AI Revolution LIVE: Are We In A Bubble?

Hello everybody. This is the weekly AI Revolution call. I'm Bryce Smith and we have John Dio here who joins us every week. John is our resident AI expert and industry veteran. And I work at 10,000 Days Capital, which is a hedge fund and we manage a long only fund and we also have our Trading with Cody newsletter service. So happy to be here. Happy with you as always, John.

Absolutely. Good to see you.

So let's jump in. We haven't had this call, I think in a couple weeks now and as always there's seems like there's an endless stream of news in the AI world. Every time you think it's slowing down, it never really is. And so we put a few things kind of in the description that we were going to hit on today. I guess one of the things that kind of jumped out at me that's circulating right now is this rumor that OpenAI is raising money at a $500 billion valuation. Half a trillion dollars in the private markets is unheard of. And so I don't know how much money they're raising at that amount, but it's probably, you know, $10 billion or something. And yeah, so it's got, you know, one of the things we put in here is like, are we in a bubble? Is this is open AI really worth $500 billion? So what do you think, John? Well, do you think, do you think the rumor is true?

No.

Okay.

I mean, well, think back. It was last year and Sam Altman was talking about they were gonna need, I thought he said $3 trillion, something like that. He said they were going to need to raise $3 trillion and I don't think they did that. Right. Maybe, maybe 600 billion or something. Just some small number like that in terms of their big, their big project that they got going with Matsu from Japan and SoftBank and.

Yeah. Masa Sun.

Yeah. Master san. So first off, no, I, I, I, I mean I think that he kind of shoots high and, and then, you know, settles in somewhere. By the way, we are also haven't spoken since, since GPT 5 came out.

Yeah. With this. So yeah, I do, I think, I don't know that the rumor's true. I mean, I think the last round was at like 350 billion or something like that. That was actually acknowledged by you know, open AI and SoftBank and everything. But I mean it is, it does make it feel a little bit bubblicious to like toss these kind of numbers around for a startup, especially a startup that has like major corporate governance problems that Needs to work out. Still, most people probably know that OpenAI is technically a non profit that's supposed to make open source AI and instead they have a for profit subsidiary that makes closed source AI. And Elon Musk loves to give them a hard time about that.

They came out with some open source.

Yeah, they finally did release their first open model in like two years or something like that. I haven't, I haven't, I haven't played with it or I don't really know much about the open model that they released.

So they went from also 5 billion in recurring revenue in 24 to 10 billion in recurring revenue this, this year.

Yeah, that's even 50 times, 50 times sales, you know, for, I mean their revenue is growing double 100% year over year and you know, they have the, probably the best, I mean they're closing in on like a billion in users I think. Right. Don't they have like 700 million users on Chad GPT now?

Yeah, and I don't think that gets enough kind of credit. I'm more and more seeing that the kind of installed base of users really matters here because whatever, you know, you come out with, whether you come out with like they just did, you know, an integrated model that kind of combines the different things they've had before into one kind of one offering and everything that look at, look at how many individuals and businesses, you know, are affected and affect them with that. So yeah, it's a big deal.

And they really haven't turned on the monetization valve yet. You know, they have subscriptions but the overwhelming majority of those 700 million users are free users. And so those, you know, I don't if anybody's played with chat, I, I canceled our chat GBT subscriptions a while back when we went all in on Google's platform and but now if you're on the free tier of Chad GBT, you get like four or five questions on the GPT5. And then it starts hitting you with hey, you need to upgrade, you need to upgrade if you want to keep getting the best AI responses, you know. So I do think they're clearly, clearly trying to drive the monetization and start to convert. And I'm sure they are, I'm sure they're converting a lot of those free users into, you know, $20 a month paid users with this change.

Then they have their, I mean they're $200 a month version, which among developers the talk is that that's the only version that matters really what it's capable Of I know I don't have the $200 a month version.

So are you a, are you a pro subscriber though? Yeah, the $20 a month. Right.

I'm, I've been a pro subscriber for a while, I guess over a year. And I'm also pro subscriber, if that's what you call it, on, on Claude. Anthropic Claude. So those are the two that I use.

I was just going to share this for, for reference here to give some context to how big this number is. If they were raising money at 500 billion, you know, it would put them at the top 20 companies in the world by market cap. You can see here ExxonMobil 454 billion right now.

So, well, it's where all the new money's going, right? You know, like it investments on top of whatever was already happening is happening at 60 to AI or something of all the money going on out there.

John, are you worried like a lot of, a lot of the market, Wall street types are concerned that these companies are, aren't going to be profitable for a long, long time. And I think even open AI has admitted like they probably won't be profitable until like 2030 or something like that. He might spend a trillion dollars on data centers before then.

So like is that a long time? 2030?

Yeah. I mean if you gotta go, if you, if you're spending, if your burn rate is, you know, I think they're going to burn five or six billion dollars in cash this year is what they said. And so they're gonna burn, you know, 50 to 100. I mean, I would guess $50 billion from here or something, you know, before they reach that break even point. And so I mean, is it a, is it a problem? Like people like, you know, in the industry worry that, you know, open AI and anthropic, these guys, like, like if, if the bubble bursts between now and then and the spending stops on all these GPUs, does AI development just flatline from there or how do you see that playing out?

So look at that scenario. What would it mean for the bubble to burst? Would it mean that we have given up on trying to beat China? Right. Just from the level of the whole country and kind of our defense posture and like it, this is not, you know, like the, the, the, the, the Internet 20, what is it, 1998, 1999 kind of bubble where oh no, people are going to stop shopping or you know, E commerce is going to go. This is not that kind of, if it is a bubble. It's not that kind of bubble. Yeah, we can't lose, like, we, we can't, it's not one that we are, we can't afford to lose. And so however much it happens from the private side at the moment and so forth, I think that people will jump in and make sure that it kind of keeps moving along. Certainly there's definitely going to be some companies that, you know, went in too deep in particular areas, but I don't think so. So if it, if there's, if the bubble bursts, we have a bigger problem than just some people lost some investment money. Because one thing is you don't have to lose it. You don't lose it until you sell it. And, you know, Amazon Lost what, 90 of their value in 1999.

I guess.

Yeah.

The difference is though, John, like with the Internet, like the underlying, when the bubble popped in 2000, the underlying technology still largely kept moving forward. You know, like you mentioned, Amazon and Google and a lot of other companies were just, were still in their nascent stages, still growing, you know, getting adoption. But with the like. If the Internet didn't require hundreds of billions of dollars of capital expenditure on a recurring basis to work, you know, and for AI to keep improving, they have to keep spending tens of billions of dollars, you know, each company does.

Not to work, just to keep impro. Just to keep improving, you know, incrementally over each other in a competitive way and so forth. So if you didn't have that money being spent at the level it's happening right now, they would still function because they, you know, what's, what's happened has happened. And the, at least the, the train time side of things, you know, has occurred to a point. How long will it take for the technology we have today to be just not worthwhile, not adding enough, you know, enough technological advancement? Probably 10 years. We've got like 10 years worth of growth that can happen off of what is already there. So I, I just, you know, even so I can see again, like it bursting in certain areas probably. You know, the other thing is that a lot of the companies that are heavily invested in this have other business. This isn't all they're doing. Yeah, right. Google, it's, it's like it's been around, it's still around. Microsoft, it's like these, these companies, you know, exist anyway and have revenues anyway. Unlike again, you know, what was it like for, you know, Webvan or whatever popping up?

I was just going to address SK's questionary as far as Cody, he is sick and so he is not joining us this week. So send him your, your thoughts and prayers and he's doing all right, but he is, he's under the weather, so. And then we have a comment here from Craig. They haven't raised the price of their $20 subscription since inception either, but they did introduce like the $200 tier as John mentioned. And the, are the API prices going up, John? I don't, I don't watch those.

