Artificial Intelligence and Problem Solving

Jay Bradford: What is going on? It’s Jay and Michael here on the IT Directors Podcast with our main guy, James Hammack, Director of development here at Clear Winds. We’ve got an awesome topic for all of our listeners and viewers: how AI really works. James, I’m pumped about this episode. 

James Hammack: Does anyone really know?

Michael Thomas: I mean, hey, we want to dig into your brain and see what you know. See if Claude’s in there himself.

Jay Bradford: Oh, deeply embedded. I was going to ask AI how it really worked before the episode, but I thought we would talk about that during the show, you know what I mean? So, James, just tell us a little bit about how you came into this role at Clear Winds, kind of your background, and specifically around AI. What’s your knowledge and working experience with it?

James Hammack: You know, as the director of an entire department that does development, you’d think that would be the main place that I actually started. It was actually my personal life — just kind of jumping into utilizing AI in some other areas of life, getting some tasks done on the side, and then realizing, hey, this thing is growing quite a bit, and then wanting to look forward into how to actually implement it. And so it was all these other tasks that I had going on outside of work, and then moving into work, I realized this really needs to be in this place. And so there was a moment where it all very much clicked — we need to be leaning into this, not just probing a little bit. So that was kind of the big thing for me.

Jay Bradford: Awesome. Man, I’ll tell you what — I remember the first time I used AI, which was ChatGPT, right? Because that was kind of the mainstream one a couple of years ago. You know, there are a lot of different tools — we have Claude, we have Gemini, a bunch of things. But I can remember the first time using AI. I was trying to plan a road trip and do some sightseeing because I love to hike and the outdoors. So I said, “Hey, we’re planning a road trip to X. Please give me some sights,” and I was amazed at what it did. I was like, “This is incredible.” So it gave me all these great suggestions, and that’s what kind of started my thinking around AI and how awesome it can be for just the everyday user and person. And you know, I think we use it a lot more than we think, James, in everyday life.

James Hammack: That’s very true. Yep. I definitely want to talk about that and why that’s a reality for us.

Michael Thomas: Yeah, and even learning more and more ways we can use it — because I myself started off with ChatGPT, and one of the first things that blew me away was this: we bought this alarm clock for my kids, one of those little light alarm clocks that tells them when they can come out — turns yellow, red, green, whatever. And I lost the instructions, and it was a headache, so I literally just took a picture of it and said, “Hey, what is this and how do I set it up?” And it gave me detailed instructions and found a video.

Jay Bradford: This is a game changer. And so we’re going to transition into the technology world because we’re trying to help IT directors, we’re trying to help organizations. James, what is AI, and how is it different from regular software?

James Hammack: Yeah, that’s an excellent question. When we think about AI, there’s this kind of big-brain picture we have out there somewhere, and that’s actually not too far from the truth. We really want to think about it more along the lines of a probability engine. What is AI in this sense? It’s a group of connected things that are then grouped into further connections, into further connections, and you have different sources that it can pull from and group all these things together. And so it looks at whatever you’re asking it and goes to where it thinks the most likely answer is. It has gathered a whole bunch of information together — think billions of examples — and then it figures out the pattern itself for what you’re looking for. We do this with our own brains, right? You say, “Humpty Dumpty sat on a…”

Jay Bradford: Log

James Hammack: A log? Is that what you go for first? That is an example of a bad AI search.

Jay Bradford: I know — that’s a hallucination!

James Hammack: That’s a hallucination. That’s right. 

Jay Bradford: I was trying to see if you’d catch that, James. 

James Hammack: Well, hopefully all of our listeners did way better and said “wall.” But that’s a big thing. When we do that, we do something very similar — we’ve heard that phrase before, we kind of know what to anticipate, and so we go and find it ourselves. AI is trained in similar ways — billions of examples — trying to figure out what something is. And because of the way it works, it’s not a rule-based thing where if this happens, then do this. It goes and searches and finds. There’s nothing there as far as a rule to follow. There are guidelines to follow, and then there’s data to search. But outside of that, it’s really — hey, no one told AI what a cat looks like. It went and figured it out based on billions of examples. Which is actually how kids learn. It’s the same thing. We repeat “orange” over and over again to a kid, and then they go, “Oh, orange.”

