. So I do a lot of leadership coaching as well. I’ve been doing this for about a decade, and before, probably the last few years have been focused on AI. Before that, it was focused in other areas of technology, doing a lot more technical design analysis, due diligence projects, things like that.

And before that, I was a serial entrepreneur for many years of CTO of various companies. So that’s kind of the gist of how I got to do what I’m doing now. But a lot of the work I do is, I would call it in the educational space in the sense that there’s a lot of need within organizations still to have a shared understanding of what AI is, what it could be, and what it isn’t as well. And so we do a lot of work like that as well.

You have told me in the past that you started working on AI in the 80s. What was AI like in the 80s?

Scott Weiner (03:54.84)

So, know, AI has been around since, you know, I guess the 50s, 60s, I guess is when it first got coined. It was much more expert systems or symbolic type of rule-based applications of AI. So it was much more, you know, if this, then that was the way we built systems, the stuff that I worked on anyway. There’s a variety of things that fall into the AI blanket, which is why I talk about education, because there’s a need to understand how this is different to what we’re doing today.

And that’s how it’s the same as well.

Yeah. Speaking of education, one of the things that you’ve told me in the past that you are involved with is you spend a lot of time working with Congress and that some of our leaders are coming to you and asking you questions about AI. Tell me a little bit about the work you’re doing with Congress.

Yeah, so I work with a trade group called Act in DC and we go a couple times a year, we’ll go to visit Congress, we’ll talk to various senators and House members and their staff and we’ll talk to the FTC, the FCC, various departments, depending on what the policy is we’re dealing with. I focus mainly on small business technology issues. So AI falls under that, but so do things like privacy.

patent law, things like that. So there’s a variety of topics we’ll talk to them about, and we’ll give them the small business perspective on any regulatory work they’re doing in that area. So last year, I got contacted by the group and they said they wanted to educate Congress on what AI is, sort of what we were just talking about. So I did a presentation for a large number of congressional staff and a few of the congressmen and senators showed up.

Scott Weiner (05:37.354)

Representative J. Obernolte, who I think he has the AI task force on the house side, introduced us. And it was really kind of an introduction for a lot of them to when we say AI, what do we really mean? And I actually showed them like how it works. So they actually walked out going, that’s not what I thought it was, which is exactly what we wanted. And so that helps with them when they’re having these conversations, at least they know what they’re talking about while they’re making these very big decisions.

Yeah, it’s interesting. Seems like in Congress there’s a bit of a gerontocracy where the people that we have in office are of a different age, of an older age. I think the general public wonders how in touch they are with technology and how quickly it changes and evolves. you feel like they’re getting their finger on the pole?

So a couple of things about that. So first of all, I was pleasantly surprised that there are a couple of very knowledgeable people within those organizations. There’s a couple of senators and congressmen that I met that absolutely, not just them, but also their staff have their fingers on this stuff and really understand at a deep level. So I’m not gonna say blanket that they all do, but the ones that are really interested and engaged in it seem to know what they’re talking about at some level. They’re not…

they’re not working off a fictional definition of what AI is and can be. So I found that very refreshing. In terms of sort of the regulatory climate right now, what I’m hearing and seeing is that, first of all, there is talk of the new leadership coming in is gonna rescind some of the orders that are already out there. the current administration’s kind of executive orders around AI are gonna kind of probably push to the side a little bit.

I think there’s a lot of wait and see going on right now. There was all this regulatory energy this year around AI, a lot of concerns and so on. And it sort of feels to me, and I’m on the outside, so I can’t tell you for sure. It just feels like it’s come down and it’s not that it’s gone. It’s just that I think they’re waiting to see what happens. There’s this whole new initiative around regulations with Elon Musk and all that going on. And then there’s other parts of the government that are looking at this. Now that doesn’t mean

Scott Weiner (07:54.51)

They’re not thinking about policies having to do with AI. It’s just that this overarching regulation around AI, I think has kind of been dampened a little bit. The focus last year when we were talking to them, this is a year ago now, so it’s not the newest stuff, but was really on the large language models or what they call the foundational models that really are kind of underpinning all of the work going on and trying to figure out how to regulate those because they represent so much risk.

And so that’s where the talk I think still is around those big models and what should happen. The problem is the smaller models or things that I might do within my own business, what happens to those? Do they get sucked up under that or not? And I don’t think they’re ready to answer that question. So national security is definitely being talked about. There’s growing focus on how AI can enhance national security. And this includes applications in defense and cybersecurity, of course.