No, they went down.

They go, keep going down.

Yeah, I mean, they're still higher than, there's other, you know, other competitors out there with lower pricing, but as compared to their own, to their own stuff, it's, it's gone down.

Yeah. And I do want to circle back on what you're saying about the, you know, these companies having other revenue. Microsoft and Google, most importantly and Amazon. Right. So Microsoft, Google and Amazon are the chief and Meta are the ones that are spending them by far other than OpenAI far and above the most on AI. And they're doing it with their cash flow from their business. It's not like they're even having to go out to the debt markets, although that might eventually happen at this point. They're able to fund all of this spending purely with the cash flow. Other businesses.

Yeah. How much, how much, how much cash does Alphabet have? How much cash does Apple have?

Yeah, I mean, they, over. Over a hundred billion for each.

Over a hundred billion.

Yeah.

Yeah.

So it's like they generate like $20 billion a quarter or something.

Yeah. So I mean, and Apple hasn't even kicked in yet and they're like, they're often a, a late comer. You know, people maybe think of them as being innovative and maybe they're innovative on the, you know, the design side or something, but they didn't invent anything that they come out with. They just kind of made it better. They didn't invent the, the, the smartphone they just made just kind of took it over to the point where they're the Kleenex, you know, like the brand that people just name it by now, you know.

Right.

So they have an opportunity to. And they'll play it their own way. And I think it'll lean heavily on, you know, OpenAI, possibly others where they'll be like, you know, we're not out there trying to, you know, compete as the cutting edge language model. We're trying to be the best user experience.

Right.

And we're going to see. And so that whole game has just begun, I think Another aspect of it is the, we call, what do you call it, the kind of operator feature which means like getting it to control your computer, getting it to like whatever your computer can do. What if you could talk to your computer the way you talk to chat GPT and have it do anything that your computer does, like say, hey, can you go get rid of all the junk mail out of my. Right out of my email and do this and the other and blah, blah. And just because it could have control over those things, it's just, it's like the next wave that hasn't even happened yet.

Yeah, one of the chief complaints that gets lodged at Google is that they haven't done enough of that. Like I've seen people like, hey, Gemini should already to just be able to do anything across the Google ecosystem. You know, go, go get my file. I should be able to talk to my Android. Hey, send this spreadsheet to John. You know, he's in my email. Here's the name of the spreadsheet. It goes into Drive and sends it to you, you know, here with a question or whatever. And it can already draft emails. It can, it can work across Drive, but there isn't like a, there isn't a unified way to have Gemini just do everything for you. And I've thought about that a lot, but I think, I think there's a lot writing on Google getting that right. You know, like these aren't like experimental little products that a few million people use. Like these are products that hundreds of millions of people use every day. And so you don't want to put out an AI product on top of that and then like ruin the user experience of Gmail or ruin the experience of Drive, right? Like they have to you, you can knock them that they're not going faster. But on the other hand, like you can't. You, you, your risk if you do it and do it wrong is also somewhat high.

Well, see, I think what you're supposed to do if you're a large company and you don't want to get like out innovated by perplexity or whoever comes along is, is compete with yourself.

Right.

You know, and I think Steve Jobs did that for a bit. You know, he had, he'll have like the Apple II team going up against the Mac team and they're both operating and it's like, go at it. Who's going to do it better? But from a, from our side of things, what that might look like is don't mess with what's there, don't mess with Gmail, come out with another communication system. It probably still has mail as a aspect to it, but those who are kind of pioneers, you know, can jump on the new thing and they'll, they'll willing to risk kind of going over there. Yeah, maybe you only get 10 million people using that, but you're off to a good start. And by the time, you know, it's really doing well, you fade people over. But it's like compete with yourselves. Because this whole thing about this is where they get themselves in trouble, that Microsoft as well is where they kind of get caught on, held up by their own success. Oh, no, we can't do something. It might damage what we have there already.

Right. Yeah. I guess my point is just like there, there is some good business sense into why they are that way at times and, and then, but it does, it runs. You, you run the risk of disruption if you're not moving at the speed of, of light or if you don't.

Disrupt yourselves, you know. Right.

No, no, totally. We did one thing on the Apple thing is we, you know, we've had all of the, the quarter, the earnings reports from all of these companies now for the second quarter, and Apple did say that they're starting to like, move in the direction of spending a lot more on capital expenditures and building data centers and things. They have this concept that they call, like the private cloud, where your data can be sent to a private cloud that nobody has access to, even Apple. And then, you know, the LLM can work in that private cloud and then send the response back. But I mean, it does seem like Apple sort of being pulled in this, in this direction. Like, I don't really know that they have a choice in the matter, you know?

No. Right. I'm saying I think that they're gonna, they're trying to determine what role they're gonna play.

Yeah.

And I think they, when I heard about that, it sounded. Right, it sounded like the right kind of a thing to do because not only are we not there, the, the kind of, the, the arc that gets us there seems like it's not as steep. Seems like it's, it's going to be longer.

Right.

I wonder if ironically that's good for the markets because getting there so fast on such a steep, whatever where everything's getting thrown into, it isn't. People don't know how to, companies don't know how to make use of that. Whereas when you get, and you're, you can look ahead and go, oh, we have our new Moore's law. You know, it's a curve that we can understand and we can watch and it's definitely improvement still happening on a very rapid basis, but we can kind of watch it happen. And you know, it happened with this kind of thing with the, it happened with pre training and then it's happening with inference and, and the test time training and then you've got like agents and what's happening with those. So there's like stuff developing that people are learning what to do with and learning how to make value out of instead of being like, oh, you know, any day now we're going to get blown away by something that just takes it all away.

Right. No, you're, you're absolutely right. I think there was this like sort of fear that you have, like the AI takeoff scenario is what I've heard some people call it.

Yeah.

That, you know, the AI gets so good that it can start to improve itself. Right. And so the first company to hit that magic spot where the AI is good enough to improve itself, then it's just like a Runway from there. Right. And then nobody ever catches them. Every AI startup, every AI competitor, everybody else is taken out basically overnight. And this one company that builds AGI wins everything. Right. And I think, I think we can confidently say that that's not.

Yeah, I, I just don't see that happening now. And I think that GPT5 is, it was, the, the evidence there was, you know, was underwhelming and like I said, it's weird, but it's that, that may be a good thing there.

Even the idea that like, you know, they call these companies rappers, right? Like LLM rappers. Like, oh, you take something that chat GPT, did you build some kind of slick user interface that's, you know, designed, oh, this is the GPT for healthcare or whatever, you know what I mean? And people like, oh, that's not actually a viable business model. But I think, I think we're learning that there are, whether you call them wrappers or not, but there are a lot of applications built on top of large language models that are incredibly valuable and have a unique value proposition that the LLMs don't have. And in my mind, the, the best example of that for me is replit, that's the AI app builder we've been using and I think it uses CLAUDE as the underlying LLM. But basically they built an agent that can help you design software end to end. Right. And it can set up a database for you and set up authentication and everything. You set up your back end Server like everything you need to build an app. Right. And that's an incredible. I think Replit is Now like over $100 million in annual recurring revenue and I think it would technically fall under the category of a AI wrapper.

It's a tough business because they do kind of provide inferences well and then that becomes the really expensive part of what they're doing and the users are insatiable around that aspect. So it's, it's like thin margins so far.

Yeah. And if, and they may even, I think at one point the CEO of Replit even said they were subsidizing the compute, like selling it at a loss just to, you know, get user growth. And obviously that can, that can't go on forever, right?

No, that's actually a bad business model.