Michael Thomas: So we’re talking about a word search. How does this differ from a Google search today?

James Hammack: Yeah. So when you do a Google search, there is actually some AI running in the background as well. One of the ways it does that right off the bat is you start typing and there’s an auto-complete. Auto-complete is a really, really early base-level form of artificial intelligence. It’s kind of doing the same thing — predicting what you’re about to say, and trying to save you some time. You see that and just go straight to the next word. It’s really quick and easy and nice. And then what a Google search does is it just gives you as close to what you said as possible. It’s not trying to predict what your end goal is. It’s not trying to come up with something itself in any way. It’s really just saying, “This is the data, and you asked for it, so I’m going to give you as close to the data as possible.” And so there is a difference between what AI is doing in the background and what it’s coming up with — and even what it’s generating, which I know we’re going to talk about as well.

Michael Thomas: So it’s not some big brain in the back of some room. Like, what does that actually look like in practice? What’s actually happening?

James Hammack: Yeah, so the codebase itself is quite large, and it’s growing because it’s taking in everything it’s learning and compiling it all together. It’s not a hard drive source — that’s what a lot of us are used to, right? We have a hard drive, and that contains the data, and therefore that must be what AI is. But no — AI is really processing power that goes and finds data off of hard drives out in the cloud and all this sort of stuff, but it’s processing it. And so that’s why whenever you talk about an AI machine, if you’re wanting to build your own, you don’t really refer to the hard drive very much. You refer to the computing power, because it’s using that computing power to actually go and do the processing of the data, not just simply storing data to recall later.

Jay Bradford: Thus why memory and CPU cost is through the roof — because all these huge AI data centers are being built. You know, it requires so much processing power. I mean, think about it — we use Claude as an organization. Clear Winds, you know, James, our director of development, you have steered us toward Claude as our pathway.

Michael Thomas: Yeah.

Jay Bradford: It is so incredible the coding and the different scripting and formatting that Claude will generate, but it’s taking tons of computing power in the background to do that. So we’re one organization, in one state, in one part of the world. Think about all the organizations using Claude globally that are hitting those data centers — and the computing power required. It’s impressive when you think about it.

James Hammack: Mm-hmm. It reminds me of going all the way back to the early days of NASA. They had an entire room filled with equipment, and it needed people to run so much of it and to do checking and calculations. And now everything in that room can be done with an iPad and AI. Which is incredible — it’s totally different now. But part of that is because instead of everything being processed in-house in one location, it is being processed in so many different places. So we have AI that requires so much, and it’s not so much a brain as it is computing power, which needs a lot of space. And so these big data centers pop up.

Michael Thomas: Yeah, and I think another big thing — we kind of talked about it and were joking earlier about prompts — is it more like you’ve got your programs and you’re putting in information and prompts, or is it like a new employee you’re training? I know it might be controversial to bring in the idea of, “hey, is it like an employee?” But how would you answer that?

James Hammack: Yeah, when you’re talking about AI — is it kind of like a new employee you’re training? That comparison is actually very close in some senses. AI is not something where you simply give it training and then let it loose to do something on its own. There is a learning process specific to the user, and that’s where you’ll hear things like LLM — language learning model. What it’s doing is compiling information into an entire model of how it is typically used. And so when you’re talking about a new employee, you’re essentially taking them through the process of, “These are the way our SOPs work in various areas, you need to understand that, you need to follow that procedure, and then move on this way.” And then after that, there is some computing — you go and do the tasks given to you, learn new things, that sort of stuff. So in a way it can be thought of like that, but with a lot of guardrails around it.

