Navigating the Realities and Risks of AI in Business Strategy

01The real through line again is that I think, and what you just said, that companies that skip the AI literacy don't want to be too biblical here, but they're building their house on sand. Because if everybody doesn't understand fundamentally what it is, why we're using it, and how we should use it, that's all part of the literacy, then it's not going to work. It's outcomes first. You know, what what outcome are we looking for? And then we use AI for that. I like the that opening line for every one of them is this changes ever. You don't we don't really know what we want to call this. We know what we don't want to call it. We don't want to call it like the salience project. Or peripheral Yeah, we're not gonna call it anything like that. So we have no idea, but we did wanna talk a little bit about Where AI is taking us long term and maybe the y the myopic focus executives have right now on what what is it gonna do for me this quarter, I think that might at a high level summarize kind of what you were thinking about.

02Is that at all accurate? Yeah, right. I think we're gonna use this as a chance to I guess ⁓ figure out if everybody is caught up with like what's going on in the marketplace, things that are happening. kind of the ⁓ the point of what we're trying to get folks to understand ⁓ as executives, are are are we really getting like the ⁓ the application of AI and not just ⁓ you know the the joke I keep making about buying two pounds of AI? Which is true. I ⁓ was at a event the other night and a gentleman that does something similar to what we do at a larger firm. They've been doing it a while, ⁓ so they have several employees and they do fractional and project based work almost as an operating partner for a lot of mid market PE companies, and a little bit larger, so pretty far up the mid-market. They We're saying that they walk into talk with the C suite all the time. So after they've been engaged and they're maybe they're doing a scoping engagement at the beginning and they say, Okay, what's the problem?

03And they go, Well, we have this supply chain logistics project that we want to work on, but whatever you do, you've got to have AI in it. Yeah. That's the that's that is exactly what they're saying. And so his point was, well, there's when everything is you know, well, he said AI is being used like a hammer right now. Every problem looks like a nail, and we're using AI as a hammer, and that's what he's taking out of those conversations with the leadership teams. Huh. Is that I feel like that angle, it's a blunt object. The golden hammer, you know, the ⁓ you know, AI's the hammer, everything needs to be hit with it and it ⁓ everything you touch with it is gonna you know, i is gonna succeed. ⁓ It's feeding into a lot of like poor behaviors, right? And so maybe one of them that we were we're going to talk about here was ⁓ around ⁓ how are we measuring success using AI? It's June, right? As we're recording this podcast today, and ⁓ headlines are saying, you know, I've burned through my entire budget ⁓ of of of AI for the year, right?

04Like it like everybody's saying now it's stop using AI. does ⁓ or or use less. And the question is, is that reality? Did you did you did you measure the right thing or not? Is AI too expensive? are are people using it too much? and ⁓ I think like those are those are logical ⁓ questions to to land on ⁓ but like I think it misses the point. ⁓ people are saying you know, shut it down because it's too expensive. And ⁓ you know, ⁓ the word that you hear people say is tokens. So, you know, do you do you feel like the people that, you know, some of the f some of your clients that you've been talking to, like, you know, does that come up like with with ⁓ the the folks that you've ⁓ have you have you done a little bit of education with folks about, you know, ⁓ what when they hear it, what does that mean? Example two weeks ago, I so I'm doing an AI literacy course for a firm. So every two weeks I go, I show up and I talk to them about how AI actually works and I try to relate that to their job. And I also try to bring it down to a level where it's a little bit more explainable.

05Last time we talked, we talked a lot about context and why people give up on AI and say that it doesn't work. Which is still super супер комен. One of the things we talked about were tokens. And I think maybe you can give a definition, but I'll tell you what I told a group of people that don't have any idea what a token is. I said, just think of it as a word. Now I know it's I know technically it's probably a little smaller than that, because I mean this these things look in chunks. But maybe you could take a second and just define token and then Perhaps touch on context lightly. Lightly. I mean, right. A token is a word. It's a fragment. It's it's characters in a row, right? I I think ⁓ as we have been chasing how smart AI is, ⁓ you know, there's a little bit it and this is super reductive, but right, AI is about predicting the next word. And so No. It's very good at it, right? It's better than the T9 texting that we all had on our Motorola razors. Yeah, that was yeah, there's so many people are not even gonna remember that.