Negative gross margins are not a business you want to be in. Yeah, but yeah, you know, I mean the, the market, there's so much wealth out there today and there's so much money in the private markets and venture capital is so well established, private equity so well established. There's a lot of money out there that can, you know, help support these loss leading businesses as they scale. And so it's, it is like if you want to talk about bubble dynamics, you could say, okay, well that money might dry up. And then you know, all of this inference and all these trillion or like Google's doing a quadrillion tokens right now in a month. Like oh, you know, 50 of those tokens were subsidized and losing money. So obviously that's unsustainable. But even, even if that was 100 true, there's still a lot of money we could burn before.

Like, yeah, I, and I would worry more about not burning the money so much as not having something to burn it on. And there is just a lot happening in a lot of areas to go put the money into. Is there? Because there seems to be enough more capital behind that, you know, that can, can keep following it up for a while. And you know, the, the I think having good government support for the the for AI growth, both for the infrastructure stuff, the energy behind it, the data centers trying to kind of get the, get, you know, rapid permitting and so forth. Because we're talking about this, this growth is, it's physical, you know, it, it shows up as you know, Tesla's or Xai's new data center in Memphis and, and the new one down in Texas that OpenAI is doing. And these things, these places are huge and they're using up like as much Electricity as a small state. So having that kind of support is going to help as well.

And there is some compounding effect to that in the economy. It's not like that money just evaporates, right? That's real jobs. It's. It's steel, it's generators, it's pipelines. You know, like, there's a lot of physical infrastructure that's needed to do all this. And like, that's actually a huge tailwind for the economy. You know, if, like, I think next year, the estimate is the hyperscalers are going to spend around $450 billion on capital expenditure. And like, I looked up, I just wanted to put some context on that number. You know, that's more than the Treasury Department spent on the TARP program. If you guys remember 2008, the Troubled Asset Relief Program, where they went out and bought a bunch of bonds off the bank balance sheets that were worthless. Right. That was a huge deal at the time, was spending $450 billion on TARP. And now we have. We have four private companies that are doing that in a year.

What was the kind of the government spend on Covid on business during COVID Because they were doing a lot of. They were doing a lot of shoring up and making sure that while nobody's able to work, we're not going to let things go under and so forth. Was that comparable to what happened in 2008?

It looks like it's much larger. It says the federal government, this is from the Government Accountability Office, had spent around $4.6 trillion to help the country.

Recover from COVID And now we, I don't know what's real or not, but we supposedly have investments coming in almost on that level from Saudi Arabia, and it's kind of the Middle east and. Right. There's lots of, lots of places money's coming from. But I think this is, that's. That's examples of, you know, the world wanting to be able to play in what's happening.

Right.

And right now they've got.

The perception is that the greater risk is to underinvest and not be a part of this than it is to, you know, than it is to overbuild just because there's no signs like, you know, the risk is that, okay, we build, we turn all these data centers on, and then all this capacity comes online, and then there's no demand for it. But so far, every time we fire up a new data center, it's, you know, the demand for the compute is instantly there. Yeah, we're invested in a company called Core Weave. Some folks may be familiar with it, which is a. They call themselves the AI Hyperscaler, their direct competitor to Amazon Web Services and Google Cloud and everything. And the large hyperscalers are actually some of their largest customers. Microsoft Azure is their largest customer. OpenAI one of the two. They're related.

Well that's interesting because that puts them down as kind of the infrastructure. Under the infrastructure. Exactly. Yeah. They're the AI cloud.

Right, Exactly. And so like when they, they reported last week or two weeks ago and you know they're, they're, they continue to be capacity constrained. As soon as they can get a new data center fired up, there's a customer, you know, lined up to take all of that compute for the next and they're signing like five year deals. Right. These aren't like, this isn't like them selling it to the spot, what they call the spot market. Like right. Where you can just as an independent developer you can just go in and rent a GPU like this. COMPUTE is overwhelmingly being purchased by the largest AI customers in the world that I mentioned. And they're doing it for five years and they're even signing, they're even signing deals to purchase the old compute like so Nvidia is on the Blackwell system now, right? The B200. Before that it was Hopper. Before that was Amper or Emperor. I can't really say the name. The A's. But they said that customers are actually signing new renewal leases to rent those old a chips, the A1 hundreds that are now, you know, five, six years old. But those chips are still useful for inference. Right. You're not going to do a huge training run on those old chips. But you know, they're still signing. There's probably two or three years left of value just to run inference on those chips.

That's interesting, right? I mean instead of it just being that as soon as there's something new, the one before it becomes useless, it's like no, there's kind of a good healthy aftermarket.

Right? Yeah. And they actually said they would like to have spare capacity because it actually hampers like their sales pipeline a little bit because they would like to diversify beyond just selling to the big hyperscalers. You know, there's, you know, Morgan Stanley is a customer now in banks and other enterprises are coming to them and would like to rent compute, but they don't have the compute to offer most of them. And, and if they had actually some spare capacity they could really build out their sales pipeline. A lot more and kind of diversify the customer base.

Yeah, there's a interesting parallel subplot going on among developers or individuals that use these. The technologies. Like the $200 version I was talking about, like the latest kind of stuff.

Where the talk is jobs, it's steel, it's generators, it's pipelines. You know, like, there's a lot of physical infrastructure that's needed to do all this. And like, that's actually a huge tailwind for the economy. You know, if, like, I think next year, the estimate is the hyperscalers are going to spend around $450 billion on capital expenditure. And like, I looked up. I just wanted to put some context on that number. You know, that's more than the Treasury Department spent on the TARP program. If you guys remember 2008, the Troubled Asset Relief Program, where they went out and bought a bunch of bonds off the bank balance sheets that were worthless. Right. That was a huge deal at the time, was spending $450 billion on TARP. And now we have. We have four private companies that are doing that in a year.

What was the kind of the government spend on Covid on business during COVID Because they were doing a lot of. They were doing a lot of shoring up and making sure that while nobody's able to work, we're not going to let things go under and so forth. Was that comparable to what happened in 2008?

It looks like it's much larger. It says the federal government, this is from the Government Accountability Office, had spent around $4.6 trillion to help the country.

Recover from COVID And now we. I don't know what's real or not, but we supposedly have investments coming in almost on that level from Saudi Arabia and kind of the Middle east and. Right. There's lots of. Lots of places money's coming from. But I think this is. That's. That's examples of, you know, the world wanting to be able to play in what's happening.

Right. And right now the perception is that the, the greater risk is to underinvest and not be a part, not be a part of this than it is to, you know, than it is to over build. Just because there's no signs like, you know, the risk is that, okay, we build, we turn all these data centers on, and then all this capacity comes online, and then there's no demand for it. But so far, every time we fire up a new data center, it's, you know, the, the demand for the commute compute is. Is instantly There. Yeah, we're invested in a company called coreweave. Some folks may be familiar with it, which is a. They call themselves the AI Hyperscaler. They're a direct competitor to Amazon Web Services and Google Cloud and everything. And the, the large hyperscalers are actually some of their largest customers. Microsoft Azure is their largest customer. OpenAI one of the two. You know, they're related.

Well that's interesting because that puts them down as kind of the infrastructure. Under the infrastructure.

Exactly.

Yeah. They're the new. They're the AI cloud.

Right, exactly. And so like when they, they reported last week or two weeks ago and you know they're, they're, they continue to be capacity constrained. As soon as they can get a new data center fired up, there's a customer, you know, lined up to take all of that compute for the next and they're signing like five year deals. Right. These aren't like this isn't like them selling it to the spot, what they call the spot market. Like right. Where you can just as an independent developer you can just go in and rent a GPU like this. Compute is overwhelmingly being purchased by the largest AI customers in the world that I mentioned. And they're doing it for five years and they're even signing, they're even signing deals to purchase the old compute. So Nvidia is on the Blackwell system now, right? The B200. Before that it was Hopper. Before that was Amper or Emperor. I can't say the name the A's. But they said that customers are actually signing new renewal leases to rent those old a chips, the A1 hundreds that are now, you know, five, six years old. But those chips are still useful for inference. Right. You're not going to do a huge training run on those old chips. But you know, they're still signing. There's probably two or three years left of value just to run inference on those chips.

