The AI Programs Being Made Mandatory at LSU with John Munsell, CEO at Bizzuka

(Intro Narrator at 00:00:00) Today, we're talking to John Munsell, CEO at Bizzuka, about AI programs he's developed for LSU and technology leaders. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:16) So I was actually very curious to talk with you about what you're doing with Louisiana State University. Your AI systems are, like, mandatory knowledge there. Can you explain that to me?

(John Munsell at 00:00:28) So what happened is, a couple—was it a year ago?—the provost of LSU came over. He had heard I was doing some cool stuff in AI, and so he came over and we started to talk about it. And then he was telling me all the things that LSU was doing. Then he asked me, well, what are you doing?

(John Munsell at 00:00:46) And I showed him this framework we had created to help people really understand how to generate prompts and how to get the most out of AI without constantly having to go back and forth. And he was like, "Whoa, I've never seen that before. Do you have the prompts to match this framework?" And I was like, "Yeah." And I showed him this structured approach that we use.

(John Munsell at 00:01:06) And he was like, "Wow, I've never seen anything like that. Would you be interested in teaching the honors students next semester?" And I was like, "I don't know that I want to do that, but I think we could have a bigger impact on the workforce if we taught businesses and their employees." And so he said, "I love that." So he connected me with LSU's continuing ed, and we put together a program.

(John Munsell at 00:01:30) And now we go through LSU to help upskill people and teach them about what we call scalable prompt engineering. It's a different kind of technique for prompting AI.

(Joel Beasley at 00:01:42) How does that work?

(John Munsell at 00:01:44) So think of it this way. I always kind of liken it to a spreadsheet. You've built spreadsheets before, right?

(John Munsell at 00:01:52) Okay. So, you know, in a spreadsheet, you don't hard-code a variable into a row and then copy it across the columns. Like, if you wanted sales to increase by 10% all the way across, you know, 24 months, you wouldn't put—you wouldn't say, "Hey, take this sales number and then do a times 1.10" and copy the 1.10 all the way across, right?

(John Munsell at 00:02:16) You'd put 10% in a cell up here, and you'd have that formula refer to that cell, and then you would copy that across so that you could change the 10% in one spot to 12, and all the formulas would change. Make sense? So it's the same kind of process with AI. So instead of creating these giant paragraphs of prompts where potentially you could get what we would call prompt conflict, we teach people how to bracket them up or containerize bits of information in what we would call a stack and a variable. And so then somebody else can look at this prompt and go, "Oh, if I take this stack out and I plug this stack in, I can get a completely different result out of it."

(John Munsell at 00:02:44) So if a container, for instance, is their persona—let's say, and your target persona—here's kind of a classic example. If in marketing you're tasked with recruiting, let's just say, a new operations director, okay? And if you're going to post that job on, say, Indeed.com, that job has to sound really sexy, right? It's got to appeal to that person, that persona. Well, HR has to write a job description too, but that job description has to appeal to the regulatory authorities, to the internal teams, et cetera. So it's got to be more clinical.

(John Munsell at 00:03:29) All you're doing is you're telling AI to write a job description. But in one role, you're just saying, "I want to target this position or this persona," and then you might give it another variable that tells it how appealing it should be. And then HR just swaps that persona out and makes it for a handbook and changes one of the descriptions. And the prompt—or the request for the AI—is the exact same thing. Makes sense? But it creates a completely different output. And there's lots more examples than that. But once people see that mechanism, then it allows them to create shareable knowledge. But if they just start writing paragraphs of prompts, that's hard to scale and hard to share. Does that make sense?

(Joel Beasley at 00:04:46) What software are you using for scaling and sharing prompts?

(John Munsell at 00:04:51) So internally, we use Notion, and we're in the process of building a database that people could share internally. So the idea is that it's not just this, you know, massive group of Word documents, but it's a structured process for approaching prompting where people could actually look at the variables. They can look for the stacks. They can get access by searching a specific use case. They get rewarded when somebody sees their prompt and shares it and uses it or repurposes it.