Jay Bradford: And that’s a great point you bring up, Michael, because I use multiple AI tools in my everyday life as a sales engineer, and also in my personal life. What I’ve found is that certain AI LLMs will learn your prompting — and I hate to say “learn you,” right, because that does make it sound like some big machine in the background. But it learns our input over time. So ChatGPT, OpenAI — it learns based on all your chats, and it will say, “Hey, Jay, this lines up right with your” whatever — your philosophy, your motivations — and it’ll start to learn you. Whereas Claude is more enterprise-driven, it seems, where the chats are more individual and not tied together. I know there are some settings you can use to tie that together. But there are tools that do learn at the surface level — they learn your inputs — and in return that makes you think they’ve kind of learned your behavior, but it’s actually just your thoughts, inputs, and questions, right?

James Hammack: That’s exactly right. It’s a calculation. It’s predicting based on past experience and things you’ve already done. What it’s doing is recognizing a pattern. “Hey, they typically come and ask these things, but I’ve seen a pattern — when they start asking this question, they actually end up here. So because I’ve seen that pattern from this user in the past, I’m going to predict they’re probably going there again.”

Michael Thomas: Would you say it’s only as good as the data it’s trained on? Or how is it limited?

James Hammack: Yeah, AI has definite limitations. Some of those are security-related. Some are related to the data that is given to it. Obviously, if you don’t give something the right data, it’s not going to be able to produce anything useful on the other side. So data is a huge part of this — it’s a big deal. We would want to use a system that can have high memory and a high context window. Context is how much it can know, learn, and keep in quick-access memory within a chat. You have different context windows across different AI platforms, and the ones with the longest context can go back within the same chat  and see the history — what does this look like? — then predict, “Yeah, we’re going this direction, so let’s go ahead and go there.” And so you are training it in that way. You can give it plenty of documentation, you can give it templates and say, “Okay, this is what I want you to do with that template,” and it will move into that. And as with anything — and this applies to us as humans too — if you give a human a bad set of data, they’re not going to know what they need to know in order to accomplish the task.

Jay Bradford: Yeah, and that kind of segues into the question Michael was going to ask next, because you’re already touching on it. But Michael, go ahead.

Michael Thomas: I’m still kind of sitting with that. Like, the aspect of — if we’re using these tools and incorporating them into our everyday life, work, home — how do we make sure we’re using them efficiently and not limiting them by having questions and prompts that don’t allow us to get the best end results?

James Hammack: Yeah. AI is goal-oriented, and so when you’re thinking through a prompt, you want to think through: “Hey, I want you to look in this particular area of knowledge, and you’re going to be acting as something specific. Then I’m going to give you all the input you need in order to follow through and have the context of what I’m asking you to do, and then I’m going to give you a mission” — which is that end goal. You can even go so far as to give it examples. This is the kind of thing I want you to do with that. Partial examples — it doesn’t have to be some long example, just partial examples on that, and then even some reasoning. “This is how I want you to think about these things.” That’s called the AIMER framework, by the way — what I just walked through there. If you do that and walk through a prompt, yes, it’s longer than going, “Hey, I want you to do this one thing,” and then spending 15 minutes of your own time going back and forth before it gets you the end result.

Michael Thomas: Mm-hmm.

James Hammack: If you spend two to five minutes doing a really quality prompt, it knows a lot more about what it needs to do in the first run, and it avoids as many hallucinations and other issues like that.

Michael Thomas: And it makes for a much quicker and better quality output right away.

James Hammack: That’s right.

Michael Thomas: Well, and if we’re talking about hallucinations, what do you say to the people — and those managers — who are like, “Hey, can I really trust it? Is this something that’s actually beneficial? Can I really trust what it’s telling me? How valid is this, and should I continue to go down this path?”

James Hammack: So the short answer is you should always trust but verify — that’s probably the best way to put it. If you are asking it to find expert-level information, it will actually go to that data set and find from that expert level. But if you have a very generalized prompt and you’re not pointing it toward the expert level of that particular knowledge base — whatever that thing is — so think, “I need it to run an executive assistant-level task.” Well, do you want it to run as the executive assistant who gets paid minimum wage, or the one who has been doing it so long that they are much higher up in their role and far more experienced? When you start to point it in that direction more specifically, it’s going to go that direction when it’s trying to find answers and accomplish the task and end result you’re giving it. So it’s a very big deal in that way. And so again, it comes back to prompting, and then it moves into trusting but verifying everything. Give it a good quality check. You can even do things like, “Hey, give me your sources. I want to see your citations. Give them to me plainly. Let me see some links.”