That's interesting, right? I mean instead of it just being that as soon as there's something new, the one before it becomes useless, it's like no, there's kind of a good healthy aftermarket.

Right? Yeah. And they actually said they would like to have spare capacity because it actually hampers like their sales pipeline a little bit because they would like to diversify beyond just selling to the big hyperscalers. You know there's Morgan Stanley is a customer now in banks and other enterprises are coming to them and would like to rent computer but they, they don't have the compute to offer most of them. And if, if they had actually some spare capacity, they, they could really build out their sales pipeline a lot more and you know, kind of diversify the customer base.

Yeah, you know, there's a, there's a interesting parallel subplot going on among developers or individuals that use the, these, the technologies. Like the $200 version I was talking about, like the latest kind of stuff where the talk is get in now and start using this technology to make money in ways that it hasn't happened. So not only just go start business off somewhere, but just even individuals going and creating value and so forth. Because you've got X amount of time left, you know, you've got two years left, three years left or something. Before what? Well, before this stuff does get too smart for you to be able to stay ahead and, and, and be making, making money with it and you got to get on that side of things where you've made the money and so that whatever, whatever comes next is like, okay, I've got that, got that in the bank. I'm ready to take on existing in this kind of post AGI world kind of thing. Because we still are facing kind of the technology getting so smart that it's going to cause some kind of inflection point in the way things work.

Right. I think as an investor though, like, it's hard to make that actionable in some ways because it's like, okay, well you can. So that means the only companies that are like investable are basically the ones that are working towards AGI, which is like Google OpenAI, Nvidia is not building it, but they're going to be the ones selling the chips to do it. So like, every software company is uninvestable, every other tech company is uninvestable.

You know, and I think it's that, it's not that I don't know about software companies that set them aside, but for many, many companies, it's not about whether they're the AI company, it's about whether they take advantage of it.

Yeah. You know, and it's, I mean, it's, it is like when we, when we were in Vail, Colorado at that investment conference in June that we go to, one of the guys presented on AI and it was an okay presentation, but you know, the, the best thing he said is that the second order effects of AI are going to be larger than the first order. And I, I totally agree with that. And so then we, we start, you know, we have a panel discussion, we start talking about what are the second or four order effects. And my point Was I'm like, it's the whole economy, right? Like every, almost every business, like, like you said, the tech space is a little bit different because you could be disrupted. But like the John Deere, Coca Cola, JP Morgan, right. These insurance businesses, all of these, the S&P 500X tech is going to be a huge beneficiary of the productivity gains from AI.

I tried out a technology that I think we may be getting incorporated into our product. So I want to get specific about the name of them, but they do, you know, a voice agent that is so good. I went and tried it. I was testing it out and using it in, you know, particular with a particular kind of scenario. And then I went and looked at it from like the scenario of it's gonna make phone calls that are lead generation, you know, it's gonna go find, find new business. And I was talking back and forth and saying, oh, just hold on a second. And I think I got this and what. But I think we're, you know, you know, not necessarily trying to throw it off, but just trying to be really natural about a conversation. It kept up with me across the board. Very good response time. You know, it tends to, and probably should, you know, say up front that it's an AI, but if you get over that hurdle, then you're on to like what's, you know, what's the value prop and why are, you know, are, are you interested? What are you, how are you taking care of this thing now? What would you do if you could do it this other way and so forth and you're just trying to get to a, you know, follow up scheduled call with a human where I, I got you a hot lead. This is somebody who actually might be interested or whatever. But all of that work on kind of, I'm forgetting the word right now, but no qualifying, you know, all the qualifying can go on. This stuff is amazing. And there are companies starting to use it at scale, you know, like they've got it doing, making a thousand calls. And it's not spam, it's on a very particular topic and you don't want to talk, don't talk able to provide the value. It's incredible.

So that's. I've been doing a bunch of work on a company called Twilio. I don't know if you're familiar.

Yeah, I used to use Twilio. This is somebody else. But yeah, yeah.

And basically they build APIs which are application programming interfaces for folks that aren't familiar, basically. John, correct me if I Say this wrong. The APIs are sort of how software talks to each other. So anyways, Twilio makes a set of communications APIs and they basically abstracted the complexity of the telecom world and text messaging and voice and video calls and all these things. And so that if you're a developer building an application, you can just tie into Twilio's APIs and if you want to make phone calls, if you want to send your customers text messages, you know, most of you probably experience like the two factor authentication, right where you sign into an app, you get a text message to your phone that all, you know, Twillio overwhelmingly runs that market. And I think it's a really interesting investment here because all of these AIs that are in all these apps that are being built, like there's going to be some incredible voice AI like you're talking about. And I still think if you build an app and use voice AI, Twilio is still the best way to get that voice AI like to actually make phone calls and that people, you know, can talk to.

Well, I think that's true infrastructure wise but you know, I, I haven't seen what they do for keeping up with the actual, you know, the voice agent, you know, being real time and very believable conversation kind of thing. Maybe, maybe they're keeping up on that front.

They're not, they're not building their own voice agent, but they're, they're a layer underneath the, the voice age. Maybe, maybe they're on top of the voice. I'm not sure what the right way to say.

Yeah, it may be that these other folks use Twilio underneath what they do and that makes it a, you know, another one of those, like a good infrastructure investment. Is Twilio public?

Yeah, Twlo. It's about, it's about $105 a share. It's, it's actually not that expensive for software Companies trading like 19 times forward estimates and growing, grow. You know, I think the estimate right now is 10 revenue growth or something the next year. But I think there's, there's upside potential to the revenue estimates if some of this AI stuff starts to them.

I use them for years. I found them to be very reliable and doing a good job.

Yeah, I'm really excited about the software infrastructure side of things, even building apps on replit so it can build the app, it can set it up for you. But if you want data. Okay, I still got to go get an API to a data source. Right. And if I want like the phone call thing, right? If you want it to be able to make phone calls, you got to go get APIs from Twill. Like there's all these, there's still all these like third party. The way somebody said it is like very few applications are monoliths. Right. Almost all of them tie into a number of other applications. And so if you are one of those like pieces of the software plumbing, I think there's, there's going to be even more demand for that as the number of AI applications skyrockets because AI is making the coding itself so much easier. But I don't think somebody is going to like, I don't like. Why would a company go build its own like Twilio platform where you can talk to all the, like. Twilio has all these agreements in place with the telecom providers. They have data centers all over the world so that they can what you call points of presence pops so that they can deliver the service very quickly. And you know, in the United States we have basically three telcos, but globally there's hundreds of these things. And so if you're a developer and you are working on a global scale, you don't want to have to go set up software so that you can talk to Deutsche Telekom and British Telecom and AT&T and Korean one. Right? Like that just, that's just not that useful. That's not a very useful allocation of your time and resources to set all that up.

So when is, when is NIA4J planning an IPO?

Who's that?

So they are speaking of infrastructure, I use them. They're Graph Database company.

Okay. And say the name again.

It's a weird name. It's one word. N, E, O numeral 4J. Neo for J.

Okay.

And Neo 4J is. They're based in somewhere, Scandinavia, I don't know, Sweden, Finland, something. And they, there's talk about them having an IPO and they currently I think have over a 2 billion valuation. And they're the, they're, they're a very important ingredient in a lot of the AI systems that are being developed that combine a language model with a, a database behind it. Where the database behind it is not your traditional kind of tables like a relational database like Oracle or you know, an SQL database. It's not like that. It's like a network of connected concepts and it's very amenable to the systems that get built with LLMs.

