Can You Still Trust Tech Due Diligence? (w/ Jason Gill & Kirby Montgomery, Code & Co.)

0:00I think that's where we're driving to. And a lot of these businesses, especially with cost cutting and headcount reductions and whatnot, is just whipping Claude into doing more and more. And we're producing more and more code that we don't understand. And as Mythos, as Kimmy K3, as some of these high end models come out, they're going to destroy some of these smaller companies who haven't kept up with security posture and they haven't done the real engineering work. All the stuff that you thought wasn't important is now even the most important because people can't replicate that pain. It's not venture capital, it's private equity.

0:29It's the private equity funcast. Pour yourself a drink and have a seat. Welcome, Funcast fans, to another great episode with the Code & Co team. This time, I'm back with Kirby Montgomery. Thanks for having me. Salute. Head of North America at Code & Co and introducing his partner in crime today, Jason Gill, who is Engineering AI ML Practice Lead at Code & Co. Welcome, gentlemen. uh and we're doing this this is one of our first good remote episodes so i'm i'm zooming in from denver today where are you all coming in from here i'm in beautifulst louis missouri and it's veryst louis you meanst louis that's rightst louis missouri jason i'm gonna get in trouble this is now we get all the hate you're gonna get canceled in the midwest that's what i want yeah that's that's what i want i want the columbia maryland people coming after me too and jason where are you calling it from today i am in north carolina which is also beautiful it is beautiful So welcome, boys.

1:24Kirby, welcome back. We had a fun time in the last episode, but I think Jason's new to the FunCast crew. So Jason, can you give us a little bit of your background, which is a terrific background? I'd love for the fans to hear about it before we dig into today's fun topic. Yeah, yeah, yeah. I appreciate you having me here. Like you mentioned, I'm the practice lead for AI and machine learning at Code & Co. Previously, in my previous life, as a founding CTO, I started a cloud hosting company that I sold to a Fortune 500 about 10 years ago. and then started a MarTech company, tech-enabled services in the SEO space that we sold to private equity about five years ago.

1:58And we built both those businesses on a consolidator model, just gobbling up all the smaller competitors and, you know, little guys in the space. And so I've been on both sides of the table and the buy and sell side a lot of times. And when the Code & Code team called me up and said, you know, we'd like to start doing more AI diligence, it really seemed perfect because I get to do my favorite thing, which is look under the hood of a business, you know, learn all about it. And I don't have to do my least favorite thing, which is operating the business. So it's really been a great change for me.

2:20So when you were doing all those acquisitions, were they primarily just competitors or roll-up or were you buying features too? Yeah, mostly just roll-ups. We'd buy the smaller guys and just add them in, whack their staff and cut their costs and just keep going. Whack the staff. That's the P.E. The phrase that I was, they were necessary, right? It was necessary that people go at some point. So today's topic is a fun topic. because so i've lived through i'm old enough as folks know i've lived through all the revolutions so i you know i started software in 81 right so i've been through mainframe to mini to client server to you know browser base to mobile to i didn't go into cryptocurrency uh and and some fads that were basically useless but ai i think has the craziest adoption rate and craziest change change rate of almost anything else i've seen so you know when you think about the change from maybe client server to browser, it came slow, right?

3:19Some new folks put, you know, we're building new systems and new platforms on that environment. But existing companies, they had a long, you know, journey to the straight browser base, right? You could throw some read-only stuff. You could put some stuff around the edge. I think everybody kind of put their foot in the water. Like they were using the web, but it was okay not for enterprise applications. I think we're well past that this early in the AI revolution, which brings us, you know, to today's topic. Like this would be crazy for something that I would say really took off, right? In the fall of 2025 and 2026 from an engineering standpoint.

3:53Like we all, we were using chat interfaces, but when it really started to take off and able to do like hardcore engineering for us, it's less than six months ago. And here we're out to talk about quality of AI at a portfolio cup, right? Who would, so six months into client server browser, no way was anybody talking about that, right? That'd be crazy. But here we are. So how did you, A, come up with this, and how did this become a priority from a diligence and sort of planning for diligence perspective? Yeah, Code & Co has always done an exit readiness as part of our sales side work. And it allows for us to come in and partner with a company and do almost like a reverse diligence.

4:31And so what's the broad categories, Kirby? If you would just say, when you say exit diligence, because we're going to add this to it, right? We've done this with you many times. for our listeners, is it DR, is quality of engineering? Just some quick broad categories I think is help. I mean, we're on so many deals. The thesis don't seem that novel, but okay, scalability. So I was like, hey, if I put more money into this thing, is it going to scale? So we'd always be thinking about performance, tech debt. How's that all play into it? The next one might be around, can I expand to an adjacent market?

5:03Is the architecture going to fall over? Can it support that? Or if it's more around add-ons, you get a platform and doing that, how's that going to work overall with the engineering team and then what's the setup and so we would be looking at that saying okay well you're going to go into a diligence and there's going to be a code and co on the other side and they're going to want to try to find the gaps here to either validate the thesis or invalidate it and jason has this great saying he goes lois never go up they always go down and so what can we find here to really help inform that now with the ai piece it's more about a narrative and so it used to become great we'll do reverse diligence we'll help you understand what you need to have in your document room help you with the narrative a little bit here but they were calling us six weeks before and so you know when jason compares us to home inspectors on the buy side and the sell side all we can do is really window dress it you say put some paint on that cover this over here or disclose something but when we come into the situation now 12 18 months out from that cell it's really about saying okay, what is your AI story?

6:04And what is truly defensible? And what does defensible mean to the most scrutinous buyer and to the best nerds who say, actually, that's smoke, and there's no value in how you're trying to view that asset. And so that's really how it shifted for us on the sell side. So as we dive in, so interesting, are you seeing that people are primarily, if you were to put percentages, are believing the AI story? I'm not going to say hype, right? Because people who are close to it, we know what it can do and can't do, right? People who aren't that technical can kind of get crazy about what they think it can do, which is fine.

6:38Like that doesn't hurt the market. Then there's the other side, what I would call the Luddite group. That's like, it's a fad or it's not going to be that important. I mean, I think we see it as important and inevitable, right? And I think there is some that goes along with this. There is some, call it tension around it because of what an LLM is, right? Capturing other content. There's a little bit of proprietary copyright data. There's, you know, creators are freaking out about what you're stealing versus, you know, adopting client server. I don't think it was anything, you know, particularly, you know, political about doing that, right?

7:12There was nothing controversial about it other than with the technology worker scale. So, you know, are you seeing that both of you? What do you think the percentages are? Are you seeing people that are just too much Luddite, too much hype? Is it a traditional percentage? Are we so early in the marketplace? You don't know. Yeah. So what I see, and I'll let Jason wait on this as well, is that when we think about first just from the buy sides, if you're in the buy side persona, it helps them the sell side know how to position things. The buy side is paranoid. They do not want to look foolish in the side where they're like, okay, someone said something's proprietary.

7:45It turns out it wasn't. They also don't want to be purchasing an asset that's not going to be relevant during the whole period. So if we take the paranoia from the buy side and then we put it through the sell side framework, we say, okay, well, how can an engineering team? And then also the commercial leadership communicate what they have in a way that the spreadsheet people can understand as well as then the nerds, the other side can say like, yes, that is true. And so from how I see it, it's a lot about narrative, uh, because you don't want to have a situation where you invite, you know, a VP of engineering, and they talk for an hour about how novel their something is.

8:20But it also can't be too surface level where they're like, no, they're just putting hype on that. And there's one kind of funny thing that we do at Code & Co where we have a swarm of agents that will go look at someone's web footprint. But what I really like to do is look at the Wayback Machine to see when that company became an AI company. Because everyone's one today, but was it before Chad Chibbi came out or before you start to kind of think about narrative versus reality. But Jason, I mean, please feel free to weigh in on how you feel like the current climate is with paranoia and also reality.

8:47Yeah, I think we're at an inflection point somewhat. You know, a lot of the finance guys, they still haven't caught up with the hype of ChatGPT. They're not, you know, they're not up with an MCP and a rag and all these other terms that get thrown around. And so it is kind of easy for some sellers to pull the wool over on them or to really impress with features that look proprietary or that sound really special. But that if you're in the know, you know, you realize, oh, OK, this is just a simple wrapper. So I think we're at the point now where part of our job really, especially on the buy side, is helping inform buyers about what it is they're looking at, what really is special, what's unique, and what is proprietary.