(John Munsell at 00:05:32) So it's kind of like gamifying prompting, but it's a very structured way to approach prompting, which just makes it so where everybody could—I could go look at anybody's prompt and immediately get acclimated to it and know what they're trying to accomplish, and I could probably see where it's going off the rails too.

(Joel Beasley at 00:05:51) Have you looked at projects like Agenta, like the prompt engineering ones where you craft a prompt and then you run it against the model, you can compare the analysis versus multiple models? You can, essentially like GitHub, you can make iterations and commits and changes to the prompts. Have you seen these groups of tools?

(John Munsell at 00:06:09) I haven't really messed with that one, but what I do—it depends on what the model is. There are, like, for instance, Claude responds differently to this specific style. They all pretty much know the same thing. They all respond the same way. It's just that the variables that we use might adjust.

(John Munsell at 00:06:30) So for instance, when it comes to copywriting, you know, Claude does a really good job with copywriting. So I don't have to put as many guidelines in a style voice or something like that. ChatGPT, I have to put a little bit more into it because ChatGPT, I don't know, it just writes differently. If I were to go into Gemini, same thing. There would be parts of the prompt that it would potentially ignore. And so I've gone into all of them and taught it how to write this way, taught it how to write prompts this way, and they all do a pretty good job. But I always have about a 20% correction that I have to make to it because it will help you refine your idea, but the execution is always short. I don't know whether that's your experience or not, but that's kind of been mine.

(Joel Beasley at 00:07:28) Yeah, I use it heavily. Oh, yeah. I built my own custom tools on top of it to help me use it better for our own internal projects. And yeah, I would say that your assessment of Claude is right. We use Claude when we need to write content. It just has a better first iteration. But nonetheless, we still have to tweak it and teach it for each individual use case. We actually built our own prompt management software because what was happening is we have multiple producers across multiple shows here for the podcast.

(Joel Beasley at 00:08:04) And so we wanted to standardize—okay, this producer had his own prompt for how he turns transcripts into show notes. This producer had his own prompt. And so there wasn't any social fabric that connected these prompts and allowed them to say, "Okay, this business, this is the prompt we use to turn transcripts into show notes," and then, you know, to be able to see that template and modify it over time. So yeah, we've had to figure that out quite a bit.

(John Munsell at 00:08:36) Yeah, but see, what you've obviously figured out for yourself is how do you scale that, right? Because if everybody's doing it differently, you're all going to get different results. But if—I haven't obviously looked at any of your prompts, but when we're dealing with an organization, that's the first problem that we encounter is that it's fragmented. Nobody's got a standard. Nobody has any kind of framework for understanding what needs to go into it. Especially when they start building like Claude Projects or custom GPTs—I mean, forget agentic workflows, just start at the, you know, the basic level, right? Everybody's doing it differently.

(John Munsell at 00:09:18) And some people have gotten to the point where they've got a routine and the rest are still trying to figure it out. And then once they say, "Okay, well, now let's try to get more of the staff using it because, you know, Jane over here is kicking butt, but Johnny over here, not so much," that's when they realize, "Okay, we have to have some standards. We have to have a framework. We have to have a common level of understanding and a process." And that's where this approach seems to help out because it's just a very easy way for people to understand what's in the prompt and then what parts need to be standardized so the output's the same.

(John Munsell at 00:09:41) But, like, you know, like you were saying, you're grabbing the transcript and turning it into show notes. That's a really specific process. And it's an easy one to standardize. But the other thing is—I don't know, you're writing, as you are well aware—the writing that these GPTs will do is influenced by so many things in your prompt. If you don't know what's affecting it, it's really difficult to get it to write the way you want it to, and you keep writing sentences differently and doing things differently to get it somewhere. And it's typically because you haven't recognized that there's some other piece of your prompt that's conflicting with the one piece that you think is going to affect the writing style.