So can you, can you explain like the different types of databases and how you would use like a Like I know there's like what Post, post gray SQL and there's.

Postgres is just like MySQL or Oracle.

Open Source standard I guess or it's.

The, it's all that, all of that stuff is. This is standard SQL databases. SQL databases and they're.

Would you call that a relational database?

Relational. Relational databases, yes. And they're the picture like an Excel spreadsheet, just like a grid of data but not like where stuff's just scattered all around on the sheet but like a clean. Yeah, dense filled in grid where every thing across the top, each column is like name, address, phone number, whatever.

Right.

And each thing down the side is John Doe, Bob Smith, Mary, whatever. So each row is a record is a different like set of information about one individual and each column is. Here's all the phone numbers, that kind of thing. So that chunk and then having typically a couple hundred of those, right. Different tables all kind of hooked to each other to store the data that's needed. Each kind of subject like contact information that or you know, employees, you know, that's one. Projects, that's one. Any, you know, companies, that's one. So each one of those is its own and the rows in it, although you think of it like a spreadsheet actually those rows will go for like a million, I see to be like a whole bunch of those rows and those columns might go for a couple hundred columns.

So it's not something where people are manually in there like inputting all that data either. Right. It's all flows from like data pipelines and things.

It's typically. Well you can have data flowing in from pipelines and so forth, but there is a lot of manual stuff happening not directly to the table. Most of the SaaS applications out there are just a UI, a user interface on the front end of the. One of these big databases. And every time you get on a screen that's got like a form with some fields on it and you hit like add, Enter, whatever kind of thing you're updating one single record and this and, and the related stuff. And that's. So that's the way it's been since the 80s.

That's what they call, that's what they call crud, right?

Yeah. Well, crud is create, read, update, delete. So that's like what can you do to one of those tables? What you set up doing is I'm going to create a new record, I'm going to look at it. By reading it, I'm going to change something in it to update it and then I'm going to delete it when it gets. So that's the life cycle of data within one of these things. But really so much of the applications out there are really boring. They are just the right chunk of data that was needed to get a certain thing done. Think of like a Redfin, you know what's going on there? Well, they've got millions of records where each one of those records is a listing and then off from that listing are other tables that handle all the pictures and all the maps and all the whatever, all the, you know, allowable values for how many bedrooms there can be or you know, every little bit of it. So that's just all the data and then you've got, well, what does that look like to the user? And something gets put in front of them. So when AI came along, the latest kind of AI, you know, the open AI and the chat GPT and all this kind of stuff, this is the first new thing to happen in 30, 40 years. That's really interesting and new. Yeah, there were other things, you know.

The cloud probably should be thrown in there. Right. The shift from on prem to the cloud.

Well, even when you got up on the cloud you were still, what were you sticking on the cloud? A bunch of crud databases.

Right.

So you really, you know, you changed how it got maintained and where it got put and so forth. But what was there real innovation in what people were actually getting, you know, what they were able to work with. Not that much. Then you had, throughout the 2010s you had the first kind of wave of AI with all of predictive analytics and learning stuff, machine learning, prediction machines. And there it was like now that database we're talking about is still there, but some of the fields in it are now getting filled in in a smart way predicted by the machine. But Now, I mean ChatGPT and all the LLM stuff, large language model stuff has just changed things absolutely completely. It's, it's the difference between, you know, orange juice and, and rocket ship. It's like there's nothing to do with each other.

So do. And an LLM can very accurately tell you that orange juice and rocket ship don't usually appear next to each other in a sentence.

Well, they'll find out that the really thing they have in common is Tang, which was an orange drink that the astronauts had. And you know, it'll go find that out. But like things to do with just the world that we live in and being able to, you know, you set aside for A minute. Well is it really smart and all this kind of stuff and certainly no it's not conscious and who cares but like it can really hit on everything in the world that we go and deal with. So it's a, it, it breaks it wide open.

So go to go back to the database thing. This is a little technical folks but I do, I think it's, it has big investing implications because there are companies like Mongo Database and those kinds of things. Supermannual recurring revenue and I think it would technically fall under the category of a AI wrapper.

It's a tough business because they do kind of provide inferences. Well and then that tell me now.

With the AI stuff how has it changed how you use the database?

So that becomes the really expensive part of what they're doing and it's the users are insatiable around that aspect. So it's, it's like thin margins.

Yeah.

So far.

Yeah. And if, and they may even I think at one point the CEO you know about hallucination.

So these language models, they can be really smart but when they're stupid they're going to rat sound really smart. And that doesn't work for businesses that rely on accurate information.

Replit even said they were subsidizing the compute like selling it at a loss just to you know, get user growth.

So by themselves they have not solved that problem. They still have they and obviously that.

Can, that can't go on forever, right?

No, that's actually a bad business model.

Yeah, negative gross margins are not a business you want to be in. Yeah, but yeah, you know, I mean.

They'Ll keep pushing on the frontier of it to be like that's not making as many of those kinds of mistakes.

Yeah.

And also the information that's in the language model is like the entire Internet but in compressed form where it can't decompress it back out to get it word for word of what it the market.

There's so much wealth out there today and there's so much money in the private markets and venture capital is so well established. Private equity so well established. There's a lot of money out there.

So it's not like knowing everything. It's having good sort of common sense about everything. But when you want to get back to actual information that can, you know.

Help support these loss leading businesses as they scale.

And so it's about lawyers, you know and they want to be able to say I want to actually say what was the case that I can bring to use on, on this thing? Well Then you better have access to the actual.

It is like if you want to talk about bubble dynamics, you could say okay well that money might dry up. And then you know, all of this inference and all these trillion or like Google's doing a quadrillion tokens right now in a month. Like oh, you know, 50% of those tokens were subsidized and losing money. So obviously that's well full data which.

The language model does not have. It used it the full data to create itself, but it doesn't have it in it. So now when you get into this world of AI applications, they have to bring those pieces together the language model or the generative AI together with sustainable.

But even if that was 100% true, there's still a lot of money we could burn before.

Like I and I would worry more about not burning the money so much as not having something to burn it on.

Yeah.

And there is just a lot happening in a lot of areas to the real data that which is in the form of all the laws that go put the money into. Right. Is there because there seems to be enough more capital behind that, you know, that can. Can keep following it up for a while. And you know the, the. I think having good government support for in. In. In the world or in the country or so forth or all the data in my company's database and everything like that. The. The for AI growth both for the infrastructure stuff, the. So for it to bring that stuff together. Well you actually need a really good database to work with the LLM. And what our company has found is that energy behind it the data centers trying to kind of get the get, you know, rapid permitting and so forth. Because we're talking about this, this growth is. It's physical, you know, it. It shows up as you know, graph database. And not, not just ours, but lots of them have found that the graph database is a great way to go for that. You know, Tesla's or Xai's new data center in Memphis. And then so what's been called rag retrieval augmented generation that goes and gets some. A new one down in. In Texas that OpenAI is doing. And these things, these places are huge chunks of text that it can pull in and use for the LLM. Now more popular is graph rag and they're using up like as much electricity as a small state.

Yeah.

So there's. Having that kind of support is going to help as well.

And there is some.

And taking a graph database and being able to use it to get the right stuff back into that LLM and.

Make compounding effect to that in the economy. It's not like that money just evaporates. Right. Like that's real jobs, it's steel, it's generators, it's pipeline.

Get as smart as it can be.