9:25And then when we're on the sell side, working with sellers to help them really differentiate and tease out any of the BS. You know, you don't need to talk about the things that are table stakes. You don't need to talk about the simple stuff that you're doing. and it's really been enjoyable to help both sides to come to an understanding of exactly where the technology is at, where it's going, and how it's helping the business outcomes. So at a fundamental level, I want to dig into that a little bit. At a fundamental level, would you both, we should do this like the old newlywed game. We put Kirby in a soundproof booth and ask Jason the following question.

9:55Do you think it's inevitable, AI from the standpoint of you can't put your head in the sand on this one, right? That it's going to affect almost every business that has technology in it whether you want it to or not would you say that's true approved confirmed yeah so that's so that's the interesting part i think for me in the luddite side of things like look i'm going to try and convince you but to the day the world's going to move without yet and i think that's what's really strange about this particular phenomenon it is it's literally the adoption is so fast uh you know compared to compared to other technologies It is changing so rapidly.

10:33So to what the framework, let's talk about that for a second, because it's also itself is changing so rapidly, right? Like if we recorded this two months ago, we wouldn't be talking about mythos at all. We didn't even know it was there, right? And yet now, man, it's a major topic of conversation over the past, you know, first how powerful it was, then it was being pulled. You know, we couldn't get access to it. well, now we can sort of get access to it and, you know, and how that plays into token pricing. You know, when you say, Jason, this, you know, this is so much more complicated in some ways than any of these other technologies, right?

11:09When you say that your average finance person doesn't understand it, of course they don't. I mean, if you say RAG, I think they're going to say, no, we actually use like a duster because we find the RAG kind of puts us, I mean, literally like This piece of this, what's a model router? What's RAG? What's MCP? This is so much more technical. So how do you guys think about trying to keep that, when you engage with companies, trying to give them a framework that's flexible enough to say, guys, look, this is, unlike anything else, this is not set in stone, guys. This has to be a living process for you because it's changing so rapidly.

11:46I would say we've been doing some teach-ins. We've been working with firms to try to inform their staff and get them upskilled the best that we can. It's hard for anybody to keep up. And even Mythos now is old news. I mean, we've Kimmy K3 just came out. I mean, it's all, you know, it's impossible to stay on top of it. And it's having lived through the dot-com bubble, you know, when I started my career almost 30 years ago, it feels like that all over again. And there's something new every day. There's something new and exciting every day. And there's something crazy every day. And at the end of the day, though, this AI thing, it feels like a natural progression.

12:14All that AI really does is what we've always been able to do just faster and better. You've always been able to search Google. You've always been able to learn programming. You've always been able to do these things. And now the AI can just accelerate that at light speed. So what we are trying to do is work with the less informed people in the finance world, especially, and get them upskilled and get them some of the knowledge and just teach them how to use the tools, cloud code, co-work, some of these other things so that they can continue to learn on their own. but it's impossible even for me to keep up with new model releases and all this stuff that's coming out.

12:45And I think anybody who tries to, you end up just going down the rabbit hole, you get the AI psychosis, you know, and so you just got to do the best you can to stay on top of it and, and just find what works for your business and, and stick with that. But so there's, there's an interesting part of that though. So when you say, you know, Jason 30 years, so this is, to me, I've got a couple of theses that plug into this, right? Yeah. I'm 63. You know, my standard joke is they let me read access to any repo, but I don't have check-in privileges. They map it to dev null, so I don't hurt anything, Jason.

13:15But I'm a good architect. I've been doing this for many, many years. You're not as close to the code anymore. AI puts us all back close to the code. How do you evaluate the architects within the business? How do you evaluate that skill? Because to me, that's becoming the most critical skill in an AI first engineering organization is are the product managers able to engage at a level that can get real stuff in front of the engineering team? And is the engineering team architect set? Because that's what I care about, right? Like AI is great. What's the reason we don't do lots of test code coverage?

13:52Because it's, we say it's because we don't have time, but it's also like, no engineer wants to write that code. AI doesn't care, right? Like, in fact, it's happy to do it. Sure, I'll do it, right? It's amazing, but you have to know, Jason, what you should test. You have to, like, it can't know that. It'll just write a test for everything, and you'll have a zillion. So how are you evaluating that? Because it's so, I mean, you've got the right experience, obviously. You know, is that part of the, you know, this is a place where, hey, the engineering team's great, but they may not be great for what you need to do next.

14:25Yeah, that's one thing we've definitely seen is that engineering teams now either they go, they tend to go too far one way or another. Either they jumped on Claude Code and OpenClaw and all this stuff the day that it came out and they're on top of it. But they're almost so far removed from the coding process now that we've talked to a lot of companies. One thing we check for in diligence and that I would recommend everybody ask about is, you know, atrophy of your skills. you know, as cognitive atrophies, where you just get so used to pressing yes, approve, auto-approve, that you no longer read the code.

14:56And, you know, you realize, oh, I worked for hours, and I didn't write a line of code, and I couldn't even tell you what this agent did or this swarm of agents did. And so having some connection to the craft is still very important. Ask me in a year, and I'll probably say, ah, that you don't even have to worry about it anymore, but, you know, it's moving so quickly. But that's one thing that's very important, is making sure that the engineering teams, even if you have product managers and you have designers and you have pods of three-man teams working with cloud code rollout features, someone still has to be responsible for this.

15:27The insurance companies are not going to accept it when you say, oh, cloud to lead the whole database. Well, who is running it? So making sure that your engineering team has a plan to prevent atrophy. They're writing the test themselves. They're writing some code themselves. They're not just looking at the code reviews, scrolling and clicking approve is really important because I think that's where we're driving to. And a lot of these businesses, especially with cost cutting and headcount reductions and whatnot, is just whipping Claude into doing more and more. And we're producing more and more code that we don't understand.

15:52And as Mythos, as Kimmy K3, as some of these high-end models come out, they're going to destroy some of these smaller companies who haven't kept up with security posture, and they haven't done the real engineering work. So I would say my top hire right now, my most recommended person, is a graybeard. It's somebody who's been through this. They've written code, they've built systems, and they see Claude code as a way to make their skills faster as a force multiplier, not as a way to, well, we're going to roll out 50 new features at the same time. It's really like, this is going to help me cut out some of the slow parts of my process and accelerate the business that way instead of just doing all my work for me.

16:28So are you seeing the tension then between semi-technical CEOs who think we should be moving faster because they just believe in the AI hype and your CTOs and BPs of engineering saying, hey, this is really helping us. But to your point, I can't have this thing writing three weeks of agent code that we haven't even looked at. We can't take that risk, right? That it's making decisions that may be decisions we wouldn't make. It may be coding us into a corner, right? Because in my experience, if you keep telling any of these models to find more problems in your code, it never says, I can't find any problems.

17:09It'll keep just turning the thing back into, like some disaster, right? So I think you're right on finding somebody who has an architecture skill, who knows to think in advance to put frames around the AI to say, look, this has to be highly scalable. So you need to have a queue in here. You need to, you know what I mean? Really architecting it. That's important. I think that's where we see smaller firms potentially falling down as you're getting people who can get the feature stuff out fast, but they don't know the underlying architecture. So now we're saying we can sort of do that, But I'm saying, or I guess I'm seeing, and are you seeing, CEOs are still not giving us the breathing room to do a lot of the fixed-up work and test cases.

17:50It is a feature multiplier. Let's get more features out into the marketplace. Or can we do a custom version for every customer now because we're not writing the code? Are you seeing that? And how high is that in your red flag? Like, oh, boy, we could run into problems here. Oh, yeah. Yeah, we're definitely seeing that. I mean, we've always had a case of there's two types of CTOs. I mean, the term CTO has two definitions. You can either be the guy who signs the vendor contracts and reviews the quarterly numbers for the budget, or you can be the CTO who is the head programmer, the head engineer, and the person who is responsible for fixing all the problems.