(Joel Beasley at 00:10:56) It's a lot like Googling. You know, you could hire 50 different people and the ability for them to Google for a result is wildly different across the individuals. And some people—like, I've been a software engineer for 17 years. I happen to be really good at Googling, right? Like, I'm really good at searching, and then I'll interact with other people in my life who, you know, don't have that experience. And I'm like, "Let's get it. Let's get it going. Let's move faster. Let's get a nice, quick, good search term in there to get what we want out of it quickly."

(Joel Beasley at 00:11:23) But yeah, prompting, I think, is going to be a huge—I mean, I don't think it's going to be—it is a huge skill. It's a massive skill you have to have, and whoever's adopting it and figuring it out. So when I saw that you had made a framework and that you're teaching organizations how to do well with prompting, I'm a curious person. I'm a learner. It's not like I've got it figured—I don't think I have it figured out. I think I've done a pretty good job of getting it to achieve the result. But I want to find out—who are smarter people? What are they doing? You know, what's going on there?

(John Munsell at 00:12:03) Well, so here's the thing. You've figured it out. You're in that level of unconscious competence, right, where you know it so well, you aren't even sure why you know it well, but you know when you look at somebody else that they're doing it wrong, and then you just kind of start doing it. If you were to document what you know into a process, then it becomes scalable knowledge, right? But you're so good at it because you've been doing it so long. And this is what we find with a lot of software engineers because, I mean, I've had a software development company for the past 25 years—and I actually, 26 years total—and sold off the agency side, as we would call it, about two and a half years ago to do this.

(John Munsell at 00:12:50) But software engineers get so good at what they're doing, and they get so pissed at anybody that doesn't know it like they do.

(Joel Beasley at 00:12:58) My wife would confirm that. Yes.

(John Munsell at 00:13:00) Yeah. My wife would too. It's so hard, leaning over her shoulder and go, "No, no, this button," where you're, like, you're almost dragging her hand to where it goes there. Oh, gosh. I hope this one doesn't make it on film.

(Intro Narrator at 00:13:17) It's a good one.

(John Munsell at 00:13:19) But what, you know, what I figured out was that if I could dissect how I was doing it so that I could explain it to somebody else, then I would be on to something. Because in my head I could see it, right? I could just see it in my head, but I'd never really forced myself to put it on paper. And so I started teaching people because I just, you know, I had this concept of, "Look, if everybody structured their prompts this way, it would be really easy to figure out where they went off the rails, and it would be really easy to correct it."

(John Munsell at 00:14:01) And so I taught a six-week course, and the first two weeks, I could see that they weren't getting it. And I was like, "Man, you know, one of my superpowers has always been to take a really complex concept and simplify it in graphics," because I'm a visual person, you know? And I used to be in the financial services industry for about 17 years, and to explain investing concepts for a business—it was really challenging for a lot of people. But I found a way to graphically depict how it would work, and it was simple, right?

(John Munsell at 00:14:40) And so I thought, "Okay, I've got to figure out how to explain to these people what needs to go into a prompt so that they don't have conflict and they can see where they can do it." And are you familiar with the Business Model Canvas at all?

(Joel Beasley at 00:14:55) Oh, yeah. Yeah, there was a whole series of books, but one of them, the Business Model Canvas. Yeah.

(John Munsell at 00:14:59) Yes. Business Model Generation was the first book. And I saw that guy speak in 2012 at an Inc. 5000 conference. And so I immediately bought the book, and I blew up the poster and would walk our employees through it and say, "Look, this is how we operate as a business," so they could really kind of see all the moving parts. Because, you know, as you're trying to scale a business, a lot of people don't see the moving parts. They see their part and they get pissed because, you know, they think you don't see their part. And so when they see the whole thing, it makes a lot more sense.

(John Munsell at 00:15:26) And so when I was struggling to explain it to them, I don't know, man, all of a sudden, the Business Model Canvas just popped into my head, you know? And I literally had a six-foot version of it sitting in my office. I'm like, "Okay." So I grabbed it, and this is a hoot because I knew all the ingredients that needed to go into a prompt. And I thought, "How do I do this really quickly and then finesse it?" And I thought, "Why don't I go to ChatGPT and see if ChatGPT knows how to map these things out?"