I think the other thing I’m hearing a lot is a lot of talk around what do we do about misinformation, concerns about the role of AI spreading misinformation, deep fakes, all this stuff. So it’s talk at this point. I don’t know of any specific regulations that are really flying through right now. I’m trying to think what else. well, how do you balance innovation with safety? I think that’s the biggest and actually the most important conversation we can have right now. There’s, depending on which expert you talk to, you’re going to hear

AI is about to take over the world and other people are saying it’s a fad and it’s not either. It’s something in between. And the question is when we talk about the global competition around AI, how do we unleash our innovation and still be safe about it? And that’s where the conversation really goes. And if you look at what’s going on in the UK versus here, I got to tell you, there’s a whole bunch of applications I use. They’re not allowed to use. So I have a competitive edge right now. Now, is that a good thing or a bad thing in the scheme of things?

I feel like it’s a good thing, but I’m not sure. That’s part of it. And the last piece that I’m hearing a lot of talk about still, it’s been going on all year long and it seems to be getting noisier, is what’s the economic impact on the workforce displacement? Because the reality is AI is not going to replace everyone’s job. I know that’s one of the fears out there, but it’s certainly gonna have an impact. And so looking forward, what’s that mean and what’s the government’s role in all of that? I think that’s another long-term conversation going.

Shawn Cordner (10:20.366)

The regulatory priorities that you’ve seen them addressing, how are they impacting the business community?

So what we see in the businesses that we interact with anyway, so this is little bit more micro because this is down to of anecdotes that I see in the businesses we work with. It’s not affecting them too much yet unless they’re in, well, let me back up. If they’re in a regulated kind of highly regulated business like finance or healthcare, then there are implications to it today. But they already have that with privacy laws. this is almost, think of it as almost an extension of privacy issues.

You know, how do we protect the data? How do we be transparent, accountability, all those things? So governance is a big part of the conversation. We’ve been trying to come up with a softer word so we don’t say governance. People get scared off when we say governance. So I started saying guidance. You need to give AI guidance. But at the end of the day, that’s just a nice way of saying you’ve got to, you have to know what’s going on with your data. So I think in any regulated environment, you’re going to find that they are interested in that. So one of the problems is,

There are over 700 bills that I’m aware of throughout the states at the state level around AI, privacy, various aspects of AI. And there may be more now actually, I’ve lost track. Now, I don’t remember how many passed last year or this year, but it’s dozens. so various states have different laws now around AI. And so if you work across, if you have a business that is doing business in all these different states,

you’re under different regulations right now. And it may be that way for quite some time. my personal hope was that they were going to come up with some unifying, know, a set of regulations that were pretty light touch, but clarified like, is your obligation as an organization? And they haven’t done that. And I don’t think they’re going to do that for some time. So you really do need to stay up on what’s going on in each state, which is, which is going to be a little bit of a overbearing thing for companies over time.

Shawn Cordner (12:25.166)

Is there a shortcut to that? Like is one particular state, you know,

have tighter regulations and if you comply with the regulations, you’re going to be complying with the others.

Yeah, I think that’s a great way to look at it is that they’re probably, know, California is definitely out there, although their big push got vetoed. So it didn’t actually go through last year. they could have been or would have been kind of pushing ahead in some areas further than other people. they’re still up there. Was it, I want to say Colorado, I think has the probably the biggest strictest one right now out there.

There’s a couple of other states that have pretty good laws. know Massachusetts is going to be, sure, I forget what’s going on in New York, but I’m sure they’re going to have some laws coming by if they don’t already. I forget where they’re at. So yes, I think that’s a great way to look at it is sort of figure out which one is kind of the most strict and do that. But I’ll tell you this, I don’t think any these laws are what you need to be worried about. I think what you need to do is just use common sense. If you have a technology that can

help your business accelerate and even interact with your customer, but can make mistakes. And it can make mistakes in some fundamental ways because of the nature of this technology. You have a certain level of obligation and accountability. And so you need to be able to own that. So that means transparency, explainability about how it made the decisions, monitoring, constant re-improvement and reinforcement on the learning of those AIs.

Scott Weiner (14:06.328)

products. I think that’s just the nature of what we’re talking about. And if you’re doing those things and you’re doing all the good practices around security, you’re probably fine.