18:23And there's obviously CTOs who are somewhere in between. Cloud Code and some of those force multiplier tools, they've become cabinet for the CTO who really was just the guy working on steak dinners with Microsoft and Salesforce. They now feel like, I'm a tech guy. I can finally do it. I don't need these programmers. you know and so they're going over yeah exactly they're going overboard on it i think and they're not worried about test coverage that was never a concern it doesn't drive ebita you know so we have those type of kind of it's it's like giving a toddler a handgun you know now they're just going crazy adding all kinds of features on vibe coding all weekend long and then we have the other end of the scale the ctos who are true engineers not really business guys who are maybe moving too slowly to jump in because they they don't want to be embarrassed by this they don't want to be you know, they don't want to get just like you said, we skipped over crypto.

19:12A lot of people see this as the new crypto. And in some regards, there are some aspects of the market that definitely smell like that. But I think you need to take a balanced approach to it. And as a financial sponsor, you need to be able to work with your target companies in your portfolio to really determine, are these guys true engineers who are using this technology to do things they could do faster, and they understand what they've built? or are they really just giving the toddler the handgun and saying, here, go play by yourself? Because we definitely see a lot of both and trying to inform our clients on exactly how to manage the situations has been a tough job.

19:50Yeah. And Jim, I think I would add to that, that when we look at the teams, we have to look at the relationship between the commercial leader, let's say CEO and then the CTO and what the motivations are. So, and I'll give you just examples from this week. so in this week we heard in a workshop someone was bragging about how 95 percent of their code is written by ai and jason followed up he was like great let's go look at your test coverage who's running the test cases and it was like crickets and jason slacks me behind the scenes he goes okay well whenever they don't have an answer that means that they have no automated testing and so in that that helps us understand that somebody is trying to solve a problem for somebody else you know and so they think the investors want to hear that more code is written by AI.

20:30They think their CEO wants to hear that. So I'll stop you right there, curve you hit there right before you finish. That is a number one thing we hear at this point is that it's sort of like the whole velocity is up, but we haven't actually shifted any software in the world in the old agile days, right? 100 % engineering teams are getting massive pressure to have high percentages of code generated by AI engines. So we were in a meeting yesterday with a really great engineering team. And we kind of know the financial plan that's going on where they feel a headcount's going to be flat. Pulled up the jellyfish.

21:08So jellyfish is a tool you can use to track productivity. It showed that they were going no more fast with AI. And also the code that was written by AI was taking 10 days longer to approve than the human code. And so there's some... Was that a technical problem in your opinion? you know, was it a process problem? We felt like they were running great. It's just more so what is the narrative? The narrative is like, hey, we're using AI to help us write code. Maybe it's better code, whatever it is. But when we say, okay, well, show us how you're tracking that. It's a good signal to us that there is tracking, but then there's a sobering signal, which says, okay, but if in the sponsor's model, they feel like they're not gonna have to hire any more engineers, but they know they need to build a bunch of integrations.

21:48They know they need to build this. It's like the data here shows us that this isn't magic, that it's helping maybe quality of life or you've got a junior programmer that's going to help you do stuff, but it's not helping the actual momentum of the team. So that's what the data kind of sits in between all of the forces and the narratives that are playing. That's a tough one. Yeah, I guarantee we're seeing that too. And you're right. But I think a lot of people don't want to hear that, right? Because in some ways, where we're seeing teams really successful, we've had teams that had very little code coverage.

22:18Our standard joke, Jason, is you go into any kind of due diligence, pre-AI anything. You say, yeah, what's your automated test coverage like? And they'll say, it's like 95, 96%. And you go, really? By the time you're done, it's like 30%, right? It's like flossing at the dentist. You tell the dentist you floss every day, but he knows you don't, and you know you don't. It's the lie that you kind of go with, right? That you might have actually really using AI well to fill in test coverage that was really lacking, right? Maybe you're trying to improve your DORA statistics behind the scenes as a CTO, but the disconnect with the senior management team and the investors, Kirby, is I don't think they've realized how much of a risk they were running with the low test code coverage before, right?

22:58So in effect, we've really improved the business, but I don't think the senior management team can see it in that model, right? It's kind of interesting because whenever Jason and I and the team are working with people on that AI quality, like that happened even before we do the full exit readiness. It's amazing to me how all of the first principles are things that were true before AI. are still even now the most true. And so quality was always true, quality and scalability. Now you just have to have a different way to prove it. We also think about money and engineers are always true, which whenever Jason and I are looking at things, we're interested in the unit economics because we want to know how much this AI is costing.

23:38And we're seeing the top engineering teams moving to self-hosted for the appropriate models because they can control the costs. And to me, that was always true as engineers were always going to find a way to make things cheaper and drive costs down with new skills. And the other piece is just things that are hard. Areas that you have to stub your toe, get a scar. You have to do a custom integration with an insurance company and you learn that they have undocumented paths. This is where we're seeing the most value created, where companies have a longer tenure, a lot more scars. They're like, yeah, we took all that learning and we put it into our machine learning models.

24:14And now AI just helps us write even more conditions. but you still have to go get all of those scars the hard way. And so when we're helping them prepare their stores, we're saying all the stuff that you thought wasn't important is now even the most important because people can't replicate that pain. So there's an interesting touch over here. So given, I'm going to steal, because he's done enough of these acquisitions, we're going to steal this expertise here, Kirby. So doing 30 plus acquisitions, Jason, at one company, right? We're looking at this at a lower middle market. If you're doing some kind of roll up, whether you're buying features or you're buying just customers and closing down that platform because all the hands on one platform.

24:50AI can be, I will never use the phrase game changer because I hate it. They use it for everything. But I might attack, I might go after a legacy company that I wouldn't have before because I potentially got AI to help me unspool that mess. As long as I have some people from the business that still understand it, right? Don't have to understand, super technical level. But would you agree that you might actually go after more of those given this tool set? Or do you think it's still the same and you shouldn't be any more aggressive than you were before? Oh, yeah, I definitely would. I think the things that were a challenge before or that were a deal killer or just added a lot of hair to the deal that you didn't want it to deal with.

25:26Legacy code, old programming language. Oh, this was made in PHP. This was made in something. And it was made by one guy who he's now gone or whatever. I've never heard that story before. It's Jim's garbage Pascal code that you have to deal with. Oh, dude, stop, Kirby. I've got everything. I've got assembly in the portfolio. I've still got Power Builder, dude. I saw that people could write COBOL, and I was like, perfect. Get off the COBOL. Yeah, but that's the point, though. I mean, if you look at a business now that was almost unsellable two years ago because they were based on some old thing, and it was going to be a million-dollar lift to try to overhaul their system, I mean, now it might be a weakened project, or it might be a nightmare, but the barrier to entry is much lower.

26:11And we've seen this already. There's been situations where, you know, we've looked at companies and they have some kind of incompatible tool or they have some program that was made in Python, but their main thing is made in Rails or whatever it is. And it's easy now just to convert between two things or to merge two systems much easier than it had been in the past. Because in the past, you have to go through, read all the code, rewrite all yourself. And now you have a good baseline. You have good code. You never want to throw away good code. So you have good code. It's just in the wrong language.

26:40You just run it through AI, translate it, you review it yourself. You have tests that you write to handle the process. But yeah, I would say that that is one good thing about the AI agentic coding, whatever you want to call it, this revolution that will be a game changer is it will open up a lot of these smaller businesses who, you know, good cottage little industries that were previously unsellable to be rolled up because now their software, their code, their proprietary thing can be normalized, if that makes sense. Yeah, I think where people get, so both ends of that are that it's simple and easy.

27:10Like when people say the weekend kind of thing. Now, look, you and I, all CTOs in the AI age, I can tell you right now, it's the main problem is divorce because, you know, I'm getting called from my, you know, like, why are you at your computer again? Right. Because you, and I've written more software that I've thrown away because you can, it's cheap to throw it. Hey, I wonder if I did something that did this. It's like, yeah, I don't need that. Right. That's a, that kind of stuff's a weekend, but real systems. I think, though, these smaller businesses that contain, it's much easier when we see that, hey, we're going to throw an AI engine at millions of lines of COBOL running under CICS for an airplane reservation system.

27:43I don't think we're there yet. Right. In fact, that probably needs to be rearchitected more than because I guarantee you there's not much the logic that's in there. You just don't understand. Right. But in these smaller pieces, I think it's true. And that's, what's going to be interesting is, you know, one saving some of these sort of older technologies that the customers love, but they're just so dated from a text tech standpoint or maintainability or anything else. It may provide my argument is always, we need more software, not less. And this makes it easier for us to build more software. I think.