Michael Thomas: Mm-hmm.

James Hammack: Math is actually one of the biggest things I would say to question. AI is still struggling with math, because it will hallucinate and come up with answers. It probably pulled it from some statistic somewhere. The problem is, if it’s not a quality statistic — or maybe you’re doing something more scientific, and it went to a study that had lots of issues, so the output is already bad and the scientific community is laughing at it — AI doesn’t know enough of the context yet and still pulls it. And then you get a number that’s wildly off. You should check. You can tell it, “Show me your math.”

Michael Thomas: Yeah. I’ve even looked at some things and asked it to verify, and you’ll get a blatant response that says, “Hey, you caught me.” But it is a tool, and you have to treat it like a tool and use it. Have you ever been burned by it yourself or seen others burned by it?

James Hammack: Yes. I’ve been burned by it a couple of times. The biggest points were when I did not verify. So — whose fault is that? That’s definitely mine, right? So owning up to where you’re doing things — maybe you feel the rush and the pressure, and you say, “Hey, I can get to this faster than doing my own research,” and then you get it in there and it puts something back, you skim over it and go, “That’s generally acceptable, I’m going to go with this.” And then you get in there and you start to present it, or you look at it a little bit further and you say, “Hmm, that wasn’t good.”

Jay Bradford: Yeah, that wasn’t good. So, look — and so the math thing, Michael, right? Basic math, formulas, addition, subtraction — all these things AI seems pretty confident with. Where it gets tricky is when you’re doing research-type work. Stats, analysis, taxes, tax codes — all these different things, because there are so many variables and it’s pulling data from everywhere. So here at Clear Winds, we developed a tool to help us quote something, right? And I had to check that math numerous times. I found things where it was trying to add different packages or offerings on its own because it just assumed, “Hey, if a customer has X, we think they need this.” And so I had to be really strategic about prompting — “Don’t do this. Do this. You are a quote tool. This is your job. Here are your instructions.” So James and I worked together on a lot of these things, and I learned a lot about prompting, Michael. AI is only as good as the person using the tool. It’s just like a hammer, right? All three of us know how to do something with a hammer.

Michael Thomas: Mm-hmm.

Jay Bradford: We’d all do it differently because of our individual skill sets using a hammer, and it’s the same thing with AI. I tell people all the time, “The AI is only as good as the person using it.” And then you’ve got to verify the sources. You can’t just take it at face value. Like my son — he’s in college — he said, “Oh man, I use ChatGPT for this and that.” I said, “That’s great, but don’t you think your professor knows?” They can tell when they’re vetting things on ChatGPT. People use it a lot for research because it’s out there. I mean, it’s not going away, right? So students all the way from elementary and middle school up through college use it a lot, but you can’t use it to write a paper because it’s just pulling something from the internet that’s already written. You’ve got to verify these things. So — this goes right into our next question, which I love from an engineering background — explain generative AI and agentic AI and the differences between the two, and then LAMs and what all of this means.

James Hammack: Yeah, absolutely. So that’s the biggest thing people run into — they hear all these terms and they say, “What in the world? I thought it was just AI, but now I’m learning it’s this term, it’s this term, it’s this thing, it’s this thing.” You can kind of think about it as layers. From the 30,000-foot view, yeah, it’s AI — that covers everything, right? You start to drill in deeper and get a lot closer, and that’s where you get into machine learning, which is the dominant approach today. Machine learning is that idea of aggregating a whole bunch of data sets that have a higher probability of not only connecting to one another but being sought out, and it forms into a large language learning model — so the LLMs.

Michael Thomas: Mm-hmm.

James Hammack: And then you get into specific types of AI as well. You have things that deal with images, you have stuff that’s text-based, you have generative AI, and you have agentic AI. The image processing, the text processing — these are self-explanatory. Generative AI is something where you give it a goal and say, “Hey, create this thing,” and it actually generates something based on what you said, something fairly unique to some degree. We were talking earlier — Nana Banana uses Google’s Gemini for image processing and generative AI. I’ve done one with ChatGPT — one of my best friend’s kids wanted to create these images, and they were going crazy with it. It was something like a unicorn with a flaming mane or something like that.