Yeah. What’s the state of AI in enterprises? know, there’s the build versus buy. The conversations I’ve had with IT leaders, you know, through our events this past year, it sounds like a lot of them are still really early in their AI journey. A lot of them don’t really have a strategy thought out yet. And so the way that they’re wading into AI is they’re buying AI products, right? They’re

buying intelligent virtual agents for their contact center platforms. They’re using Microsoft Co-Pilot. That seems like an easy and safe way, but I bet that that is fraught with some risk that they’re potentially not thinking about. So, what do you see right now? Or is it still in the buy phase or just the largest enterprises building?

we move into 2025, it seems to be getting closer to 50-50. Also, there was a report that came out from Menlo Ventures in 2024, the state of generative AI in the enterprise, and they were highlighting significant surges in enterprise investment and generative AI. I don’t remember if they broke it down by build versus buy, so that’s a good point. But what I’m hearing from different reports and so on is that it’s coming out to almost 50-50. And the interesting thing is I look at it

When I go into an organization, look at it this way. You’re either at the stage where you’re saying, what is this AI thing I’ve heard of? And so they’re in the educational phase of just learning about AI and what it means for their business and how it could really impact their business. they’re in the early stages. Then there are the companies that say, I want to use AI like you just said, within productivity tools and other products that I use to just optimize what I’m already doing. So maybe I get like Microsoft Copilot and now I can build

Scott Weiner (16:02.734)

PowerPoint presentations faster, right? Something like that. They have a certain level of understanding and knowledge that they need to be effective. And one of the things we find at that level is, so I’m sorry, that’s like the slow, the low level. I’m finding even at that level, the problem they run into is they don’t invest enough in education. And so now they’ve given these people these powerful tools and they’re expensive, frankly, and they’re not really learning how to leverage and maximize the value of those tools. So that’s.

the first phase. Then you get into companies that say, we want to use AI to optimize our internal processes, increase our efficiency, maybe reduce our need to hire more people and have our people do more better work. And so they’re looking at AI from a custom perspective. How do we integrate it with our current systems and make our own processes better? And then you have the companies that are actually building products with AI.

And that can be the ones that integrate AI into something they already have. I call it sprinkling AI on your problem. Or the ones that are actually doing full native AI applications, new applications. And I think what you’re going to see is more more trending towards that over time. Everyone will kind of move down that path over time.

Even at the lowest level of adoption, I think there’s risks that might not be considered. For example, data leakage is a problem with your workforce using ChatGPT. Another example that, real life example somebody told me they experienced in their business was they enabled AI transcription on Zoom without thinking about it.

they would start up a Zoom call and they would have conversations before other parties entered into the Zoom bridge. And then that transcript got sent out to all of the participants, right? So talk a little bit about the risks even at that lowest level.

Scott Weiner (17:58.87)

Yeah, and I don’t want to pick on zoom, but in particular, but it is a horror story. hear a lot where people are like having a conversation and they forget there’s an AI listening. And then that AI has different ideas about what it’s going to do with that information because it was told to do it. It’s not like it’s it’s got its own mind. It was programmed to go off and maybe submit that conversation to everybody.

And we should say it’s not unique to Zoom either. A lot of platforms have the same feature, right?

every, every, yeah, any AI platform has this risk that you’re talking about. and, so what happens is at one level, you can think of AI is just another tool in the environment. You need to understand the data flow. You need to understand where the data is going. You need to have data policies around, around data, you know, acceptable use policies around the application and all of that. Those are basic fundamental things you need to have with any application. It’s just that with AI, it gets a little bit more tricky because it’s the first technology we’ve had that can kind of

make decisions for you or on its own without very specific rules. And that’s where it really varies from what I talked about with expert systems in the early days is that it’s not based on a set of rules that are very concrete and strict. And so the unexpected can happen. are ways around the walls we put in front of it. So in a lot of ways, it’s like any other tool you need to have an education and policy. So the difference here is that

you’re allowing a lot of access without asking a lot of the questions. I talked to one company, was the credit union, he told me that he, this was the CIO, I think, or COO, I don’t remember, but he said that he had about 120 vendors. And I asked him, I said, of those 120 vendors, how many do you think in some way are now using AI in providing you their services? And he said, probably most of them at this point. And then I said,

Scott Weiner (19:51.158)

And how many of them that are providing you products that you actually work with, your counting systems and ER, not ERP’s, counting systems and so on, financial modeling systems, how many of them are providing you AI tools that are kind of coming up to you that you’re actually engaging with and trying? He said, a lot of them, we see it. And I said, so not only do you not have policies yet in place for how you’re going to work with these things, how many of those vendors have you vetted? And what’s your vetting strategy? And the answer is we don’t have one.