28:16So if we dive in, let's, let's, I want to talk about the actual deliverable. Cause everybody always asks about, you know, like, okay, okay. What do I get? And this is always soft because it's not, you know, either people complain about, you know, it's just a PowerPoint presentation or it's too detailed. But I want to get a couple of things that we touched on that I don't think we, I want to dig into the answer. So when you say defensible AI strategy, I think that I've got a system of record or a set of logic that's really complex. That's the core of my system that AI unlocks value or use of that.

28:50is different. That's defensible to me, that I've got unique data, unique processes that AI can leverage is defensible. What are the typical, I think we're defensible, and you look at it and go, no, boys, all you are is a light wrapper around AI, and you're going to be engineered out of this thing by a tier three model released by Meta next month. Yeah. I'll go first, and Jason, you can correct it on is, to me, it all goes down to prove it. So how can you prove it either for good or for bad? And thinking about some work that our team has done just very recently is that if you have a proprietary something, you say this is proprietary, you have to then benchmark it against open source models, against LLMs, and prove that like, hey, I don't know what the state of the future is, but even today, I have to prove that our custom OCR solution that we've built or entity resolution is truly proprietary.

29:48And so to me, there's a, our smoking gun is if someone says proprietary, prove it. If you are deep into a vertical and it's not like a proprietary thing, but it's more of a, how hard is it going to be to replicate? It's more about how fast you can run and what accelerates you. So you have to prove that the more data that you get, the more customers you get, the more unique points that Shure models are actually improving with that. And so an incumbent, they could probably write the software in whatever time horizon you might agree on, but they're not going to be able to stay out his head as far as you are.

30:20And so these are two things that I think sit in the perfect category. But Jason, how would you add to that? Yeah, I would say we see both cases, right? We see businesses that claim that they have all kinds of cool proprietary stuff, and it's really just a prompt. It's a prompt, and it's a couple pieces of software that they put together. We also see businesses who kind of poop on themselves and say, well, we're not really AI. We don't really have that much. We looked at one where they didn't feel like that they were very AI. And when we looked into their system of record and into their database, we found they had over 25 ,000 human-written support tickets all about their integrations with the type of companies that they integrate with over the course of 10 years.

31:02And that's amazing training data. And we were able to work with them to basically come up with a plan. Like, if you have this as training data for an AI, you have something here. So sometimes it's really just coming up with that narrative, coming up with the story. And obviously, we're always going to get the companies. There's always people who are trying to pull the wool over on finance guys, right? There's always going to be companies who just started last week, and they say, oh, I have an amazing innovation in the insurance industry. And all it is is just a prompt that does something silly.

31:29But there's a lot of old school businesses out there that I think that with a little refresher, a fresh coat of paint, and maybe just a closer look at the things they have Stockholm syndrome for, you know, for them, this, this database of all these support tickets and all these integration history that they've built over the years. I mean, that's just work. That's boring stuff for them. But when you enable that with an LLM, now you really have something, you have a whole asset that they've been able to spin up. And it's just what the work they've done over the years that they almost were throwing away.

31:56As you said, you know, you've written more code than, or you've, you've thrown away more code than you use. I think businesses create more data and then throw it away than they use. And so that's one part of the process we go through is going through and looking what data assets these businesses have to try to see where AI can unlock additional value in the businesses for them and tease that out from the wrappers and kind of the cloud-coded, vibe-coded BS. Yeah, so we always look at like, look, if you have proprietary data, right? You have unique data. If you're a hard system for your customer to get rid of, then you've got some kind of a moat.

32:31Now the question is, how can you use AI on top of that? I think I've seen two angles that are not core in engineering, Jason, that I think are, because you lived through a long enough stuff here, I should get a laugh out of this, right? But business intelligence, okay? And user query tools, business objects, Brio, Cognos. In my opinion, those are all largely a failure that in the face of AI, because all we were ever trying to do is make it easy enough for business people to get to the data without involving engineers, which LMs can easily do, right? So that tends to be a first thing that we tend to look at is, hey, if you're not plugging against that data already that you have, you're probably missing something, right?

33:11Because any of our businesses, they have to have sales data. They have to have. So if your own internal team isn't using the data for just your operational systems, that's a great place to learn. The second part of that we see is every single software company has to have years of ticket data. And if you can't make something out of that with an LLM, I'm not sure you should be in the software business anymore because you have to be recording your tickets, right? And people are answering. Now, I get it. Some people are so bad. They never answer the tickets. They have years of tickets. You know, we've gone into businesses, but this is actually really funny because you know Nori Kirby on our team.

33:49Nori beat me to doing an analysis of a deal we were looking at. She pulled the ticket data out of JIRA before I got to it and ran it through an LLM. You know, I had a history of, oh, we've got a way to extract that data. You know, you've got ETL for that kind of data and Jerry had dumped it into a database. He was done before I could even start, right? Just because LLMs can do that now, right? So to me, that's, I think, you know, how can, when you walk in the door, are you seeing people that are like, I haven't even tried it against my support ticket data or I haven't even plugged it against our, you know, our internal CRM stuff.

34:20I mean, are we really seeing software companies at that level? That minimum level? Yeah, we definitely are. We definitely are. I mean, and you think about like GetHistories, you know, even before that, We had Subversion. We had CVS. I mean, you might have 20 or 30 years worth of change history for your software that doesn't matter to anybody now, but the LLMs can read that and they can understand the architecture decisions. They can understand why did they do it this way? Why did Jim change his mind about this? What was the conversation? And when you can enrich that with the emails, the tickets, the other things that we don't normally consider to be big data assets because they're not in the BI database.

34:58They're not in BigQuery. They're not in some kind of snowflake or something. But they're just messy, unstructured life of the business over the time that it's run. Taking all that and really taking a hard look at it and seeing what it is, that's a real asset. And something you just mentioned there actually resonates with me. We talk to, you know, as much as I like to crap on the young people, they come in, they use ChatGPT. They did all their homework in college. They graduated last year. And, you know, I prefer to hire a gray beard who's kind of taking the slow and steady approach. a lot of these younger people, the first tool they reach for is the LLM.

35:32And maybe for people like you and me who are on the backside of our career, it's not. And we're dinosaurs for that, you know? So I've seen the same thing where it's like, wow, I didn't even think about that. I didn't even think we should just use the LLM to check on this or to try this. And obviously, it's a fool's errand in some cases. It won't always be. I'm sure soon enough, context windows will be unlimited. You'll be able to upload that, you know, 50 gigabyte CSV file and get the insights out instantly and stuff. I hope I'm retired by that time by the way no you don't you're always going to work by the way if you're listening to this and you don't understand that last sentence that Jason just said you should definitely call code and code because you should know what that sentence means all of us should know everything in this day and age in AI for sure also if you're listening to this Jason has a beard he's the only one in the color of the beard he doesn't have much gray though it's getting there every call I have to go on with Kirby I get another gray hair we'll get you involved with a couple of portfolios That's tech debt on Jason's face.

36:28We can gray you up right quick, Jason. We got plenty of situations. When we think about terminology as well, what we heard a month ago, Kirby will echo this, every call we had, we're worried about a SaaS apocalypse. We're worried about the SaaS apocalypse. It was like everybody's worried about, oh, my business, my portfolio, my new deal that I'm looking at, it's going to get whacked by Claude. It's going to come, Claude Design came and killed all the whole design industry. And nobody says SaaS apocalypse anymore. What they're saying now is token economics. I think that a lot of these businesses that a month or two or three months ago, their CEO said, hey, we got to do more AI.

37:02And their CTO said, wow, I've been vibe coding all weekend. I need to get the whole staff to do this. Now they're getting their first bill where everybody's been running Opus and maybe Fable. And they've got a$100 ,000 AI bill. And they're saying, whoa, whoa, whoa, we need to talk about the token economics. You know, so, you know, just as much as context window is important to think about, thinking about some of the other technologies and some of the other pieces that I think firms right now they're touting. We talked to a lot of companies say we do custom model routing. We do this model routing to to make sure that we're not wasting tokens, you know, with everybody wants to be tight with their tokens and stuff.