Michael Thomas: And that’s how it’s supposed to be — let’s be honest.

James Hammack: Exactly right. And I was — this was when it was just really coming out strong — I was so impressed with its output. It gave this fantasy-like illustration of this white, long-maned unicorn with fire coming from the mane, just this majestic creature kind of look and feel. And it was really cool to see it do that. But I didn’t give it parameters — I didn’t tell it the unicorn needed to be white. It just naturally assumed unicorns are white, because why? Because so much of what’s out there depicts white unicorns. And so it assumed that by going in and saying, “This is the goal — I’m going to hit the goal of what was said.” Now, if I wanted to be more specific, I could have said, “It needs to be a blue unicorn,” and it would have output a blue unicorn because I gave it a specific direction. But that’s generative AI. Agentic AI — this is the frontier. This is where we are moving into. It’s here.

Michael Thomas: Mm-hmm.

James Hammack: It’s not widely used yet. It’s becoming more widely used, especially on the development side of things. This is where you are able to give something a goal, and instead of just answering your questions or completing a task within the parameters given — like generative AI does — agentic AI is going to take action. So it’s actually going to move through. I can say, “I want you to create a WordPress plugin, and it needs to be able to do these five things.” And then it will go out, search for its own answers, reason through the process of how to code, figure out the best way to code given what you’re looking for — and it does all of this without me having to put any additional parameters in at all. So it’s constantly taking action. It can also ask you questions — a lot of these tools will go through and say, “Hey, I need to change this sensitive file in the project folder. Will you let me do that?” And it’s asking for permission, and you can approve or deny. I’ve had AI do both — where it’s like, “I need to change this,” and I’m thinking, “No, you don’t, and I don’t want you to — that’s a huge security issue — so no, you can’t do that.” And then other times when it’s fine: “Of course you can do that, I actually want you to,” and I just allow and approve it and move on.

Jay Bradford: It has to have that human touch to all these things. I think agentic AI, James and Michael, is where we’re heading. For sure. I think the next one to three years is where we need to be in that space as a technology company — and as a consulting and MSP firm — to help our customers with this. Because agentic AI is where AI can be used so widely. And we were talking about it pre-recording today — there’s this idea out there that AI is going to replace humans, replace jobs. But AI will replace certain jobs only at the hands of people who know how to use the tools to do it, right? And so, like agentic — for an MSP, there are agentic tech bots that will change passwords on their own. We don’t need a help desk person for that anymore. “Hey, James, this is Jay with so-and-so. You need your password changed?” “Yeah.” There’s a chatbot that will do that agentically — in the scripting and you know. So there are great tools that have the security framework in place to do these things. I think we have to learn and open our minds — and open people’s minds — to accept the agentic, because it’s only going to grow.

Michael Thomas: Yeah, it definitely feels like there’s so much untapped potential because I think we’re on the edge of it. It’s been out there for a little bit, but we’re learning more and more what the implications of this tool are. It’s been created, it’s out there, and we’re finding more and more what it can and can’t do. What would you say — because I know our audience wants to grow in this field — what would you say to those who are on the cutting edge of it and want to get more experience with it? They want to learn how to test it, how to think differently. Because I know we’ve talked a little about it — there’s a lot of distrust out there because some jobs have been displaced, but a lot of that displacement is coming from those who know how to use AI. That phrase that’s out there is: it’s not necessarily AI replacing you, it’s people who know how to use it. So with that idea — because we’re here, it’s not going anywhere — now that we’re here, what would you say to those who are in a position where they need to learn and adapt? How do they need to think differently, and what can they do to help themselves?

James Hammack: Yeah. So I think when it comes to knowing where you’re at and then thinking, “Okay, I need to jump on this — when, how, and what am I going to do?” — the first thing is you’re probably already paying for a tool that has AI incorporated.

Michael Thomas: Correct. Mm-hmm.