Well, that’s where we come in. But that’s kind of the point is that, you know, they’re not thinking about what’s the implications of this because the things that come up with AI that are different are obviously, it’s harder to find out where your information went and how it’s being used within that AI system. It’s not like a database where I can go to a record and say, there’s my record with my information in it. That’s not how these large language models, for example, and we’re mixing AI and large language models together, but that’s not how they work.

they kind of create connections between information. so access to your information becomes much more difficult to identify. Other problem with large language models is they provide a level of bias and a risk in the bias area. So in other words, as it gets fed information, if it goes into the learning for the next iteration of the model, so the training of the model, it’s biasing the model. So if the data is good,

fine. If the data is not good or unbalanced, it’s going to start to make decisions based on that. And that can impact people. You one example is an HR system that starts rejecting people based on their race or sex because it has biased data that trained it on what to look for.

Right. Because historically, say males were getting hired for this one position more than women were. Right. So then the model learns this is a position for

Scott Weiner (21:48.014)

Exactly, exactly. And that’s just one small example of this. So it’s very important to understand how to work with data within AI. And it’s not just for the engineers. It’s actually for the business people as well. there are policies you need to have in place, training that should happen. One of the major things I recommend to every single organization and every person in the organization that’s going to touch anything to do with AI is they take some level of prompt engineering training.

This is simply, sounds like it’s a technical thing, prompt engineering. It’s really just a language skill. It’s the ability to formulate your request in a way that is going to make it more structured for the AI to understand the context of what you’re asking. If you say to an AI, what’s a nice color? There’s a universe of colors you can pick from and it’ll pick a color. That’s the context you’ve given it. But if you say, what’s a nice color for a flower? You’ve limited the context. What’s a nice color for a daisy?

you know, as you limit the context, you’re going to get more and more specific answers from it. And that’s really what it prompt engineering is, is creating specificity around what you expect out. And the more you can do that, the more value you get out of any AI system you use, any LLM type of AI system.

It almost sounds like really the first place you have to start is with data management. Then you have to give it the right data and the data has to be structured in a way that it can get to the answer that you’re trying to prompt. Is that fair?

Yeah, we can go through a more of how we when we go into a company, what readiness looks like for AI. We have a whole structure I can kind of walk you through. But one of the major pieces of it is the data. And what’s interesting is we’ll go into a company and say, OK, you want to do something with with data. What data do you have? And they’ll start to list off the data they have and they say it’s great data. But then when you look at it, you realize, well, it’s not really great for AI because for AI to work it.

Scott Weiner (23:45.622)

Like for instance, let’s say you had time series data and there was pieces of data missing. Well, you have to figure out how you’re to fill that in. Otherwise, there’s going to be assumptions that the AI makes about what happened in that time period. And that’s going to skew everything for predictive modeling and so on. And there are techniques for this, but that’s just one example. Another example is, how does the AI know if this is a good record or a bad record? You have to tell it.

Is this a good trend or a bad trend? You have to tell it. have to give it that information. There’s lots of different techniques for training that make the data useful for training, but you have to do that. so the other problem is it’s not a one-time issue. It’s an ongoing issue. One of my favorite examples recently this year was there was a massive change at the Supreme Court level where they had a law that

that basically set precedence for the last 40 years. And what happened was they overturned it. Well, think of all of the filings that were based on that precedence for 40 years. It was used in a lot of cases. So now you want to go ask a legal question of the LLM. And as you ask your model the question, it’s basing it off precedence from 40 years, which is no longer valid. So all of a sudden, all that training is invalid. And so what do you do?

Well, you have to wait till the next version of the model comes out. Because the problem with AI today is these large language models are, they’re fixed in time. They’re not open to new information in the way that you need for real-time information. So there are techniques to bring real-time information into systems, but you have to do that deliberately. And that’s a whole other thing. And that is also what you just said, data management.

One of the areas that I think enterprises see a lot of value from AI is in gleaning insights from their customer interactions, for example, or just business intelligence. It would seem to me that having real-time data would be critical to that being actually valuable information to get from the AI.