06⁓ that's gonna that's so great. Okay, but it's it's about predicting the next word. Now, what we used to be able to do with predicting was it was just kind of a a dictionary and the likelihood of the next letter, and ⁓ and now we're predicting words and the way that we predict those words when you hear you know, and maybe we'll do a whole thing about like how LLMs work, but the for every token, a ⁓ Big math equation slams against what's called the LLM. The LLM is a set of mathematical data points that you can either host on your computer or ChatGPT or Claude have up in the cloud. And ⁓ and these ⁓ these math equations just smash against this, like think of it as a giant cube of numbers. It's like a giant it's a matrix on top of a matrix and top of a matrix. Math slams through all these numbers. And what happens is a token pops out the other end. And for every single token that pops out the other end, you do this incredibly complex math equation. And sometimes the token comes back and you pr try to predict the next token based on the token that you just used.

07There's there's a lot more there, but ⁓ that's what you are paying for is this computational time and ⁓ You know, that's why NVIDIA is doing so well right now as as a as a company, because they've got specialized chips that are are very, very good at this type of computational high order math. ⁓ but Tokens are just that. It's just a measure of utilization of the system. And what we saw, ⁓ I know you and I were talking about ⁓ you know, news articles happening where people are saying we have these token dashboards. Amazon had a token dashboard, right? They were like, who's the top employee who is using the most tokens? ⁓ and and the bet, not a dumb bet, but I think a short sighted or a or a little bit of a blunt bet was if they use it, ⁓ it's gonna yield some cool new things, ⁓ internally and things are gonna get better. And ⁓ That's not really true. Yeah, so if companies are burning through all these AI budgets though and complaining about costs and then starting to freak out that they don't have what they thought they would have.

08Did they have an idea of what they were doing in the first place? They I think, you know, it really was like field of dreams, right? Build it and they will come, right? Like use it. But ⁓ you know, I I was thinking through as we were like preparing for this and I I was trying to remember the story. And then there's two like kind of historical stories around ⁓ the ⁓ it's ⁓ perverse incentive. There's a perverse incentive here, right? There's one that happened that there's like it's whether it's myth or real about like cobras in India or ⁓ the ⁓ There were rats in in Hanoi, Vietnam, but like ⁓ there was an infestation. The government said, ⁓ we're just gonna pay a bounty ⁓ on ⁓ on them, and ⁓ and the locals will will take care of the problem and they'll bring in, you know, ⁓ you know, the tails of these ⁓ you know, ⁓ of of these rats and prove to us that they've killed the rat. And ⁓ no, that's that people looked at it and said, I will breed rats. And cut their tails off and bring them in and collect the bounty.

09And and I think in both cases, right, the like the if i i i even if we treat it a little bit as like a little moral, you know, ⁓ kind of a story, ⁓ the moral of the story is is ⁓ it didn't work because people were greedy. And at the end, the problem became worse because when the government finally said, Hey, we're not gonna pay for this anymore, ⁓ what what did they do? They just released all the stuff that they had. out into the wild and the the problem was even worse, right? Because breeders were ⁓ breeding these animals. But we kind of did that. I mean, I I read these articles the in the in in you know newsletters and on the news about what's going on at Amazon. And I feel like ⁓ is it called token maxing? Yeah. ⁓ yeah. Like the ⁓ the the young kids are doing looks maxing. That's true. I just wanted to let you know you're you've got a lot of aura today. ⁓ sorry. I I I mean I I I'm I'm trying to make sure that I have rizzed. ⁓ yeah. Well Yeah, no, that one's you missed that one. All right. Sold you sold on that one, brother.

10Yeah, I you probably don't even know what that meant. I love you know, you know, but I mean ⁓ only got I two middle schoolers, ⁓ and they're young middle schoolers. ⁓ yeah. Yeah, you're you're you're getting there, but by the time you do that that word'll be gone too. Yeah. So These guys are token maxing. Did it pay off? It doesn't sound like it did. One, because I don't think they extracted value out of the thing that they thought they were gonna get. But also They didn't get corp like corporate value out of these tools, like like new things didn't get built. But I kind of challenge that because the incentive was just use tokens, the other value that, you know, maybe we that that you know you could have gotten out of this is people would have been built a fluency for these tools, right? Like like the the this is, you know, developers getting at bats, right? This is you on the bench like doing more reps. and getting better at your game.

11But if the game was not to really build good things and it was just to kind of play around, ⁓ or burn tokens, come up with unique ways to burn tokens, ⁓ I don't think you got that benefit either. Yeah, but as we teach people about these things, and this this is something I kind of just thought about, but we're teaching people things like context, tokens, you know, you're talking about layers and nodes without saying the words, you know, you're you're kind of talking about that and probability weighting. When we teach people these things and they hear the word and they get a little bit of They grasp at the highest level what it means. Does that make them too confident? Do they now think they understand AI and then dive in too deeply? I think there's ⁓ danger of trying to understand it and then ⁓ you know there's the overcorrection, right, that's happening ⁓ at this point, right? Because it now, okay, we spent too much money on AI. So what should we do?