James Hammack: Yeah, and so really do an audit. Just step back, look at your systems, see what’s actually available in something you’re already paying for. Especially from an IT director standpoint, you’re most likely paying for a product that has AI built into it.

Michael Thomas: Mm-hmm.

James Hammack: And so you need to know what that is, what its purpose is, what it’s good at, what it’s not good at — and then make a nice long list of those things so you can get your overall picture. And then ask: do I need to purchase something that fills the gaps? That’s the next step. It’s really, really important to do that. If you don’t do that AI audit, you’re going to go out and purchase things you probably don’t even need, wasting money and resources when you could be leaning into something you already have, and maybe even putting some money and resources into that because it already answers the problem you have.

Jay Bradford: You know, James, you brought up something right there about auditing your [00:29:00] tools, Mike. I was thinking about VMware — which is now Broadcom — the number one hypervisor for the last 20 years. In VMware, there’s a tool called vMotion, and I think it came out in vSphere 4 — a long time ago. It takes machines that are running and moves them to different hosts live, as resources are available or not available. And I thought about that as we were preparing for this episode. Man, that was AI 15 years ago in the code, because it was doing predictive learning and agentic moving of things. I didn’t have to manually go move the server, shut it down, and put it on another host. It was doing it on its own. And I thought, man, these tools have had AI in them for a long time. We weren’t using it as a consumer model — it was more enterprise-specific. But it’s been in there, and so evaluating the tools you have as customers, as an enterprise, and saying, “Hey, we already have some of these” or “we’re using this, we’re using that” — that’s a good point. Really good point.

Michael Thomas: Yeah, and even in the IT world, there’s a network manufacturer that incorporated this into their product a couple of years ago, and they took off. I don’t even think everyone actually knew what made it stand out the way it did — how it could resolve some of its own networking issues. I don’t think everybody understood that at the time. Now we see a little bit more, and now there are other manufacturers doing that as well. But it is interesting — when you take that look at, hey, what’s actually happening within this tool? What is here? Um, but I know one of our big topics is also security.

Jay Bradford: Oh, yeah. Absolutely.

Michael Thomas: So, Jay, I know you’ve got the security expertise since you’ve been in that seat for a long time. Hey, look — that’s a vast topic. But James, in terms of security — and this is a really wide topic — what are some guardrails and measures that organizations, IT directors, and leaders in the industry can take to protect their organization around using AI tools? Because look, there’s sensitive data we manage in IT. From an ERP system, you’re talking about Social Security numbers, payroll data, demographic data. We don’t want that going out into the cloud where other people can use it and access Michael’s information, yours, and mine. So what are some guardrails that organization leaders can put in place around AI from a security standpoint?

James Hammack: Yeah. Some of the basic guardrails involve using the enterprise level of whatever you’re going with — that’s normally what you want. There are many paid versions that are maybe not enterprise level but are business level, and they have a lot of the same protections. The biggest thing you want to determine is not so much should I use AI, but rather, when I come to this: where’s my data going? That’s a big one. Is it being used for testing? And then, who is legally liable if something goes wrong? So these are three big questions you want to ask about any product you’re going to incorporate with AI. When you’re talking about consumer AI tools versus enterprise-level tools, they are built very differently. At the consumer level — especially on the free level — you’re talking about a tool that is actively learning from what you’re doing, and it’s being used to develop the entire framework, which is great. That learning does in fact help the enterprise level of that particular tool, whether it be Claude or ChatGPT or whatever. But what it also means is that any information you put in is accessible across the board. If you have anything that’s proprietary, anything that doesn’t need to be shared — if you’re giving it SOP documents, if you’re giving it those kinds of things — it is ultimately out there. So that’s a huge security risk. You can fix that by having enterprise-level AI, and that’s all good. But the bigger risk for most organizations is really this idea of shadow IT.

Michael Thomas: Mm-hmm.

James Hammack: It’s people who are going to use the tool without asking. Then you have free tools being used and your proprietary information is out there. There are people who will put passwords on these things, put API keys on these things — and from a data security standpoint, [00:34:00] those are just huge no-nos.