37:35You're not doing that. You're you're you're not doing your fiduciary duty. It's like if I said, hey, I have a networking business and we have firewalls. great, you're supposed to. Model routing is like table stakes at this point. And so I think financial companies need to really inform themselves about some of these techniques and some of these tricks that have moved from the fringes of, wow, it's something we should explore, up to everybody's got to have this and make it part of their checklist when you're looking at a deal that you should be saying, if you're not doing model routing, you're leaving money on the table.

38:04If you're not using cloud code for your engineers, you're not getting their full productivity. And that's one thing that we obviously help our clients with is trying to get them the framework they can use to evaluate a business. But it seems like every day there's something new you have to consider. And it just goes with what you had said about, you know, the terminology is always changing, what you have to keep on top of. There's a lot to it. And this week it's token economics. Next week it's going to be Kimmy K3. I almost guarantee it. It's five days from now when we're taping that they're supposed to release this as open source.

38:31Game changer is going to happen. And a week after that, it'll be something else. So it's really tough to stay on top of, but it's important to try to do. But if you know the broad categories, you can, I think that's the point, right? And we're seeing some of our CTOs now say, look, we don't need leading-edge foundational models to help us with code generation and tests. So it's not just even how we're using different models as AI as part of our production product or in a production workflow. It's really as part of even engineering, modeling what is the right model for us from a cost standpoint, right?

39:04Right. It's changing every day because there's no way that the big foundational guys can keep selling dollars for quarters. They can't. Right. So at some point it's going to have to rationalize itself. And when it does, I think you're going to start to see some big bills come in for people who just aren't understanding what it costs to do some of this. So, Jim, I would say like when we do the AI quality as part of the exit process, those are the stories that need to be fit and proper. than with the data to back it up because it's going to be asked. And there will be new things even when this episode airs that will be asked.

39:39But for that team to have a foundation that they can stand on as a engineering organization, a CTO, and to be aligned with the CEOs to what the narrative and what the proof is, is the reason for the work that we're doing. And so they're going to be pushed in the process for like, you haven't looked at this, you haven't looked at this. It's about unit economics at the product level. What are you doing on this? And for them to be able not only to show what they have today to explain where they're going and then why and they have data to prove that to us really creates then that defensibility and it might be an internal um you know corpus they're building for just how their product works it can be unit economics for their most valuable feature that they've launched because they've proven that it's successful they need to optimize the cost these are the things that kind of pull into the bigger bigger narrative yeah i think it's I think there's so many moving parts.

40:27That's the challenge, right? That, and it's changing fast and there's a lot of moving parts. And this is, I, so at the end of the day, this is one of the places where you can't get a yes or no answer on even analyzing all that stuff, right? It requires, it's all subjective. It requires that it depends, right? Which is part of why you need this exit readiness. There's no, there's often not a super clear answer on some of these issues why you would do something or not something or, or, you know, we have businesses that say we don't ever have to worry about using anything with a foundational model because of X, right?

40:58What they're doing is so token efficient. Not a problem, right? But again, you got to prove that. Or you got to at least prove it on paper. And I'll use your favorite term now throughout the rest of the podcast, game changer. You know what a game changer is? I know you guys are. I shouldn't have said that. You know why? Because everything is game changing. You know, it's like our plastic bags have an extra string. It's game changing. It's a plastic bag. I think the game changer on this is whenever you don't create the risk of someone's narrative souring the overall understanding. And so you don't trust the lead engineer to describe something.

41:39You don't trust the CEO to describe something. You say, you know what we're going to do? We have this sim. Everyone knows it's all prettied up and it's nice. But we're going to have a separate document that says that game changing thing that we have. here's one line that talks about what it is and here's one chart that proves what it is and i'll give you the backup data so you know the chart's not hallucinated and then everyone's on the same page so whatever that asset is the assets you're putting together um when jason and i are thinking about what is success what is success is simplicity and cohesiveness which is this is the game changer and let me just prove it to you right here with this chart but sometimes it won't part of it sometimes you know i think kirby will be not every business has a game changer right like i could hit a bunch of signals and I can take a business like AI has made this software business better, but it's not going to be game changing for it.

42:26You know what I mean? I'm getting a 25, 30 % lift and that's good enough for this business. Right. And there are places where that's true. So I would challenge it. I would agree with you. Not everything has to be a game changer, but there are things that they can highlight to emphasize the thesis for who's going to be making that investment or who's going to be that acquirer. And I think about, you know, we're looking at we're working on a project right now where it doesn't sound like a sexy company niche software deep vertical whatever but their thing that needs to be emphasized in their exit readiness is they have built a really nice agentic development pipeline and they're a prime person to then have to do no more add-ons they've proven that they've done add-ons in the past they've incorporated them but now they don't need to go buy features they can build features and so as part of the bigger thesis for whomever that buyer is going to be when we put it together, we can show proof that they can continue to win inside of this deep niche and they can take on new incumbents that aren't going to have their customer base or the data that are collecting because of their footprint.

43:29But everyone's got their own little narrative. So I want to dig in something that's, again, this is a hold onto your seats, boys and girls, silly rabbits, segues are for kids, right? Because there's one piece of that that that i don't think we've solved yet kirby and this is the user interface for many applications is going to change as a result of ai right because you can literally yeah i find myself communicating with my claude sessions embarrassing like like i'm literally doing pair programming right yeah that you know in a sense you can sort of hallucinate yourself like it becomes your you know we used to engineers used to talk to ourselves jason now we of somebody that's actually answering back in a certain way, right?

44:09But this is a part that we're really worried about in diligence as an industry, that for plenty of applications, human text, actual speech, you know, are the real, the easiest way to do this. And there are plenty of places where that's not true, right? Like, as an example, I put together my own exercise app because I'm a lunatic fanatic and I had a way I wanted to do it. It's check boxes, whatever. And you would think the easiest thing is to just keep in track of your exercise, just talk to your clawed running on your watch. That's not the way I visualize it. Like, that's not right for me. But for so many other interfaces, it's real human text.

44:46It's talking. And that's, I think, one of the things we're most worried about trying to diligently prepare for is, yeah, the guy down the street can now, they got great agentic AI and they can develop a new screen really quickly. But the point is, the users of those applications never wanted a screen in the first place. It's just the only thing they have, right? What do you think about it? Are you seeing it? Or am I like really worried about something that's way, way, way over? Okay. I'll give thoughts and Jason, you know, challenge me on them. The first thing is, you know, my old product designer self would say, okay, well, especially in B2B, the best software is a software that you don't actually have to use.

45:24And so what can run in the background and I trust that it's working. And so sometimes the best interface is just an audit log where everything's green and nothing's red. So that's a component where you say, okay, well, what kind of automations are people setting up? And then how can AI actually help set up those automations? The other piece is that, you know, I used to work at SAP and SAP has got all of their kind of old legacy interfaces. And where that's going is allowing people to build their own interfaces. And so Jim's, whatever manic thing you've got going on for all your checkboxes is probably different than what Jason needs.

45:57And so a lot of people to have their own interfaces. I also think that I spent a lot of time in the developer experience space and it used to be beautiful API documents. Oh, we have this great API document, it's a perfect collection and now it's like, hey, can I integrate very quickly and then know it's going to work and maybe there's no interface there and so I would allow for all of those modalities to be true and just know that interfaces are now cheap and we were looking at a company last week and they were touting this great new UI I see actually two. And you know, it's the Claude UI. It's got the left bars, the whole thing.

46:31And so if you think about first principles, if you're going to waste somebody's time by making them look at something, great design was always true and great design is still true. And so that's kind of where design is also now at a premium if you're going to ask somebody for their attention. If you're going to allow somebody to all be different, then you can let it be flexible. But here's where you look at the two ends of this thing, right? We always said that if you, I'm not a pilot, but you look at a pilot's dashboard and it's like, holy shit, like, what am I looking at? Right. But to the pilot, it's the perfect user interface because everything's a touch away.

47:04Right. Like the learning curve for me would be impossible, but for them, it's like, they know where everything is when they sit through this thing. Right. Muscle memory. Even, even that, you know, the Maverick, you know, Top Gun Maverick is like, it's been a minute mad, but he still knows how to fly that jet. Right. That, that to, to any other user, So that's horrific. Now, to like a user who's using an application only occasionally, and a lot of people have to use it, that kind of lead wizard interface was kind of the way it developed. Like, yeah, it was a little slow for me, but I only do this once in a while.