(John Munsell at 00:16:10) And so I went through a couple of exercises with it to get it to map them out. And I was like, "It's not quite what I'm seeing, but it's pretty close." But I was like, "Hey, if AI gets it, then these people will get it if I just get it in the right framework and right organization." And so I started to manipulate it because the one thing that ChatGPT couldn't figure out is the sequence, you know.

(John Munsell at 00:16:31) So in the Business Model Canvas, if you remember, there are two blocks at the bottom, and one's for revenue and one's for expenses. And then across the top, there are market segments and all these other things. It's divided into debits and credits, just like an accounting system, you know. And I thought, "Okay, well, you know, when you're constructing prompts, there are certain things that you need to have in it. And if you don't have it in there, you're going to have to realize that your output's not any good, and you'll have to throw it into your next request or your next request or your next request until eventually it has what it needs."

(John Munsell at 00:16:57) But if you knew what those things were upfront and you could structure them in blocks of information, then you could swap out those blocks, and it becomes a more scalable framework for prompting rather than paragraphs. And then once I put all that together and I showed it to them in the third week, everybody clicked. You know, everybody was like, "Oh, wow."

(John Munsell at 00:17:37) Now I get it. And at that point, they understood what the prompt stack framework looked like. They understood why it was scalable. They understood what to put in it, and they understood where potential conflicts would occur. And the next cohort, I was teaching it again, and everything went smoothly.

(John Munsell at 00:17:59) And then we came up to an office hours because we teach it, and every week we have office hours for people to bring their questions to it. And the end game is they have to do a capstone project, which is a custom GPT. Well, this gal was creating a chatbot for the ophthalmology industry. And she says, John, I got a problem. I'm like, okay, what's that?

(John Munsell at 00:18:21) She goes, I can't get my chatbot to stop talking like a pirate. And I was like, what do you mean? What do you mean? A pirate? Like a legit pirate?

(John Munsell at 00:18:29) She goes, yeah. And I go, show me what you're talking about. Because I was like, I don't know. Yeah. And sure enough, she puts in her request, and she goes, oh, hey, matey. Arr. It just starts doing all this junk, and you're like, what the heck? And I said, well, show me your prompt. And she had fallen back on the paragraph style prompt. And up here she had told it, you're this educational helper that teaches people how to learn or how to sell ophthalmology products inside the clinic. And then down here, for the voice, she's, and again, all in the paragraph, she says, I want you to use a little bit of humor, kind of like that funny uncle, and then she goes on to say some other stuff about how it's best to write.

(John Munsell at 00:19:15) Well, in the one stage, she told it to be a helpful assistant, and then she told it to be a funny uncle. Right? Well, the funny uncle then tells the AI it's talking to a child. And apparently, AI thinks pirates are funny to children. And so it starts talking like a pirate, a very helpful, funny, stupid pirate.

(John Munsell at 00:19:39) So I have to say, okay, let's go back. First thing we want to do is create a block called role. We'll just define what that role is very specifically. We don't want to get all flavorful or anything. Just say you're a sales trainer. Okay? That's it. We don't have to get fancy. Then in the voice, let's calibrate the voice a little bit. So instead of saying, I want you to talk with a little bit of humor, every AI tool is going to gauge its own level of humor.

(John Munsell at 00:20:11) So if she changes models, she has to change it. But if you say humor equals two out of 10, every model understands what that is. Right? So it's a highly calibrated way. And then there are other, we have like 56 different style of voice kinds of parameters that we could use.

(John Munsell at 00:20:30) We only use, like, I don't know, somewhere between three and seven in a prompt. But when you see something off the rails, you can look at what those adjustments are and shove it into your prompt, calibrate it, and it works. But, anyway, we did that, and instantly, it started to talk like she wanted it to. And it was because she was doing a different method, and there was what we call prompt conflict. Right?

(Joel Beasley at 00:20:54) So who are you teaching this to? Is this just like the average operations? Yeah. Yeah. Just average employee?