So does do you know if what. What type of database Mongo do you know? Like there's a lot of physical infrastructure that's needed to. To do all this and like that's actually a huge tailwind for the economy. You know if like I think next year the estimate is the hyperscalers are going to spend around $450 billion on capital expenditure. And like I looked up DB does I think they have a. No, I'm just looking at that. It's a no SQL.

No SQL. Yeah it's called NoSQL database and it's a property value database. And so it's another kind and it gets used a lot. Where they're really strong is in projects.

I just wanted to put some context on that number. You know that's more than the Treasury Department spent on the TARP program. If you guys remember 2008, the Troubled Asset Relief Program where they went out and bought a bunch of bonds off the bank balance sheets that were worthless. Right. That was a huge deal at the time was spending $450 billion on TARP and now we have. We have four private companies that are doing that in a year.

What was the kind of that need to go ramp up very quickly and need to push. Put some kind of storage on their back end. It's good for the government spend on Covid on business during COVID because they were doing a lot of. They were doing a lot of shoring up and making sure that while nobody's able to work we're not going to let things go under and so forth. Was that comparable to what happened in 2008?

It looks like it's much larger. It says the federal government.

It's documented oriented data.

That's one of the things that they're supposed to. I think. I think Mongo database does this. Well is that like if you're dealing with unstructured data. Right. That's not neatly in columns and rows like if you just have a random invoice or a picture. This is from the Government Accountability Office had spent around $4.6 trillion to help.

The country recover from COVID and now we. I don't know what's real or not.

But things like that aren't right. Then, then you use a no SQL database to store that kind of we.

Supposedly have investments coming in almost on that level from Saudi Arabia and it's kind of the Middle east and Right. There's lots of, lots of places money's.

Coming from structured data is that.

Well yeah they kind of ended up getting no SQL to mean something like Mongo when I've always taken it to mean not SQL. So like I would put the graph database in as nosql but I think this is, that's, that's examples of you know the world wanting to be able to play in what's happening.

Right. And right now the perception is that the, the greater risk is to underinvest and not be a part, not be a part of this than it is to you know, than it is to over build just because there's no signs like you know the risk is that.

As well it's not a SQL database. But yes there they are document oriented for a bit there you had kind of this, they call it a vector database but there's not much database there.

Okay. We build, we turn all these data centers on and then all this capacity comes online and then there's no demand for it. But so far every time we fire up a new data center it's you know the, the demand for the.

You're kind of just having chunks of text and they're getting indexed in a.

Way that's probably too compute is, is instantly there. Yeah we're invested in a company called.

Coreweave something technical to talk about but that technical way of indexing them or.

Like semantics may be familiar with it which is a. They call themselves the AI Hyperscaler, their direct competitor to Amazon Web Services and Google Cloud and everything. And the large hyperscalers are actually some of their largest customers. Microsoft Azure is their largest customer.

OpenAI, one of the two based indexing is now being used across the board. I'm sure Mongo's using vectors related. Well that's interesting because that puts them down as kind of the infrastructure under the infrastructure.

Exactly.

Yeah. They're the AI cloud.

Right, Exactly. And so like when they, they report.

The graph database is using vectors it I'll just say it's like picture again the spreadsheet but where a single term anything pogo stick you know any, any.

Term event last week or two weeks ago and you know they're, they're, they continue to be capacity constrained. As soon as they can get a new data center fired up there's a customer you know lined up to take all of that compute for the next and they're signing like five year deals, right. These aren't like this isn't like them selling it to the spot, what they call the spot market, like right. Where you can just as an independent developer you can just anything, whatever is.

Represented by like a thousand numbers go.

In and rent a GPU like this. Compute is overwhelmingly being purchased by the largest AI customers in the world that I mentioned. And they're doing it for five years and they're even signing just for that one concept.

And those numbers, there's math that can be done on it so that jump and pogo stick end up with numbers that are like close to each other.

Okay. So sort of the way that LLMs work, right. And they're even signing deals to purchase the old compute. Like so Nvidia is on the Blackwell system now, right. The B B200. Before that it was Hopper. Before that was Amper or Emperor. I can't really say the name the A's. But they said that customers are actually signing new renewal leases to rent those old A chips, the A1 hundreds that are now, you know, five, six years old. Just kind of that same vector, relational.

Vectors like in the vectors like in.

But those chips are still useful for inference, right? You're not going to do a huge training run on those old chips but you know, they're still signing. There's probably two or three years left of value just to run inference on those chips.

That's interesting, right? I mean instead of it just being that as soon as there's something new, the one before it becomes useless, it's like no, there's kind of a good healthy aftermarket, right?

Yeah. And they actually said they would like.

To language models and LLMs.

You know what interesting. John watched the interview with the anthropic CEO and I got to say I didn't really like him going into it. But after I watched his interview, he came off very likable, knows his business. He seems like just kind of a quiet have spare capacity because it actually hampers like their sales pipeline a little bit because they would like to diversify beyond just selling to the big hyperscalers. You know there's you know, Morgan Stack confidence. Like he's not really worried about all these things that everyone else seems to be worried about all the time with AI. And Lee is a customer now in banks and other enterprises are coming to them and would like to rent Compute but they, they don't have the compute to offer most of them. And if they had actually some spare capacity they can really build out their sales pipeline. A lot more and you know, kind of diversify the customer base.

Yeah. You know, there's a, there's an interesting parallel subplot going on among developers or individuals that use the, these, the technology.

Yeah. One of the things he was talking about was like.

Dario Amade.

Yeah, Gosh, where was I going with that? It has to do with the vectors. Gosh dang, I can't remember what I was saying.

So you saw him, you saw an interview with him and was he talking about the vectors?

Yeah, it had to do with what you're saying. Anyways, it was good, it was good.

Like the 200 version I was talking about like the latest kind of stuff.

Where the talk is interview and like, I don't know, it makes me think like the, like I think about Oracle even and like that is.

Get in now and start using this technology to make money in ways that it hasn't happened. So not only just go start, you know, business off somewhere, but just even, you know, individuals going and creating value and so forth.

Product has been around forever. Right. The gorical database was in the 70s, I don't know, 80s?

Yeah, yeah, I think of 80s, but it could have been as early.

We still use a lot of.

Because you've got X amount of time left, you know, you've got two years left, three years left or something. Before what? Well, before this stuff does get Oracle databases.

Right. I mean, I don't know. That's a rapidly growing part of the market, but that's still a huge business.

That'S out there, too smart for you to be able to stay ahead and be making, making money with it. And you got to get on that side of things where you've made the money.

And just because like these new things are coming along doesn't necessarily mean that the old way of doing things goes away either. Right.

I think of it as inertia.

Yeah.

You know, I mean it's not gonna go away quickly. They have a ton of customers and they. And it's another reason connected back to that. Whatever, whatever comes next is like, okay, I've got that, I've got that in the bank. I'm ready to take on existing in this kind of post AGI world kind of thing because we still are facing kind of the technology getting so smart that it's going to cause some kind of inflection. We're saying earlier about having the installed base is so important. That's one of their biggest assets is their pinpoint in the way things work.

Right. I think as an installed Base. Yeah. And I do think that, like, when you're talking about. Because they're really focused on working with enterprises, I think is what the. Where I was going investor though, like, it's hard to make that actionable in some ways. Claw, he's taking anthropic, like, almost not. I don't know that they're yielding because it's like, okay, well, you can. So that means the only companies that are like investable are basically the ones that are working like the consumer space or like giving that to OpenAI. I still think they would love to compete in the consumer space, but towards AGI, which is like Google OpenAI. Nvidia is not building it, but they're going to be chat. ChatGPT as an app is the consumer app for AI, and it's orders of magnitude larger than Claude's app or Grox app, despite Elon's threatened loss of Apple. And he's just talking about, like, when you're dealing with enterprise, the ones selling the chips to do it. So, like, every software company is uninvestable. Every other tech company is uninvestable, you.