47:33And I didn't want to know how the screen works, right? That, you know, and now this idea of I've got this great interface. Well, first of all, my pushback, Kirby, is if you are the system for everybody, right? So let's say you're SAP. And I heard the company's using SAP. We're a portfolio company. we're selling a software product that fits into a bigger environment that people are using, right? We're not the only interface that those people see like SAP. What did we do in those cases? We made it look like what else to look like. So when we owned a legal software company, everything we wrote sort of looked like Outlook because lawyers spend most of their time in Outlook.

48:06So it made sense, right? They probably trust it. It feels familiar. Right. And that's what they're kind of trying to do with Claude. Claude and the rest of these chat interfaces where I'm saying, oh, guys, we're thinking the wrong way, right? That the user can't think of the way that's easiest. That's part of our job, right, as engineers and designers is think of an easier way. And to me, the whole talking to the computer or agents running in the background to your dashboard, it's a completely different way of thinking. And I'm not sure we're not seeing too many people being ahead on that piece of it.

48:35Right now, it's still about generating better, faster interfaces. But they're not thinking about, you know, holy moly, are we even thinking about this the right way? Right. Are we going to end up like Scotty on Star Trek doing the whole computer? I think you could isolate it to the task because business is a team sport. So the systems that are used have to operate for the team. Now, how individual team members interact with that system, you could have variance in that depending on their jobs to be done. And I believe that's where the flexibility sits. Your point about good design, good design takes time.

49:05It's about removing things. And so whatever I can dictate with my Whisper flow to my Cloud app to build an application, that's cheap. But if you spend the true time to make something that's more streamlined, that's the kind of token for innovation. So I want to dig into that because that we've never diligenced, right? So you hit a really key point, Kirby, which is the cost of a prototype is cheap, right? And it gives me two things. It gives me something to hand directly to my prospect or my customer to say, I think we agreed on what we were doing, but is this it? And they can say, nope, because they're pointing at something, not a diagram, it actually works on the one hand.

49:38On the other hand, I can go down an alley to say, is this even something we should try and do? And I can figure that out pretty quickly, right? That is a skill that I don't think lives in a lot of organizations because it wasn't available. Yeah, you could do a Figma design or Balsamic, but it's not the same as like me as an engineer going, I can bang out a pretty good specification and let Claude go at it for a day and come back and go. So like literally, you know, the example I let you draft over is here because I had a bunch of health issues. I was tracking my medications. They have it in the Apple health app, whatever.

50:13Yeah. But I'm so stubborn I'm going to start myself. And the best thing was I realized I shouldn't be doing my own application and it's too complicated. But what that cost me, it cost me two hours of making myself learn health after what that I didn't have to do. Right. So are you finding the teams receptive to product managers kind of being evaluated as part of this readiness? And how do you do that? How do you tell whether somebody can do that or not? So I'll pick on Jason in this one because I think he's really good at it. That's not picking on him when you're complimenting him, Kirby. Well, okay.

50:48You're really bad at picking on him. Okay, I'll be sure to shave him in a different way. I'm going to run a picking on him seminar for you, Kirby. I mean, so I think, so I've led large product, product organizations. And I believe that the ability to visualize something is the old IDO adage, you know, a prototype's worth a thousand meetings. And so the ability to have a thing to converse over a working prototype is excellent. And I think it's speeding the product craft. It also does another thing for product managers, which it lets them go on a bender. They could just scope creep the hell out of something.

51:16And then whenever I would like say, first of all, call up Jason, Jason, look at this amazing thing. that's got all these kind of bells and whistles, like that old Homer car in the Simpsons. Me even explain, I go, you know what? Actually, we can cut half of this scope. Yeah, this isn't really the kind of piece. And so it allows people to kind of get it out and they can detox and get back to the core of the thing that they were wanting to solve. And then from there, when they leave, if they've done the right kind of setup, it aids in then spec-driven development. Markdown files, HTML files, things that they can give to a real engineer.

51:43And the engineer's like, thanks for the framework here. Now I'm going to go actually really engineer this in a way where the product can be true. so in product management where we're diligently one is the craft how'd you get to the conclusion faster which is a sport of communication within a team and the second one is and this is where I pick on Jason is we're always looking at the feature uses metrics to say of your building you built 100 % of the things but people only use 10 % of it do you know that and then if you put AI on it how efficient is that AI because that's the true thing that matters if you put more money into it, more skill into it, more users.

52:20That's the one area where your price is truly valuable. The rest of it is just hype in a sales meeting. Well, that's an interesting part, Jason, from you slipped into something dangerous here, Kirby, because part of it is when you look at that, we still as an industry haven't swallowed basically putting metrics on all of our applications, right? I guarantee you most companies don't have their front ends all pendo eyes or whatever for tracking purposes. But let's, you know, so now we're doing that and we're figuring out, you know, what features get used and not used. A lot of the time is because you built one thing for a set of customers, one thing for another.

52:55So it doesn't look like usage because it's, it's a, it's 10 features spread out of a 10 customers. Right. And as you start to rationalize that, do you think Jason, that we can as an industry start cutting off those prehensile tails of these applications? Cause historically, if we added in for a customer and nobody's using it, it still stayed there. It never left. You know, it was like a room in your house you never used, right? Do you think this makes that possible? Yeah. The most fulfilling thing to do as a programmer and engineer is delete code, especially if it was somebody else's code and you never liked it, you know?

53:27So, um, being able to delete code and remove old stuff is great. I think that the agentic development tools make that easier now because it's like, okay, nobody ever used this feature, find all the references to it, find everything about it and wipe it off the face of the earth. And it's helping us to cut down on some tech debt, helping us to clean up old code bases. That's a good thing. But I also think that, you know, it's okay to have features that only a few customers need if you have a reason for it, if you have a business case for it. And you shouldn't really leave the engineers in charge of that.

54:01Some of the businesses that we've looked at, they're doing the best, have adopted a new method where they have a pod of a product manager who understands the customer and understands the business case, a designer who has an eye for design, because even with CloudCode, as Kirby said, if you're a programmer and you don't have an eye for design, the stuff you make is going to look like CloudCode made it. And we can sniff it out in a heartbeat. We see it all day long. And then an engineer who can take the designer and the product person in the business case and harness that into something that actually works.

54:35And then they all work in the same cloud code sessions. They all work together closely. I think that's a paradigm we're going to see even more of going forward where you have the business guy who understands the customer, the design guy who understands that it should look good, and the engineer who understands how to make it work, and enable them to roll out features rapidly. And then not be afraid, like you said, to remove those. A lot of these businesses are using feature flagging tools. They're using analytics tools to track feature adoption. And so just like it's always been fail fast, I mean, get rid of the stuff that nobody uses, get rid of the features nobody uses.

55:05To go back to your earlier point, Jim, I'm old school. I don't like the chat either. And if I had, and we actually built a business that was AI tools just before chat GPT came out, marketing tools. And when it came out, it was like chat. This is, there was a, OpenAI admitted it was a weekend project that they had done this chat GPT on a whim. And we thought no one's going to want this. They're going to want to go, you pick your tool from the toolbox and you use the right tool. You don't want to chat with it. It's the most inefficient, the laziest, just the worst way to interact with software that I've ever seen, honestly, because all your pleas and thank yous and all your little messages and stuff, it's just wasting time and energy.

55:40But now we've had it for four or five years. We have a whole cycle of new engineers and of new people who that's the paradigm they're used to. And so I think as much as I would like to get away from it and go back to targeted tools, I think we're stuck with the chat paradigm. But I can't tell you, I mean, Kirby, every single week We see another company, oh, we just wrote out a cool new chat feature. Does anybody want this? Does anybody use this or do your customers really hate it? So when you're looking at a company, when you're evaluating it from a financial perspective, you should be asking them, show me the data, show me the adoption, show me the usage, show me the NPS scores.

56:10Prove to me that this thing you added is actually driving value. And it's not just you added it because it was an afternoon with Cloud Code and you were able to knock it out easy. And there's two recommendations that make it easy because PostSog, PostSog's for free. So you can just put post talk in there and you can put it as part of your agency development cycle just to be tagging all of your features. And the second one is in just to kind of build off Jason's chat and turn your RPGM is that anytime we see someone's using a chat, we say, what are they using it for? Go look at those traces and turn those into agents because then the agents are basically purpose built.