Scott Weiner (25:51.276)

Yeah, so there’s different techniques for getting real-time information in. And the areas that we see, the common use cases we’re seeing today are definitely predictive. Customer service is one. I’d say AI-driven chatbots, virtual assistants, handling different inquiries, enhancing customer engagement, reducing response times, that type of thing. That’s one area. Another one is definitely predictive maintenance. So the ability to predict when something will happen.

is really important for businesses. It’s great for supply chain optimization. And I will say from that’s more of our, where we see it a lot is in the supply chain, there’s lots of opportunities to enhance through AI. Fraud detection. So in the financial space, especially fraud detection is phenomenal. The things you can do, because the thing that AI is really good at is pattern recognition, pattern matching. So it sees patterns that we can’t see very easily. That’s one of the most interesting parts of AI is

give it a lot of data and let it try to sort through it and figure out where the patterns are rather than telling it this is what I’m looking for specifically. So cybersecurity, another area where AI is very valuable. Personalized marketing, that’s another great one. Think about the idea that when you come to my store or my online store, let’s say, and I know about you and I can put all kinds of correlate, all kinds of issues and figure out what’s the best way to not just what would you be interested in for my store.

but the best way to present it to you. And one of the things AI is really, really good at is influencing. This is both exciting and scary at the same time, but it’s incredibly powerful at influencing people. Great for product recommendations, things like that. But then you get into financial planning. AI algorithms analyze financial data really well, can help support budgeting and forecasting processes, strategic planning, anything like that. And the way to look at it is,

The AI represents your assistant, not your replacement for this. I would never, personally, I would never say I’m going to have the AI make the financial decisions for my business. I would absolutely say I’m going to have it go and look at the data and give me recommendations that I can consider. That’s the way I would use it today. And then another area that we see a lot of AI usage is human resources. Lots of tools coming up for AI-driven.

Scott Weiner (28:12.442)

assistance for talent acquisition, employee engagement, performance management, streamlining various HR processes. those are kind of the big areas we see it in. yes, absolutely being able to glean insights from data is where it shines.

Let’s talk about where the rubber meets the road when it comes to AI. So there’s an IT leader who was listening to this. They’re interested in getting value in one of the areas that you were just talking about. They need to have an AI strategy in place. So you kind of alluded to a framework that NuEon uses. Can you walk us through that framework?

Yeah, I’m trying to figure out, I’ll do it in a very high level detail and if you have questions we can drill into each one. So when we go into an organization, and again, you gotta understand, we’re looking, just to clarify this, we look at mostly mid-market type out companies. So these are not the biggest companies in the world, but they’re not startups either, they’re sort of in between. And they tend to have infrastructure in place, they tend to have,

products and revenue and so on. So when they look at AI, the first thing we want to make sure of is that they have strategic alignment with their business goals. In other words, it shouldn’t be, let’s do AI because it’s a new thing. It should be, we have this goal and we think, we believe AI will help us achieve that goal. If they don’t have that, then we start there. We don’t even want to go through the whole framework because until everyone’s aligned on what are we trying to accomplish, it’s not worth it, honestly. There’s too many things that go wrong.

So that’s where we start. Once we have that, it can go into different directions based on, like I said, those different categories of businesses. Are they in the, I’m learning what AI is, or am I in the, I’m building an AI application, depending on the gamut of problems we’re trying to solve. But essentially, we do look at their data. We start to look at the data. So we try to do sort of a feasibility on, you have a strategic plan, and do you have a vision for this thing you wanna do?

Scott Weiner (30:18.712)

For instance, have you have metrics for ROI? How are you going to measure success on this project? Then we go into data assessment and management. That was the part we were talking about before where they say, this is the data we have to work with. And from that data, we’re going to make a few decisions. One is, is the data applicable to the problem we’re trying to solve? Is the data complete enough for the problem we’re trying to solve? Maybe we have to go acquire more data from somewhere in some way or make data.

And then the third thing we look at is talent acquisition and skill development. How are they going to support this AI project? Because AI is different in a few ways. And so they may not have the skill set internally, or they may not have enough of what they need. Nine times out of 10, they don’t have the skills they need internally. And I will tell you, your first AI project, if you’re on your first one, I would highly recommend bringing in external resources. And there’s lots of great teams out there. We work with a lot of

that can do the work of helping you get your product vision up and going, version one, let’s call it, your pilot, whatever. Because you don’t want to be guessing at what’s a good practice in this area. You want someone with experience building your version one. Once it’s up and running, sure, let’s have a training program, let’s get your people up to speed, all of that, that’s fine. But I wouldn’t go into this with people who are figuring out as they go.