Know, and I think it's that. It's not that I don't know about software companies that set them aside, but for many, many companies, it's not about whether they're the prizes.

He's like, okay, I'll go meet with the CEO. And they're like, yes, I'm in. We would love to do a deal.

With you, EAI company. It's about whether they take advantage of it.

Yeah, you know, okay, let's make it happen. But then he's like, then I have to deal with like a hundred thousand employees at this enterprise and somehow get them to use Claude and do it in a way that is complementary to their existing workflows and stuff. And like, that's challenging. Right? There's like, there's a lot like they have, they have. And it's, I mean, it's, it is like when we, when we were in Vail, Colorado at that investment conference in June that we go to, one of the guys presented on AI, and it was an okay presentation, but, you know, the, the best thing he said is that the second order effects of AI are going to be larger than the first order. And I, I totally agree with that. And so then we, we start, you know, we have a panel, discuss old Oracle databases. They have some MongoDB, they have data pipelines, they have on prem, they have cloud. You know, like, we start talking about what are the second or Forward effects. And my point was I'm like it's the whole economy, right? Like every, almost every business like, like you said the tech space is a little bit different because you could be disrupted. But like John Deere, Coca Cola, JP Morgan, right? These insurance businesses, all of these, the S&P 500x tech is going to be a huge beneficiary of the productivity gains from AI. Very few companies can just like okay, flip a switch. Now we use Claude in every workflow at the, at the company, right? Yeah.

I wonder if we're skipping a step, you know and it's a little bit too much to expect to have something like anthropic Claude, which is basically the other, the other chat. I, I tried out a technology that I think was we may be getting incorporated into our product that so I want to get specific about the name of them but they, they do you know, a voice agent that is GPT, you know to have something like that just show up into a large company and all those people. So good I went and tried it. I was testing it out and using it in you know, particular with a particular kind of scenario. And then I went looked at it from like the scenario of it's going to make phone calls that are like you do you have a ton of workflows and how do you actually get in there? How do you actually change the culture despite this thing is great. I use it every day. You know, how do you get the culture to change at a large place like that?

Right.

Our lead generation, you know, it's going to go find new business. And I was talking back and forth and saying oh just hold on a second. And I think I got this and what. But I think we're you know, not necessarily like.

And I think there's a lot of like companies like Anthropic and others and even like trying to throw it off.

But just trying to be really natural about a conversation it kept up with me across the board. Very good response time. You know it tends to and probably should you know, say up front that it's an AI. But if you get over that hurdle then you're on to like what's you know, what's the value prop and why are you know that's honestly that's probably.

Why IBM has done what been able to do what it's been able to do all these years that like they are experts at getting in and helping you know, make, make, make, make art out of a pile of paint. You know what I mean? And like you are, you Interested?

What are you, how are you taking care of this thing now? What would you do if you could do it this other way and so forth and you're just trying to get to a, you know, follow up scheduled call with a human where I, I got you a hot lead. This is somebody who actually might be interested or whatever. But all of that and all this.

Stuff and like IBM comes here and like holds your hand and shows you how to.

Work on kind of. I'm forgetting the word right now, but no qualifying, you know, all the qualifying can go on. This stuff is amazing. And there are companies starting to use it at scale, you know, like they've got it doing, making a thousand calls. And it's not spam, it's on a very particular topic. And you don't want to talk, don't talk. Organizational transformation.

Yeah. You know, and so there's, there's, there's clearly a place for that. I've always been like sort of down on IBM, but I totally get the, the issue there is that you need these big firms that can get in their hands into this monstrosity of a corporation and actually change it and able.

To provide the value. It's incredible.

So that's. I've been doing a bunch of work on a company called Twilio. I don't know if you're familiar.

Yeah, I used to use Twilio, this is somebody else. But yeah, yeah.

And bring new technology, you know.

Yeah, we used to be like a McKinsey too would be.

Yeah.

To get.

And basically they build APIs, which are application programming interfaces for folks that aren't familiar, basically. John, correct me if I say this wrong, the APIs are sort of.

How so. Yeah, there's something missing there because those.

Companies, Oracle Software talks to each other. So anyways, Twilio makes a set of communications APIs and they basically abstracted that.

Like let's say Oracle not as a technology company in this case, but more as a technology user. You know, there's huge advantage they are going to get as they allow these things to figure out complexity of the.

Telecom world and you know, text messaging and voice and video calls and all these things. And so that if you're a developer building an application, you can just tie into Twilio's APIs. And if you want to make phone calls, if you want to send your.

Customer how to get these things in there and they just, they got to figure out how to do it. But that's true across the board.

Text messages, you know, most of you have probably experienced like the two factor authentication, right where you sign into an app, you get a text message to your phone that all, you know, totally overwhelmingly runs that market. And I think it's a really interesting investment here because all of these AIs that are, and all these apps that are being built like there's going to be some incredible voice AI like you're.

Torg with different, you know, companies out there for their AI companies. They're doing AI for the non AI companies. They have to get into AI by finding the right investment.

Talking about. And I still think if you build an app and use voice AI, Twilio is still the best way to get that voice AI like to actually make phone calls and that people you know, can talk to.

Well I think mints to me for bringing it into their business.

Yeah. And it's funny like despite what we might think, not every single employee at every corporation is focused on AI all the time.

Like we that's true infrastructure wise. But you know, I, I haven't seen what they do for keeping up with the actual, you know, the voice agent, you know, being like if you work.

In tech, which we are, I would say we're at, we're in finance but we invest in Texas. You know, we spend all our time in things thinking and talking about it. But like the average, the average business user is not just going to go.

Out there and be real time and very believable conversation kind of thing. Maybe, maybe they're keeping up on that front.

They're not, they're not building their own voice agent but they're, they're a layer underneath the, the voice agent. Maybe, maybe they're on top of the voice. I'm not sure what the right way to like oh there's this, I heard about this new technology. I'm gonna like try to bring it into work. Like they're, they're just gonna go do their job and if unless their job requires them to do that, they're probably not gonna just natively start using say.

Yeah, it may be that these other folks use Twilio underneath what they do.

And that makes it a Claude in the business somehow.

You know, it's, it's not easy but I think the, the more it's, it's, it's being talked about, it's kind of, you can't miss it. You know, every day more and more is being discussed around this. It can't be avoided. So they're the early comers and they're the ones who hold back and the ones that watch others do it then the ones who realize, you know another one of those like a good infrastructure investment is Twilio Public.

Yeah, TWLO. It's about $105 a share. It's, it's actually not that expensive for software Companies trading like 19 times forward estimates and growing, growing. You know I think the estimate right now is 10 revenue growth or something the next year. But I think there's, there's.

We can't wait any longer.

Right. And I guess that just kind of goes back to the beginning of this conversation. Whether we're in a bubble or not. It just when you think of it like that, it still feels like we're so early, you know, like if, if AI, if the models themselves stopped improving. There's still so much work to be done getting this technology side potential to the revenue estimates if some of this AI stuff starts to them.

I used them for years. I found them to be very reliable and doing a good job.

Yeah, I'm really excited about this. The software infrastructure side of things like there's a, like even building into the right places and into the hands of the right people and it can make productivity like it can improve productivity as is if it's, if it's implemented correctly. There's a couple of commenting apps on replit like so it can build the app, it can set it up for you. But if you want data. Okay. I still hear from Craig the risk of under investing is maybe what's adding to the hype. Therefore the bubble talk. The FOMO seems to be the fuel. Definitely. I mean I would agree with that, John.

Yeah, I guess it comes down to like how you jump in. You know, it's like there will be parts, there'll be little mini pops that happen. I don't see a, you know, dominoes tumbling.