12⁓ we should find cheap models, we should ⁓ we should host our own AI and not pay ⁓ Anthropic and Claude and Gemini for for for their tools and ⁓ and so now ⁓ I I see a lot of you know, I see a lot of stuff on social media around p the AI folks about like run your own ⁓ agent. I mean that it's kinda evolution of what we've seen over this year with like open claw into ⁓ you know other agent based tools. ⁓ you're playing with ⁓ is it Hermes or is it herme's like like the fancy, you know, jewelry and and ⁓ and and Birkenbags. I have no idea because it's not on my computer any longer. Yeah. That's a whole nother of itself. Yeah. Whole other thing. All right. Well we'll do agents another day. Epic fail. But people are chasing these and ⁓ I'm kind of you know looking at it as ⁓ you know turning into a coupon cutter type of economy. I gotta cut you know, clip coupons. And now I'm instead of building value for my company, I'm just trying to figure ⁓ out What's the most unique way to clip coupons, save a dollar on ⁓ inference budget? Do I really need to save a dollar?

13Like you know, is that really moving the needle for our company either? Don't think that it necessarily is. I don't think I think this is really about what are we building these for? And are there outcomes that we have sat down as a team and laid out so that we know what the end looks like. We know what we're trying to accomplish. Right now, I think it's a lot of Exactly what we talked about earlier. Give me two pounds of AI in a bag, or you can do this project, but whatever you do, consultant, make sure it has AI in it. Right. Like it w it it's a it's a ship adrift with you know without a North Star. And ⁓ I think what you know you and I have talked about in other contexts, and I think for the folks who listen to our podcast, you know, might maybe as as executives, they've heard in their careers, right? Is ⁓ you know, we are trying to avoid theater. And what this is creating in a lot of ways is a kind of new AI theater, whether it is it's about using AI, being the guy who says, Did you hear me in that meeting?

14I told that vendor they have to use AI, or, you know, or being the engineer who's who's, you know, trying to in their, you know, status report, trying to prove to their boss that they are doing their part by ⁓ cutting coupons and finding you know, cheap ways to do AI internally. ⁓ and they're missing that North Star. And like where I was getting at was that they're missing this fluency, right? You and I are trying to train companies, you know, to g to build some level of fluency on How to do work that is meaningful and that actually like, you know, that you've sat down and had a a a a real grown up discussion about what what's what's the outcome? Agreed. I think this whole idea of coupon cutting and doing things, you know, on the cheap, self hosted, I it could create kind of two tiers. So the large companies who have a North Star, who understand how to think strategically, can afford McKinsey, Bain companies to come in and guide them and really give them a focus. That that could be a real separator because mid market companies don't always have access to that.

15And there's only so many people like you and I who are thinking about this in the long term that are out there and available to them. That's that's my thought. Yeah. I think the what I don't want folks to lose, and I mean we will spend time talking with I mean, I'm talking to clients, you're talking to clients about how to use these tools, how to be efficient in using these tools. There is there is value in using tokens smartly. And ⁓ you know, w we're we're teaching folks as they build those skills how to be better, but you know, we're also not trying to teach those out of the gate. ⁓ trying to teach people to optimize their token budget when they don't necessarily understand what are some of the use cases that they could use to ⁓ get benefit to do some of the things that you and I are doing. ⁓ you and I are try are doing some optimization things. We've talked about buying local inference and ⁓ buying ⁓ NVIDIA Spark devices and and running local models and and ⁓ you know building ⁓ skills and we'll talk to folks about that.

16But I think this idea that Because it's easy to to maybe lean into AI's expensive, and so let's ⁓ optimize it, or I'm not using enough, so use more, or telling people that you they have to use it, that's a those that's chasing a trend. It's not actually talking to our to our peers at the ⁓ you know, sitting at the s at the at the C suite table and saying, well, what is the outcome? You know, what's what's the th, you know. as I feel like we're often talking to our PE clients about like what is the what was the thesis that was what was made ⁓ into these portfolio companies and are they trying to really figure out ⁓ you know how they're gonna deliver on it. Yeah, pick one of those levers, decide how you're going to execute those levers on that lever. So let's say, hey, one of the levers was we were gonna, you know, have a loyalty program installed, and we know that based on industry research, it's going to increase ticket size and frequency of visit by X percent, whatever that is.