(John Munsell at 00:21:02) So yeah. Good question. Yeah. We're teaching it, well, the majority of people that come to us are C-suiters that are trying to figure out how to scale it in their organization, and so they just want to understand it. And now we're starting to get, like, we've taught, I don't know, 10 LSU professors. We've taught, I don't know, 20 different CEOs, CFOs, and CXOs of some sort. We've taught marketing agencies. We've taught a bunch of different, it's all over the board. But what I'm in the process of doing, well, I'm done. I just have to do one more thing.

(John Munsell at 00:21:42) I've written a book on this. It's called Ingrain AI, and it teaches businesses how to scale an AI-first culture in their organization from strategy all the way through execution. So it's not one of these books that says, AI is important and it's coming for you. It's one of these things that says, AI is important, but here's exactly how you do it. So it's literally walking you through how to do it.

(John Munsell at 00:22:08) So it's more like an instruction manual. So if you want to learn how to do scalable prompt engineering, there's an entire chapter on that. If you want to learn how to use the AI strategy canvas to run strategic discussions and define AI initiatives for your company, it'll teach you how to do that. So it does all of that, but not quite to an LSU's academic record or scrutiny to where it's sitting in your classroom. It's a business book for business people or for people who are just trying to use it on the front end. Right.

(Joel Beasley at 00:22:47) When will that book be out?

(John Munsell at 00:22:49) Well, with any luck, it'll be out next week. That's the dream. Inside the book, I tell people where they can go if they want to download stuff. And, you know, I've got some resources like, you know, prompt database and other junk. I have to finish building that website.

(John Munsell at 00:23:07) So it's really the automation behind it that's holding me up right now. I got my final, my editor going through the final pass, which will be finished this afternoon, and then I'm going to push it out there. But what I've done for other people, if your people hear about it and they want a copy of the book, if they go to Bizzuka, B-I-Z-Z-U-K-A, dot com slash Ingrain, and they tell me they heard about it on your podcast, I'll send them a digital copy of the book for free. Because I just wanted to get out there. I want people to...

(Joel Beasley at 00:23:42) They need to buy it on Amazon.

(John Munsell at 00:23:45) Well, hell, if they'll buy...

(Joel Beasley at 00:23:46) You'll make some money.

(John Munsell at 00:23:47) Yeah. There you go. Yeah. Yeah.

(Joel Beasley at 00:23:50) Put some Amazon links in there. We have a lot of CTOs, like chief technology officers, VPs of engineering, those types that listen to the show. And we do have some people that aren't in that category that listen in. But for the technology leaders that, you know, maybe they're not programming every day. They haven't been programming for five, maybe 10 years, but they've seen the AI come up. Maybe some of their staff's using it. What sort of tips can you give them that would make them look good? What areas should they be looking at? What should they need to know in order to catch up and get ahead of the curve here?

(John Munsell at 00:24:27) Good question. Well, every AI tool is good at something different. You know? And the reason I brought up DeepSeek's R1 is because you can actually access it through Perplexity. And he was looking to, don't upload confidential stuff in it. You're okay. But the reasoning stuff behind it is insanely fast and deep. You know? It's different than O3, but it's fascinating to watch it think through problems. So I would really recommend going to something like that to help you think through problems.

(John Munsell at 00:25:08) The other thing is, as a CTO, one of the things, if you're a CTO, you're in a company that's got more than six people in it. Right? So you've got some employees. You've got some responsibilities. If you want to scale AI use, then you really need to be in a position to help your organization do that methodically.

(John Munsell at 00:25:36) And that would be the key. Right? And so it's not just about, hey, Kim, which one of these tools is going to help us write code faster? Because, you know, O3 is amazing at writing code. O3 Mini and all those are, they do insane jobs with code writing.

(John Munsell at 00:25:53) So that would speed a lot of stuff up. But to really get AI scaling in your organization, you need an AI champion, you know, whether it's a chief AI officer or whatever. But that person needs to have a really large creative bone in their body as well as a structure bone. Right? You can't really get AI scaling inside of an organization if everybody's afraid to use it, and you can't get it to scale if the CIO puts too many handcuffs on people.