56:39You know, I want to get this outcome, make them the agents. And the agent library is a lot more valuable than a chat interface. That's a whole other topic we can talk about for another day because the tracking of those. So you get a lot of business people today that are using, you know, Claude or, you know, we don't see as many using cursor. They tend to be using either ChatGP or Claude or Codex or, you know. Good luck with the MD files, skill files moving around. I mean, none of that is secure. None of that is backed up. None of that is, I mean, we're all developers. I got a GitHub repo for all that stuff.

57:13Like, I'm not taking any chances. That's going to be its own other mess. But that's a story for another day with my own industry. Let's talk about one thing. So you make a key point about how do you buy yourself a year? So you talk a lot about, all right, if I'm behind, like, so we're doing this quality of AI and we're doing it before we sell, right? What do we do? What do I do if you consider me behind? What are the steps that I would be taking to say, at least let me start catching up or at least getting in the game from an AI perspective? Like, where would I start? And how do you evaluate that?

57:48tracking on all your features to prove that they're actually valuable. And, you know, one of the recommendations that we make is PostHog. There's also other tools for that. But hold on one thing that I'm going to challenge you. One thing is we own a portfolio company. This is a real story, right? We own a portfolio company that in the mobile days, they had desktop application. There was this, you know, real interest from some competitors had talked about mobile. And we had all of our customers banging to do a mobile interface, right? So we built it. Customers raved about it. We did put usage statistics on.

58:24It was the number one reason they bought it and no one used it. And they walked around going, it's awesome. It's on my phone. And so that's the hardest one of like, literally the customer said, we bought it for this thing. And it sounded great. And they loved that it was on their phone. They just realized they were always going to do it from their desktop anyway. Right. And that's one of those weird things where like, man, so tracking the value of that feature, like if I track, if I went based on usage, we would have killed that. Right. So you got to get your data correct. So if, you know, if we're in a situation and we're trying to buy somebody a year, it's about getting, we'll call them variables, getting all the variables in line and track.

59:03So then you can then line them up to tell the best story. So in this example, we'll play on it. You've got real features. so we're saying like hey we're looking at the data this being real we want to make sure that they're efficient so if they're ai enabled we're like okay those are the ones that go down the unit economics route buy yourself a year by putting tracking on the features and then also knowing understanding you know the cost if you have sizzle features that win deals which i love those features you have to then reconcile say okay great we interviewed 10 customers or 100 customers there is we're reconciling that they love that mobile app um and then you just own the narrative and we know that no one uses it so we've proven to you with like the you know our commit history or whatever it is that we don't over invest because we're not fools we haven't been you know deploying a bunch of expensive engineers to go on a fool's errand to build that so you build the right kind of narratives based on what the true situation is but if you don't know what your situation is um jason i would say telemetry on usage the why and the cost like that will buy you time Right.

1:00:01That's a perspective. Yeah. I think the weird thing in this case with customers is, you know, so the connection between, and we'd love to say that there's a complete feedback loop, right? And if you're some large, yeah, I worked at Oracle, that product managers upon product managers was the same product, right? And they covered every piece. We knew everything about the products. A lot of the small companies, like it's at the particular case, I'm thinking about colloquial or just on the back of our, you know, we knew that customers would tell us this, right? But there wasn't a survey that we did.

1:00:31It wasn't. So the proving it was harder in this case because it was just that, you know, you could point to the support logs, potentially, whatever. But sometimes you're going to run into a place where the customers just want it for a particular reason, right? Well, yeah. And pay them a hundred to$500 for 20 customer interviews and just validate that for yourself. You know, that's kind of where you do have to collect that information. I think the other - Yes, but is it true or is it not to like go back to the database wars, you know, Sybase started winning because they had a single server architecture, right?

1:00:59Did it, was it more valuable? Was it better product? Yeah. The customer said we want single server and my, for what? Like faster, better, you know, better transactions, whatever. In a lot of cases, they just didn't know. Right. Or it was, it's, it's like you could, you could never back it up. Right. Because on the other side, I was competing against them. Like, I don't know why this, they think this is a key feature. They don't even have row level locking for God's sakes. So there's that, that kind of stuff. Right. That I think becomes harder. There are going to be places, especially in AI, where sometimes the metrics are a little soft, right?

1:01:30And that's where I guess we're relying on your experience and seeing a lot of different businesses that this is enough data. It's good enough, right? So if metrics is one lane and even if they're soft, there's other lanes to pull. And so the next one is just being able to talk about the true health of your platform because that's going to be a question that comes up. You know, the amount of people that ask us, they say, well, what's the tech debt story? Because they know that they're going to be asked that in an investment committee. And so they want to have a clean answer on that. So what does it look like?

1:01:59there. So you obviously you can build a story around that with performance reports or whatever your narrative is to that. But if you look at them, you say, okay, I've got metrics, I've got tech, that story. And then the other one is just what are the stories that Jason called earlier this talk? What are things that you do every day that other people would think of value? And you're going to forget them. You're not going to mention them and you're going to skip over it. And these are all actually true assets. Individual assets can have their own valuation and you're just not going to highlight them.

1:02:28And so having an X, like an outside opinion to come in and be like, you know, one of the, Jason, I'm going to quote you. I hate flattery. You know, is that whenever you, you were in a call and you said, Hey, looks like you're doing some old school machine learning on that. And they said, Oh yeah, it is. You know, they're all bummed out because it wasn't cool. Sexy. And Jason's like, no, there's a lot of depth to that. You've built out this machine learning with all these business rules and things that you guys, we need to tell the story because machine learning is cool again. And that's a piece where they may just not know how to talk about what's actually truly valuable.

1:02:56Well, that's a really good point because when you think LLM, right? AI is in everybody's mind at the surface level market is LLM. They don't know that. But when they say AI, that's the only thing I'm talking about. There's all different kinds of machine learning. LLMs aren't what generates those images for you boys. That's a diffusion model. It's not even the same industry. That's the part that people just simply don't understand, right? They go, oh, it's AI. To your point, there are some different kinds. I mean, every credit card fraud system is a machine learning application. And man, that's got to be right, right.

1:03:28It can't be like hallucinate all over the place that, you know, you're right. Although you could actually say the new LLMs could help you generate better ML models. Like this is all starting to become this kind of crazy thing where, you know, it all molds together in some ways. Yeah, so I'll build on what we were talking about earlier or where you should point the LLMs and where it makes the most sense. And so if you're buying yourself a year and you're not pointing somewhere, it's a good time to start that to then build up that narrative. So you guys mentioned a few earlier. I'll just add another one.

1:04:01Onboarding integrations, great opportunity for AI-generated code. And then even in that machine learning model, if you're already building out machine learning, proving that you're able to now – you don't have to hire a bunch of data scientists or ML engineers now. It's like, hey, we can actually make this model better without having them kind of pay that premium PhD price tag for that. So these are things where I would – Well, Kirby, they made that mistake. I don't know if you've seen it. We've seen it. We've seen that mistake where somebody says, I want to hire a data engineer. No, you don't.

1:04:31You're not building your own LLM. So you don't need it. But I think a lot of people wasted their money. They've got a PhD. Now, I have a degree in applied statistics. In 1981, that was completely useless. Now I can just call it machine learning, right? Game changer. It was applied statistics, right? I think that is one of the biggest problems that we're going to run into on this go-forward basis. We've seen lots of people blow up by hiring a data scientist and wasting a year with somebody who's brilliant PhD, but it's more research. It's not really what I would call practical or usable AI, right?

1:05:05I think a lot of people have made that mistake. Yeah, we hired a machine. We hired a data scientist team and we didn't get anything out of it. So if you boys are ready for a lightning round, you ready? Let's do it. Jason has to go first. All right. One diligence document you never trust anymore on site. in the age of AI? Well, I don't trust anything anymore. You know, I think we always couldn't trust your disaster recovery plan because we know that you, we know that's still the board. Yeah. But I don't trust anything anymore. Everything we get is a summary made by Claude. It's all got the same cornflower blue.