Jay Bradford: Well, because what happens, James — and for the listeners and viewers, Michael — let’s say someone who’s more novice puts an API key into personal ChatGPT for an enterprise-level organization we’re doing business for, trying to connect two data sources. Well, guess what? Michael has access to that too. He can go in and chat, “Hey, I’m trying to connect — oh, whoa, we found this one because Jay put it out there,” and now we have a key. And now that’s how so much data gets exposed. But what’s interesting — because I know in my line of work there are some really proprietary softwares designed only for government and municipality use — I tested ChatGPT the other day and asked it a question. I said, “Hey, I need some information about some database tables in Software X. Do you have any information?” And it started and said, “Oh sure, you’re talking about Software X. Here are the tables. Here are some SQL scripts.” So that means someone put that database information out there in the cloud for everyone to see. I was blown away because I thought, “What a security issue — it’s no longer proprietary.” But that’s why we need guardrails, because these tools are here, people in organizations are using them, and we can’t stop that because it’s only growing. But we need to put some guardrails in place. The more you put out there in your personal AI, the more it compounds. That’s why it’s a language learning model — it’s learning from what’s put out there in the cloud. And so I found some very specific, truly proprietary stuff showing up. Yeah, and I know that’s why

Michael Thomas: some people are skeptical about using it. Correct. Yes. They’re skeptical because — hey, what are the implications? One, if your team is using it, you could potentially expose things that don’t need to be exposed. And I know we’ve talked a little bit about this, but understanding that skepticism — what would the cost be if you stayed on the sidelines and chose to just avoid it altogether?

James Hammack: Yeah. So you’re talking about something that is already here.

Michael Thomas: Mm-hmm.

James Hammack: And if it’s already here, it is being used, it is going to be implemented in your particular market by your competitors. So the question at this point shouldn’t really be: are we going to use AI?

Michael Thomas: Mm-hmm.

James Hammack: It’s: are we going to manage the AI that our team is already using? And that’s really the question that should be asked. And so if we focus on that and then say, “Yes, I am skeptical. I do have concerns. I need to understand some things first” — sure, I would do that very quickly though. There are a lot of resources out there that walk through and answer all your questions. The terms of service and terms of use for a lot of these big, well-known tools — like Claude, like Gemini, like ChatGPT — they have terms of service that show you everything you need to know. Review those things. If you need a legal team to review them, do that. But do it today, not tomorrow or the next day, because you really need to lean into this. The reality is it’s already embedded into the things you’re using. Think about it — if you’re using Microsoft 365, Copilot is already there. If you’re talking about your CRM, at this point, if you’re talking about any CRM, it probably has AI in it unless you’re talking about one that is very old and doesn’t update. It’s there. It’s not really a question of should we use AI. It’s how are we going to use it.

Jay Bradford: Well, and to Michael’s point — how these tools develop, how we’re going to use them. You know, 40 years ago, if you went in for surgery, there was a high probability the surgeon was smoking a cigarette while operating on you in the operating room.

Michael Thomas: Yeah.

Jay Bradford: That doesn’t happen anymore, right? Now there are guardrails in place. They understand smoking causes different things. And it’s the same thing with AI. When it came out originally to the public, everyone was like, “Whoa, hold on — everything I’m putting out here is going to the cloud?” “Okay, if I’m using this for business, we might want to pull back.” We might need to have a data governance committee around AI. We might need to have some MOUs signed. We might need to look at our legal team. Because as an organization, we are protecting other people’s data. So we can’t have that going out.

Michael Thomas: And I mean, our conversation is shifting — really, I think you can see it — from should I be using this, to how can I use this tool securely so that it actually benefits my organization? It’s here. So how do I use this well? How do I be a good steward of this tool to get the benefits from it? Um, so as we take that step further — I can think of one real-life example. Hey, you brought up healthcare. I know someone who just had surgery. Previously, they would have had to be cut wide open — it was invasive — but because of advances now, they literally had two minor incisions and a robotic tool go in and remove part of a kidney. The healing time was dramatically reduced.

James Hammack: And sometimes the surgeon is in another state or country.