It’s it’s too. Here’s the reason AI is changing so fast. In fact, Sean, we were originally going to talk about three weeks ago. We’re going to have this conversation, right? I had an idea of what I wanted to talk to you about that. In the last three weeks, there have been so many innovations and releases that we might talk about today that didn’t even exist on the market three weeks ago. That’s how fast it’s going. mean, major things, not little tiny, you know.

products. So the idea that I would spend months and months trying to build something with a team that’s figuring it out, by the time they do, it’s probably a completely different thing. So you want people with experience helping you get it off the ground. Then go and train your people alongside them. Maybe they’re just part of the process. All right, so that’s number three. Number four is technology and tool selection. So do they have the right technology and tools? This is where, frankly, Nuveon is really good at helping companies.

Scott Weiner (32:42.446)

figure out the right tools and vendors to work with. And that’s why we love partnering with Amplix, actually, because you add a lot to that formula for us. And then ultimately, picking a pilot project and a prototype. How are we going to assess that this is going to be worth investing more in and getting to that point as fast as possible? So creating the smallest possible project that is high value and is highly feasible.

And so that’s one of the things we look at is, you have that in place? If not, that’s another piece. Now, once you have that pilot project in mind and you’re starting to formulate those data pieces and you have the people in place, the next thing is ethical considerations and compliance. How are you going to make sure that what you’re doing is both ethical and compliant? And compliance within your industry, but also just, like we said, just have general data principles because you don’t know what laws are coming.

And then from there, I’m really just giving you a top 10. From there, you probably go to change management and cultural transformation. What do you need to do within your organization to have your organization embrace AI? What type of training, what type of alignment do you need to do and so on? So there’s all kinds of components under change management. Identifying what partners you want to have and how you’re to collaborate with them on your project, because a lot of these are not in a vacuum. You’re usually leveraging other…

technology and tool sets and integrating with other products. Having cross-functional teams within your company are really important. And then as you start to go beyond that, now you’re into how are you going to scale and continuously improve? Now you’re into the real creating a pipeline for moving forward. And then finally, what are your performance measurements and ROI analysis? How are they doing in measuring them? That’s kind of the framework. And then we have an assessment that we do along seven dimensions, basically. There’s

strategic alignment, leadership and governance, which we talked about. There’s the organizational piece, so culture, talent and change management, data management, governance and infrastructure. And then you have AI development, training, updating and monitoring. And then you’ll have ethics, compliance, risk management, security and privacy. And then finally, performance measurement, ROI and stakeholder impact, just like I said, and ultimately…

Scott Weiner (35:01.09)

the external partnerships and collaborations. And that’s kind of the map.

That’s quite a lot to think through.

I know, I know. And that’s one of the challenges when we go into these organizations is that it is a lot and it’s overwhelming. And so what we try to do is not overwhelm them. We start with simply, do you have a goal? Can we get something up quick and and running for you? And I will tell you, if they’re open to it, the first five of the things I mentioned, we can knock off for them very quickly. We can get them to the point where they have a proof of concept.

and it’s running and they can just see if this is something that they want to invest more in really quickly and at low cost. And that’s really our goal because beyond that, it’s sort of then into how do you want to build and deploy.

Yeah, I think it really highlights the value of a partnership with a company like Nuveon and Amplix. appreciate the plug for Amplix earlier with the checks in the mail. But we do have a workshop that we collaborated on and we’re offering at a discounted rate to anyone that comes in through this podcast. So we’ll put a link to that in the show notes. I think something that

Scott Weiner (36:09.578)

that but so

you hit on was the ROI for this, right? We’re not doing AI just for the sake of doing AI. There has got to be some sort of return on investment in that effort. And to get to that return on investment, we need to understand the total cost of ownership of AI, right? So I think a lot of the IT leaders that I spoke to, they don’t have a full view of what the actual costs of AI are in terms of their resources, their time, as well as budget. So maybe you could speak to the costs.

Yeah, that’s a great point. there are a lot of costs. So if you think about it from the cost of AI, first of all, it depends on if you’re talking about proof of concept or you’re talking about full blown, we’re doing our own development initiative. So I would separate out the two because a POC shouldn’t be too bad.

I hear a lot of people talk about them in the 30 to 60K range is something that is, you can get something that covers your idea mapping and your business model and you can get your validation and your prototype development, all that. so, and we have programs like that to get proof of concepts out the door. But if you’re going to get into a full blown, we’re now going to be a full AI shop that has our own products and services that are all AI based. That’s the other end of the spectrum.