17Then trying to solve that problem the old fashioned way and then figuring out points along the process where AI can then help accelerate that process is the way that I tend to think about it. You know, start with old school. What is the process? Understand from start to finish what you're trying to accomplish and then Find the places inside of it where you can have the biggest impact. Right. Not just where you can have, you know, biggest impact, but you know, as leaders, how do you educate and find ⁓ ways to bring your team along with you, right? To join that like, you know, mission. I think ⁓ 'cause I think a little bit of this is ⁓ like we said, like with w when we're telling people buy two dollars worth of AI, it's a little laissez faire like, you know, like saying the words is like Harry Potter and ⁓ You know, you said it and therefore magically the AI wizard will show up and ⁓ you know, something great'll happen. Yeah, I think you have to stick with it. That goes back to your point about AI literacy and a real program around it.

18Like how do you educate, teach them enough, and then let them experiment. So I don't think and maybe we're circling this, the two are mutually exclusive. There can be experimentation and everyday use of AI. in a responsible way to speed up processes that aren't necessarily touching the customer. I think when you move further towards this is going to be customer facing, this is a touch point, this is something that externally will be relevant, that is when you have to get really, really intentional. So I don't think that you can do one without the other. You have to let your your smart employees experiment and go after it. And then you have to give them the tools and the knowledge, really the knowledge. The tools are easy. That's click a button buy. The knowledge to use the tools, that is where I think companies, in a lot of instances, aren't spending their time. They don't want to. They just want the AI. They don't want, ⁓ we actually have to learn a little bit about this. AI looks AI looks really simple. It's a chat box. It's a box where you type and things come out.

19It looks super simple. It doesn't look like some of these other systems we've used in the past. It doesn't even look like email. How many buttons and widgets and levers do you have inside of Outlook? There's so many. This is deceivingly simple in that you type something in. Right. A lot of times you record something in, you just talk to it, and something comes out. Yeah. It is deceivingly simple and I think the value that some people see at the very first brush or blush with it is is ⁓ It's like getting a piece of candy. Like it like or you know, like it ⁓ for some reason it like like it it it it immediately satisfies, you get the little dopamine hit ⁓ and ⁓ It has the ⁓ danger of distracting you from thinking at a higher level with it. and I think that's where, I mean, one, we do want people to go play with it. We don't want people to experiment and try things, but even those experiments, if they if they stay at, ⁓ it's just a chat bot, right? I'm gonna ask it a question, it's gonna give me an answer.

20I'm gonna give it a really hard question, it's still gonna give me an answer, right? Is it's still kind of this ⁓ you know, first order type of a ⁓ of a of a problem. You know, I think I got super nerdy with you the other day talking about like, you know, like ⁓ you know, whether or not this makes sense to kind of like, you know, with calculus, right? We you know, you you do kind of your first order derivative, your second order derivative, and those are ⁓ math affects us in the real world. I mean it I mean it it it it it's one of those things where like the physical world d is kind of tied to, you know, distance becomes speed, w ⁓ which then becomes acceleration, you know, based on the math that you do. And and ⁓ the the funny part of that is that you have to know distance and speed in order to get acceleration.

21Like those things are they compound and in order to like solve problems and ⁓ here we get stuck at that very bottom level of like I just understand distance and and you and I and I think people who listen to our podcast are being paid to do that second and third order thinking. What we're we're trying to do like that's where strategy lives, right? Operations live and and neither one is not better than the other, but they have their place. And I think we're trying to get people to think this step above just, ⁓ I, you know, it it is a simple tool. It's deceptively simple. But we could build things that are not simple, that take ⁓ data and derive it and ⁓ do more than just give me faster results, right? I d ⁓ faster dashboards. I mean, you you're looking at it from you know a a point of view of pulling in that you know, retail franchise data and a trying to figure out ways to make it actionable. Right? That and that's where AI can help. Absolutely.

22I mean it speeds up the process of organizing the data, which is another thing people People are still they don't have the right staff or talent on the team to take advantage of these higher level. You know, I don't know, you're calling ⁓ the second and third order Problems. They don't have the people. I mean, it all goes back. You have to go far upstream and say, okay, do I have talent that can even do this? Because, you know, I'm here I still hear about RFPs for enterprise data warehouse. And, you know, how do we get these? And what do what do people really do when that happens? Well, they bring in a group of consultants and then the consultants do the actual work, which is really a lot of times just saying. Okay, tell me what you want me to do, client, and then I am going to hook up your data that lives here into a database that lives here. That's it. And that work to get the data back and forth. Smart people with no technical background are doing that work now.