(John Munsell at 00:26:37) So I think a good CTO is going to be one who's learned how to be creative and has learned how to teach people how to use AI to help them with their own creative processes and to execute their jobs better. Right? And that's where I think coming up with a system, a framework that everybody can understand makes a lot of sense. And that's kind of what we try to teach is, look, first, get everybody on the same page. Let's let everybody speak the same language when it comes to AI.

(John Munsell at 00:27:13) And then that's where the magic really happens. But if everybody's self-taught or everybody's going to a different YouTube channel to learn how to do it, that's going to be herding cats if it really starts to scale in an organization. And that's kind of what we keep hearing out there. I don't know what you're hearing through your audience, but when I talk to a company that has, you know, more than 100 employees, they're actually starting to get tense because they realize that they don't have a handle on how this is being used. And you've got some people that are outliers and some people that are just too afraid to use it, and they know that if they don't get everybody in sync, then they're going to get passed by.

(John Munsell at 00:28:05) And so I think a strong CTO is going to say, hey. Let's get everybody in sync. If we get everybody in sync, you know, it's kind of like one of those, you've seen those big, what do you call them, scullers? You know, the those long boats where you got 10 people rowing, and, you know, they used to be in Tampa Bay all the time. But if you're not rowing in sync, you know, you're either going off track or you're slow.

(John Munsell at 00:28:35) But once everybody's in harmony, things just fly. So I think that would be my advice to a CTO is if nobody else is stepping up, you step up. But get it organized.

(Joel Beasley at 00:28:47) Well, John, I want to wrap up here on a couple questions about the book. Is the book for sale today on your website?

(John Munsell at 00:28:55) No. The only thing they can do today is request it. And then when it's available, then I'll either send them a digital copy or I'll, you know, tell them where they can get a hard copy. So but it should be available, well, the dream, Joel, was that it would have been available in November of last year, but now I'm focused on February 15.

(John Munsell at 00:29:18) So...

(Joel Beasley at 00:29:19) There you go.

(John Munsell at 00:29:20) Should be...

(Joel Beasley at 00:29:21) Valentine's Day. Maybe.

(John Munsell at 00:29:22) Day after Valentine's Day. Yeah. Yeah. Right. Right.

(John Munsell at 00:29:25) I don't want to get in too much trouble.

(Joel Beasley at 00:29:27) Right. Right. What's your one tip for writing a better prompt?

(John Munsell at 00:29:35) Structure in a single word. The thing that you have to try to do is think about what your goal is for the output, and then think about what the AI needs in order to do that job effectively for you. Like, you can't just go, I've seen all these little prompts that go, oh, I want you to write a marketing analysis for this industry, blah blah blah blah blah. Or I want you to write a blog post about topic X. The problem with that is that everybody can do that, and it's not going to be injected with your own thoughts, your own beliefs, your own company information, your own products or services.

(John Munsell at 00:30:23) All that stuff you have to provide it. If you learn how to provide that in a prompt in a, what I call, a hot swappable block, then your prompting goes to a whole new level because you can use those blocks over and over again. And you're not constantly reinventing the wheel when you're trying to write a prompt, and then somebody else can use it. So to me, it's structure. It's how you structure it. You know.

(Joel Beasley at 00:30:53) And if people are interested in doing business with you, how can they reach out to get you to come train or talk about what you're doing here with the prompt?

(John Munsell at 00:31:00) Yeah. Thanks. They just go to Bizzuka, Bizzuka.com, B-I-Z-Z-U-K-A dot com. And like I said, if they want to get a copy of the book, they go to Bizzuka.com/Ingrain, I-N-G-R-A-I-N, and we'll send them a copy.

(Joel Beasley at 00:31:15) Cool. Josh will set that up. Thank you so much for doing this. We made a podcast. How do you feel?

(John Munsell at 00:31:20) Love it. I appreciate you having me on.

(Joel Beasley at 00:31:23) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email, [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.