You’ve got to think about data management. So you have your collection acquisition, your cleaning and labeling your storage. So that’s your data, your data piece. when you develop development integration, you’re going to have to think about what kind of model you’re going to use and the software development process and all the tools that go in your typical software development and system integration. So those are typical. and then it gets interesting because where AI is different somewhat than traditional software development today is the hardware investments are higher. If you’re going to do your own.

Scott Weiner (38:02.314)

model and do your own training and why would you do your own model? The main reason is you want it to be specialized to your business or you have data in it that you just don’t want to take a chance on a you know an open AI or Microsoft storing your data for your Google or whoever. Those would be some of the reasons why we hear people want to do their own models but some of the performance may be a reason. However I doubt it because it’s really expensive to get high performance. You need to buy models that have

hardware that has very high GPUs, TPUs for model training and inference. So a lot of people go to cloud computing where they can get a cloud infrastructure. They’re not cheap, but they’re probably cheaper than you’re trying to host your own servers and all of that. And these are big machines. The other thing that people underestimate with AI cost is networking. It turns out that if you want to have these things be high performing,

They need really high capacity networks to get the information between the different pieces. So that’s something that I think is underestimated a lot in the cost when you’re costing this out. And then there’s the talent, right? Hiring specialists, training your employees. We talked about that. Those are real costs. It’s a new skill set that they get it trained on. It’s not exceptionally high.

But it’s definitely a cost that they underestimate a lot. started out by telling you, even with just, want to use Microsoft Co-Pilot, but we didn’t teach people how to do prompt engineering and now they’re not getting value out of it. We hear that a lot. The testing and validation of AI is another area that I think is underestimated in the cost because if you’re going to build, again, we’re separating out building your own from, I’m just going to use some third-party product that they’re doing it all behind the scenes for.

So if I’m building my own, have testing and validation issues. then maintenance and support for models is an ongoing thing. The one mistake I hear a lot is, we’re going to build this model and use it. No, you’re not. You’re going to build the model, and you’re going to constantly monitor it and train it and fix it and update it. So there’s an investment once you go into AI that’s ongoing that you just have to be ready for. The value will be there. You basically create your ROI or you don’t.

Scott Weiner (40:20.142)

But if it’s there, that’s what it’s going to take. then lastly, you get into regulatory compliance issues, but that’s typical for most businesses. You have to deal with your compliance issues. We talked about governance and then I would say security, data protection, safeguarding the information, obviously, but also threat mitigation. So typical security issues. We deal a lot with cyber threat issues. And one of the things about AI that’s a little unique is prompt injection is a new thing. So the idea that I could actually

attack your model with a prompt that tries to get around whatever safety pieces you put into place. That’s new. Being able to inject into a model is a new threat. Now, it’s not any different in terms of how you would approach it from organizationally, but you have to consider that. And then the final one, and this is a big one, is energy consumption. Sustainability within AI is a big conversation right now. The operational costs are high for managing AI.

I heard one quote, someone did some study, they said that generating one image with one of the major image generators took as much energy as charging your cell phone. So just to put that in perspective, sustainability is a big deal. There’s pluses and minuses to what’s going on in sustainability area, but I just wanted to make that point.

Scott, we’re getting towards the end here, lightning round. I want to get through some predictions. What do you think are going to be the trends in AI for 2025?

Yeah, so in 2024 the number one thing I predicted was agentic AI. And we are there and so you’re going to see throughout 2025 a lot of talk about AI agents, the AI agent revolution. So we started it this year. It’s going to continue into next year and the thing that you need to watch out for is. What people are calling agents is going to change overtime. So the sophistication of these agents and what they’re capable of doing. I’ll give you an example.

Scott Weiner (42:22.242)

Today, I would love to have an agent that can go out and plan my trip for me. I am not ready to have it push the buy button for me. I want the human in the loop. And so for most of these agent type of activities, I’m looking for human in the loop, meaning that it can go out and do a lot of research for me and figure out things and maybe interact with other agents. So that’s the other thing is agents talking to agents, but I still want to be in the loop. And so we’re not quite there yet. AI spending and adoption across the board is just going to keep going up.