Michael Thomas: Yeah, which is crazy. But talking about real-life examples of AI — what are some that you can bring up to help us continue to move that idea forward?

James Hammack: Help desk is the one that just jumps out right at the beginning. You’re talking about all these tickets that come in. You have people working on the weekend, amassing tickets that are going to the help desk. So Monday morning hits and you go into the help desk and see all these tickets. There are AI tools that can help categorize the tickets, get them into a good place, even draft an initial response for your review — and then you click send if it looks good. So Monday mornings are full of tickets, but the work that has already been done by AI is organized. You feel so much better just seeing that queue. And so you’re talking about getting things done in a very quick way. This is productivity math, right? This isn’t saying we need to replace people with AI. This is saying if we use AI well, we’re actually giving our people the tools they need, and those tools are super helpful in pushing things along so our people are able to do their jobs better.

Jay Bradford: Oh, 100%. And quicker. I mean, you talked about the help desk, right? I used to manage our help desk. We had 23,000 devices, 4,000 users, so as you can imagine, we had hundreds of tickets every day — not every other day, hundreds of tickets daily. So we had AI working in our help desk system. Michael puts in a ticket — “Hey, thank you so much for reaching out to X. We got your ticket. We’re working on it, and an agent will be with you shortly.” You know, it gives a human touch to the customer, right? Takes that off a human having to respond to every ticket. But a human looks at the tickets in the background. That help desk technician is now organizing and categorizing — what’s tier one, tier two, tier three. But the initial response has already happened for the customer. Which is huge, because the customer knows, “Hey, they got my ticket.” It didn’t just go out into la-la land. That’s a great, real-world example when you talk about that, James.

James Hammack: Yeah, I think another one is documentation. You know, there are people — and I don’t want to hate on anybody — I actually like reading a bunch of documents. I’m the weirdo in the room a lot of times. I don’t mind documentation. Thus, you’re a development director. Yeah, there you go. So I read it and I don’t mind it — hundreds of pages, whatever. But most people are, “Please don’t make me write any documentation. This is the worst thing ever.” AI can do it really quickly, especially if you have a really strong prompt and you move through it. You give it the resources it needs. It develops the documentation, and again — we’re still talking about a tool that people are using, so you should go back and verify. Go through and make sure it’s following all the processes that ought to be in the documentation. And then overall your product is done, and you’ve got documentation that’s really solid. What would have taken you four to five hours has now taken you an hour.

Jay Bradford: Yeah. And that freed up time where you can do other tasks.

James Hammack: Oh yeah.

Jay Bradford: Well, it frees up time. Michael is our sales director and marketing director. There are tasks that AI can do for him that he’s learning, and that frees up his time to be creative and visionary and to do the things that we’re here to do to make a much larger difference — rather than sending out spreadsheets with tasks. AI can help us do those, which is why we have to learn these tools. Like you said, James, we have to. So in closing, what is one takeaway that IT directors all across the world — who are listening to this podcast and viewing it — should leave with about AI? What is one thing or one concept we should leave our listeners and viewers with?

James Hammack: Yeah. I think we still have so many people who view AI as a threat, and moving from that mentality — from “this is a threat” to “this is a tool, and it’s a very powerful one” — is key. And it works best when smart people are in the loop. That’s called human in the loop.

Jay Bradford: Yeah.

James Hammack: When smart people are in the loop, we can take a problem, start with it, fix it well, and then build from there. And so it becomes a progress machine and a progress tool, rather than something that sucks up our time. So using it well and realizing it’s not a threat — but that it actually empowers our people to do their jobs better — I think that’s the biggest thing.

Jay Bradford: Man, that’s fantastic.

Michael Thomas: Yeah. That’s good.

Jay Bradford: Well, look, hey, we are so glad that you joined us today. I mean, this is going to be an amazing episode for our listeners and viewers. I know Michael and I have gained a lot just from talking to you about these topics and working through them. So look, hey, go follow us on all the socials. Go listen to us, go watch us, go follow Clear Winds on all the socials. This is Jay and Michael on the IT Directors Podcast, and we are out.