1:05:41It all looks the exact same. So, so my, my model now is trust, but verify, you know, like, thanks for the summary. Give me the raw data. Give me the Git log. Give me the copy of whatever you got this from so I can run it through my cloud and we'll double check this. So I don't trust anything anymore. But traditionally, it has always been the disaster recovery. It's always been the procedures and stuff that nobody had. And the way to sniff those out is to say, hey, when you're on a call with the whole engineering team, say, oh, Jim, I know you guys have this engineering disaster recovery document.

1:06:13What is the hotel that you guys go to when there's a hurricane that hits and it's over a certain number of mile an hour winds? because it's in the document and everybody's supposed to know it. And they say they practice it every quarter and you just, you get deer in the headlights looking. It's always a lot of fun to do that. It is. We had a portfolio company in the hurricane zone, the Gulf side. The data center was a laptop and a desktop machine in a closet. It was pretty secure. That was a disaster recovery plan with somebody who would drive to the office and grab the laptop. All right. So that's one.

1:06:46All right. a prediction that the rest of the PE market will think is crazy when it comes to AI, but you guys would bet your business on. I'm calling the top here real soon when it comes to data center development and expansion, and hopefully in the RAM and HBM prices. I feel like with these open weight models coming out from China, that sadly, Anthropic and OpenAI, if they only would have IPO'd a month ago, they'd be in a lot better shape. So I feel like where we're at now is we're going to see a lot more democratized LLM stuff. We're going to see a lot less foundation models where you're having this huge token economics, you're paying less money to Anthropic.

1:07:24It's going to be a lot more decentralized. And as that happens, it'll hopefully drive down the price of RAM, the price of GPUs. You'll be able to run a lot of these models on your computer, on your MacBook or whatever, and we'll start to see a lot more diversity and less concentration on one or two providers that we have now. Now, I said that a year ago and it didn't happen, so we'll see what happens next. I think it is happening though. We've seen yeah you can say there was very little traces of it last year but we've seen a lot of traces of it this year um i have at least kirby would you would you have a different answer to that um i'm just going to bet the trends so what i would bet on you know with whatever the table stakes are here is that security is going to be super cool we're already seeing a trend right now people are cool so hiring engineers um to do quality work is valuable then also whatever humans in the loop can generate now is proprietary.

1:08:15So we do a lot of work with service companies that are coming online with more kind of backend systems, it backbone, but overall when humans interact with human and they create something bespoke, if you can save that now it's valuable. And so if I was going to bet then on top of that, data brokers are really going to be moving again because once proprietary day can come online, if you can sell it and selling as easy, then there's more value to be created to like train more models and do more things. So those are some of my spaces. I think that where you hit it is there is so much data that we still haven't pointed these engines at.

1:08:51That's going to be the amazing part. You had a post about something like ancient company that was, you know, valued because they had all this offline, you know, centuries of information or something like that. I was like, it's right on point. Where's stuff being accumulated that's not online yet? And then how can you value it? Yeah. I mean, and I think that's an unlock for the world too. Like literally, I mean, I hate that. It's not a kumbaya thing. It's game changer. It is. It is a lot of, well, it may not change any games, but at least we do a bunch of research. You know, we're going to have a war about this now, Jason.

1:09:19It's, you know, we've drawn a line in the sand. He's going to throw a game changer in every conversation from now on. I'm going to have to defend against this. So, you know, Kirby, Jason, what have we missed? This has been a great conversation. I love having our portfolio companies go through exit readiness long before they have to exit because you don't have time to fix anything. Back to your home inspection, Kirby. If you bring the home inspector in the day before, you know, or you let the buyer do the home inspection. Yeah, you're going to get a death of a thousand cuts. You have to paint this.

1:09:51You have to stop that, whatever. Like if I knew long before I sold the house, the things would be fixed or I'd have an answer to say, I'm not fixing that because it doesn't need to fix it. And here's why. Yeah, I would say that we didn't get to talk about it, but I would just, you know, encourage people to challenge their legacy thinking of what the process is. if you have an asset and you don't know if you should take it to market call for a second opinion even something light just to get a view that's outside of your own kind of silo and then you'll know hey do we actually want to go through X awareness do we think we're actually going to go to window or do we want to keep holding but right now if there's no clarity on that or if it's just an exchange of different opinions and narratives for everybody in the echo chamber you're just going to get stuck so I think that just having a second opinion is what I would encourage anybody to do is because there's money on the line you know for the team and the firm like let's get a clear view as to when is the right time to go to market based on some objective information not just opinions yeah and smaller companies i think part of the challenge is often they don't get to control that you know founder-owned businesses where right you know they suddenly have a life event you know hopefully it's a positive life event although i don't know deciding to buy a boat is that really a positive life event i don't think so but you know that gets to the point where hey i'm in my late 60s early 70s my kids don't want to run the business You know, sometimes the timing kind of, yeah, they can't control the time, but I agree with you that it's always good to, hey, you know, what's the market look like for this business?

1:11:14Because I do think to Jason's point, some small niche legacy businesses that have, you know, may not be PE backable, but have business successful businesses could get disenfranchised by a newcomer that can generate the application that they've long built, you know, in a modern stack, et cetera. We're at a point in time where maybe that moat isn't as strong as it was for a really small company. I was going to say, yeah, I would agree that now is the time, you know, the best time to work on thinking about selling your businesses today before you're ready to sell. And that's where we really help companies find success is looking at what they have now and preparing for the future.

1:11:53Nobody knows what tomorrow is going to hold for LLMs. No one's going to have the crystal ball. And so if we can look now, we can start to help you determine exactly what steps to take to prepare for any eventual future. And then I think we've worked with a lot of businesses who just, they thought they would never sell. They thought they missed their window. And after looking at what they had and looking at the pieces that they've got, it really kind of jazzed up, fired up the team, got them excited about some of the things they had that they'd forgotten about. And it's really fulfilling work for us as well.

1:12:23So we definitely like to help businesses not only try to time the market, but also just learn about the market. Because when you're a business owner, you're running a business, you're in it every day. You're in there all the time. You're focused on it. The last thing you're thinking about is, you know, what's going to happen a year from now? What are the latest trends in LMs? It's like, I got to get my numbers now. I got the PE guys riding me as they need a board deck made by next week. So for us to come in as the outsiders and to be able to look at a business and really play good cop, bad cop sometimes against the PE firm and work with the company, it adds a lot of value.

1:12:57And it makes everybody money in the end. That's why we like doing it. Yeah, I mean, I think, you know, back to at the end of the day on this particular topic. Yeah, I wouldn't have had a quality of NFT process because we could answer that. Well, we could. It's zero. um you know that this is one of those that folks you may not have you know cryptocurrency bitcoin nfts yeah yeah don't care this is one you can't miss right that's not going away so to your point there may not be a ton of data centers being built but llms we can't predict the future other than it's not going anywhere like these models are not going to be thrown away so you know we're going to have to build this into our businesses into our our workflow how we develop software and run businesses, whether somebody wants it to be that way or not.

1:13:41It's lead, follow, or get out of the way, right? Yep. Or that was the old Rossboro, was it the Dana Covey Rossboro? You want, I'll tell you about it. You want, I used to be able to do that so long ago now. Gentlemen, always a pleasure. Jason, how do they get ahold of you if they want to, if the world wants to reach out to you, how do they get ahold of you? They can reach out to us at codeandco.com and I'm sure that Kirby would be happy to set up a meeting and have us dig deep into your business and tell you what's good and what's bad about it. You see that that that was well done. Somebody who's done sales engineering, he handed it right over to the sales guy.

1:14:14He didn't try and step in front, which was perfect. He didn't give his home, his home email address and home phone. Well done. Kirby's eye started to twitch for a second.

1:14:25They can get directly to you, Kirby. Go to your billboard. They can go wherever you want, right? We got a great delegation. Delegation is an important skill to have, Jim. That's right. Absolutely. Got true leaders over here. Thanks, guys. We'll ciao for now. From the heart of Chicago to all over the globe, a couple private equity geniuses, they share what they know. They love to mess with technology where the future is sink or swim. It's the Private Equity Funcast with Devin and Jim. technology issues middle market peep back companies they bicker with each other and they don't take themselves too seriously it's not venture capital it's private equity it's the private equity funcast pour yourself a drink and have a seat