I would say, what is it, like two thirds of companies are starting to embed AI into their core business operations the next year or two. You’re gonna start to see them just figuring out ways to use AI for things that matter. What’s happening right now over the next year, I think you’re gonna see this huge shift from experimentation to operation. It’s just starting to happen. Another area is multimodal AI integration. So you’re already seeing this, but the idea that AIs are no longer just language tools, but they’re able to see you, hear you, look at video.

sentiment analysis and pull all these things together to provide much more comprehensive solutions and responses to you. Workforce transformation. So one quote I heard was that 46 % of organizations will be scaling AI across their operations in the next year or so. But the increase in focus will be on upscaling employees, not replacing employees. It’s going to be on how do I get my employees to be

better because they have AI as a tool. And that’s why I was making this big deal about prompt engineering. So I do think that that’s an area that can really upskill everybody. You’re going to start to see more new hybrid human AI workflows happening. So routines that take AI and people into account, human in the loop. So emphasis on workers doing more high value work. I would say if you have a job that’s just redundant, repetitive, task oriented, those are the jobs I worry about.

The jobs where you have to think as part of the job, I’m not. I think you’re to have much more interesting work to do. There’s going to be business model innovations. We’re going to have new business models, new revenue streams we haven’t seen before. AI native products are going to start to pop up more. It’s already happening. And we’re going to redesign some of our existing services. It’s going to transform the customer experience over the next few years. Model advancements are going to continue to happen. Reasoning is the big thing right now. And that’s just going to get better and better in domain. So you’re to start to see more industry-specific

Scott Weiner (44:47.958)

models that are better than humans in certain areas. That’s gonna happen more and more. And more of a focus on AI security, as we talked about this, you’re gonna hear the term responsible AI more and more, and accountable AI as well. And the other big one that I think is interesting, and I don’t have a good prediction on if it’s in the next year or so, but is the quantum computing is starting to really emerge as a player in the conversation. So for those who don’t know,

Google just announced their willow chip, was a huge breakthrough in quantum computing. And if nothing else, it kind of shows that this is not a theoretical science anymore. It’s not at the point where we can just use it. But just to give you an example, just in case you don’t, just to give a kind of perspective on this, they took a math problem that would take billions and billions of years. The way they described it is longer than all of time, since the beginning of the universe, to solve.

and the will of ship solves it in about five minutes. So if we start to have that kind of capability in AI, the sky’s the limit. We’re not there, I don’t know if we’re, I don’t know if we’re five, 10 years away or if it ever happens, but that’s where the promise of quantum is. And so I guess what I’m really trying to say is we’re starting to see emerging of hardware and software, AI and hardware technologies that are both accelerating this conversation or what can AI do?

And then finally, the last two things I’ll say on predictions are large world models.

faster training for models by having these large world models. So what they’re doing now is they’re simulating environments and then allowing, well, that was weird. You popped over. So with large world models, they’re simulating the different environments and then they’re trained AI on that simulated environment. Imagine, for instance, you were trying to train a self-driving car and the way you would do it is you would have it, you know,

Scott Weiner (46:49.356)

either drive the car and make mistakes or watch video of cars driving and making mistakes and would learn from those mistakes. That’s how it learns. But you can’t simulate everything. Like I can’t simulate a child jumping out on the road and running them over. How do I train that? Well, with a world simulator, you can do that. And not only can you do that, you can do it millions of times faster than real world scenarios. So our ability to train models is going to get faster and faster with these large models, which are only possible because the hardware is catching up. And the last thing I’ll say is robotics.

Robotics robotics merged with AI is absolutely going to change and it’s already changing manufacturing But also you’re to start to see it in consumer consumer environments, too I mean I’ve been seeing these these robots that can make your dinner for you and clean your house that are like right out right out of the cartoons so There and the thing that’s that’s interesting about them is not their capabilities, which are amazing It’s their cost the costs are coming down to the you you’re talking a few thousand dollars

for something like this now. You’re not talking about millions of dollars. So it’s getting more more interesting about robotics.

Scott Weiner, thank you so much for all of your knowledge and for sharing it with the audience here. Last question, when exactly is AI going to take over the world?

We don’t know. If you ask any five experts, you will get four or five different answers. You’ve got people saying we’re two years away from what’s called artificial general intelligence, and you’ve got people saying we’re 20 years away. And I think that we’re probably somewhere in between. And here’s the thing, once we hit artificial general intelligence, which is an AI that can kind of work in areas that it’s never experienced before, do things that’s never been trained how to do.

Scott Weiner (48:39.47)

sort like a human can do. The leap from what’s called AGI to what’s called super intelligence is not long because what was really going to happen at some point is AI is going to start training AI. And so at that point, it will escalate very quickly. So I don’t know. Stay tuned. We’ll see.

I’ll keep my fingers crossed that it’s not anytime soon. Thanks, Scott. Yeah. All right, you too.

Yeah, I don’t think it is. Appreciate it.