23However, I still hear about these RFPs that are going on at these companies for hundreds of thousands of dollars where they're hiring another team, but one or two incredible people on staff in IT, or maybe they're in FPA and IT, they can do this now. They just don't know how, or they haven't been shown that it's possible. think they need to work together. I mean it's IT and FPNA. And I mean and starting it out in FPNA, I think makes sense because there's there's you know knowledge of the underlying data, what it means and what you want to get out of it. ⁓ you know, and and ⁓ and then we let the the the folks who have architecture and security chops come in ⁓ to make sure that it's handled well but ⁓ We can move really fast if we get the right folks involved. And those people are the right people, like you're saying, right? They have an intellectual curiosity and a certain amount of drive and grit to make it past the obvious and the easy into something that is ⁓ you know, outside of their comfort zone and at the same time related to their, you know, field of of ⁓ expertise.

24Yeah. Yeah. Hundred percent. Yeah. So maximizing token use, probably not the best thing that we've you know, that we we're seeing out in the marketplace. Clipping coupons. That's, you know, the other side of that one. Again, not trying to shame people, but you know, the two ends of the spectrum are are not places that you want to hang out. no. Finding the Usually I mean, in life you usually don't want to. No. But I mean that's what that's what's getting the news, right? That's what that's what that's what we're seeing ⁓ you know, in in both of our ⁓ news feeds. I mean, so maybe that's something that we kind of like talk about is like, you know, here it is, it's the middle of the year. ⁓ we've hit a point of evolution of some of this technology that we maybe didn't see coming at the beginning of the year. Things have happened. ⁓ you know, we're seeing ⁓ IPOs now, which which have got a lot of headlines. Yeah, there's two two specifically. I mean, ⁓ Anthropic and OpenAI are both kind of in a race to see who goes first.

25Anthropic, at least as of last week, was was it nine hundred thirty-ish billion dollar valuation. ⁓ so ⁓ a little higher than open AI. So they are they're winning that race at a couple of different Events that I went to over the past week and a half, that is where most of these mid-market companies and larger companies are, you know, they're hitching their wagon to anthropic and clawed. So open AI is still, you know, kind of seen as a consumer product in some ways. ⁓ although I know that there's a lot of good things happening on it. However, there was ⁓ we're So we're gonna try a little bit of a news segment. New and news. So it's a but just talking through a lot of the different Topics that are relevant right now from just ⁓ since the last time we left our heroes on ⁓ the the previous podcast. So speaking of anthropic and AI, they're both kind of embroiled in in some legal situations. Anthropic, I don't know if you saw this, you don't use Claude. I use Claude. Claude. You do.

26I think we ⁓ like the the you you you are the yin to my yang with ⁓ claude use versus my codex use. No, and I think that's good. And I've been using both. You mean you you've got me using Codex. I think it does a I think it does a nice job. I'm still I I like I I see some benefits of Codex that I don't see in Claude Cowork. and I can get into those a little bit later. So I've kind of seen both sides now. But the big story over the weekend was anthropic. got shut down in terms of using its new model. So I got to use it for about five minutes is what it felt like. And then it was gone. And they have all over their website, you know, Claude Fable is currently unavailable and learn more. You can't get rid of that. So usually inside of Claude, if they have a pop up Typical one is, hey, you're at 90% of your usage. ⁓ slow down or give us more money. Now, this one, and then you can click an X and it goes away. This one is Claude Five, it Fable Five is currently unavailable. Learn more. No X. Can't get rid of it. Just lives there. It's with you the entire time.

27So clearly they're trying to get the users to get you know ginned up. Over this, post on X, get out there, you know, let the government know what they're thinking. Because I guess this all stemmed from Amazon saying that Fable Five, they thought it was if you prompted it a certain way, it could find vulnerabilities in software. And then that was supposed to be a guardrail in Fable Five, which is the same thing as their like they call it Project Glasswing. Which is what they took mythos, which was w what kind of like Fable Five is supposed to be a little bit of a guardrailed version of Mythos with another enthropic product. Which is weird too, because not only are there guardrails, but sometimes they're invisible. They don't even tell you. And it just falls back to one of their older models. So that's frustrating people as well. The bottom line is this is the first big kind of government intervention. So Yeah. Anthropic tries to launch a model, government says, Whoa, that's too dangerous. Stop.

28And it's specifically they said stop for foreign nationals and foreign companies. So the problem was is Anthropic has a lot of those foreign nationals on staff. And so they just had to shut it down for everybody because there's no way for them to be that. Right. Turning into a bank, it's a know your customer exercise. Yeah.