I gotta go get an API to a data source. Right. And if I want like the phone call thing. Right. If you want it to be able to make phone calls, you gotta go get APIs from Twilight. Like there's all these, there's still all these like third party. The way somebody said it is like very few applications are monoliths. Right. Almost all of them tie into a number of other applications. And so if you are one of those like pieces of the software plumbing, I think there's, there's going to be even more demand for that as everything over. Yeah, EPIC has been around forever and should have implementation down to a science in the healthcare world. But even their implementation isn't seamless. I'm not familiar with epic, are you John?

No.

Maybe you want to tell us, Craig, what company that is. The number of AI applications skyrockets because AI is making the coding. I can't imagine these programs just up and changing to one of the new coding tools. I do think that in the, in this, if you're, if you're tech adjacent, you know, like those clearly are the first adopters of this stuff. And if you're in the software business, you're using AI.

Oh, you're absolutely. There's not, it's, it's not that they're you know off all using something like replit necessarily. You know, there's a lot of those tools are kind of, I think being used more for smaller things itself, so much easier.

But I, I don't think somebody is going to like, I don't like. Why would a company go build its own like Twilio platform where you can talk to all the like Twilio has all these agreements in place and more.

For stuff that's starting from scratch, prototyping. But for existing code houses like where I work. Absolutely.

Are we with the telecom providers? They have data centers all over the world so that they call points of presence pops so that they can deliver the service very quickly. And you know, in the United States.

We have basically using AI and whether it's Cursor or Windsurf, you know, or CLAUDE code, you know, we're using AI to get things done quicker.

What's your main ide, John? Where do you spend your time if you're actually drafting and editing code?

Well, my IDE is Visual Free Telcos.

But globally there's hundreds of these things. And so if you're a developer and you are working on a global scale, you don't want to have to go set up software so that you can talk to Deutsche Telekom and British Telecom and AT&T and, and the Korean one. Right. That's just not that useful. That's not a very useful allocation of your time and resources.

Studio VS Code but that's just the editor. But my go to language model for.

Code set all that up.

So when is. When is Neo4J planning an IPO?

Who's that?

So they are speaking of infrastructure, I use them. Their Graph database company is Claude Anthropic Claude and I also have you.

Okay, and say the name again.

It's a weird name. It's one word, Neo because used a little bit cursor, a little bit windsurf as IDEs that are actual AI IDEs and I'm just starting to play with Claude code which is a numeral for J. Neural 4J.

Okay.

And Neo 4J is. They're based in somewhere Scandinavia, I don't know, Sweden, Finland something. And they, there's talk about them having an IPO command line level interface for doing this kind of stuff.

Yeah. I've tried to play around with because we've spent so much time on replit.

And they currently I think have over 2 billion valuation.

I've tried to play around with, with Windsurf and Cursor and it's. That's over my head. You. I think you've got to be a software developer to know.

Yeah, it's for developers to know how.

To do that and like what I wouldn't when I was.

And they're the, they're, they're a very important ingredient in a lot of the AI systems that are being developed that combine a language model with a, a database behind it. Where the database behind it is not your trick.

Playing with Windsurf, all I was end up, all I ended up doing was talking to an LLM, asking it how do you to do whatever Windsurf was telling me what to do, you know like down, you know the, like you know you're doing pipe on and PIP install.

You gotta know additional kind of tables like a relational database like Oracle. What you're getting, you gotta know what you're trying to make happen. But.

It'S sort of like doing like almost like building build like make or.

You know, an SQL database. It's not like that. It's like a network of connected concepts and it's very amenable to the systems that get built with LLMs.

So can you, can you like maintenance on a building versus building a new building from scratch? Right. And if you're, if you're, if you have an existing 60 story skyscraper, you don't use the same tools to maintain that as you would if you were building a new house. Right.

A new two step and it's a two story house, right. 60 story skyscraper. Because we're talking about zero code and then go make it versus hey AI.

Go explain like the different types of databases and how you would use like a, like I know there's like what a Post POST gray SQL and there's.

Postgres is just like MySQL or Oracle.

Open source standard I guess or it's.

The, it's all that, all of that stuff is. This is standard SQL databases.

Okay.

SQL into my 400,000 lines of code and you Know, see if you can understand what's going on. And it actually really can if you help it along.

Really?

Yeah.

How do you.

Databases. And they're.

Would you call that a relational database?

Relational. Relational databases, yes. And they're the picture like an Excel spreadsheet. Just like a grid of data but not like where stuff's just scattered all around on the sheet but like a clean, dense, filled in grid.

Get. How do you get all that context to the AI? Like how do you.

Where every thing across the top each column is like. So that's the real. That's a million dollar question. So like wind surf can read what it wants. You give it your whole. You expose it to your whole code base and it's doing little name, address, phone number, whatever.

Right.

And each thing down the side is like text searches of parts and looking around like ah. Until it's like aha, I found it.

Okay.

When I'm John Doe, Bob Smith, Mary, whatever, you know. So each row is a record, is a different like set of information about one individual. And each column is. Here's all the phone numbers, that kind of thing. So that chunk. And then having typically a couple hundred of those, mostly doing it. I am finding the pieces where I know I was going to be making changes and I'm doing a lot of copy paste and I'm providing code files or pieces of code files into the kind of project area of a.

Of a describe. Like the workflow. Like you have a repository, right.

Different tables all kind of hooked to each other to store the data that's needed. Each kind of subject, like contact information, that or employees, that's one. Projects, that's one. Any companies, that's one. So each one of those is its own and the rows in it, although.

More like GitHub or something. Right. And then you. Okay, I want to make this change. Look to the, to the interface or something or the database or whatever. And so then you would go maybe download a couple of.

Think of it like a spreadsheet actually those rows will go for like a million.

I see.

There'd be like a whole bunch of those rows and those columns might go for a couple hundred columns.

So it's not something where people are manually in there like inputting all that data either. Right. It's all flows from like data pipelines and things.

It's typically. Well you can have data flowing in from pipelines and so forth, but there is a lot of manual stuff.

Those files.

Well no, when you're. I mean when you have GitHub, you've got already all the files locally as well as up on the server. So you're doing things git, push, git, pull kind of things that kind of keep it updated, but it's all right on your own kind of machine. And then you're reaching in and because you've been doing it happening, but not directly to the table, Most of the SaaS applications out there manually. And if you are a coder, you know, you're dealing with all of those, those files. And I said it's like, you know, whatever 400/2 million lines of code or something, it's in the form of problem are just a ui, a user interface on the front end of one of these big databases. And every time you get on a screen that's got like a form with some fields on it and you hit like add, enter, whatever kind of thing you're updating, probably 500 files or 800 files or something of code, it's like, I know where to look, I know what the stuff is because I was going to go do it myself anyway. It's just going to take me a week and instead it can take me two hours.

Right. And so when you are talking to Claude code or, or whatever, like you're, you might tell it, hey, go into this, I'm working in this file or.

Something, like one single record and this and, and the related stuff. And that's. So that's the way it's been since the 80s.

That's what they call, that's what they call crud, right?

Yeah. Well, CRUD is create, read, update, delete. So it's like, what can you do to one of those tables? You know this, what you set up doing is I'm going to create a new record. I'm going to look at it by reading it, I'm going to change something in it to update it and then I'm going to delete it when it gets. So that's the life cycle of data within one of these things. But really that.

And point it kind of in that direction.

Yeah, well, if I'm doing a wind surf or CLAUDE code or things like that, it's in a place where it can just. I can tell it about them. And it's like, thanks for telling me, I know all the stuff that's there, I'll go grab what I need. But if I'm working with more claude, which is like more like GPT, then I'm having to provide those files sort of as attachments. So much of the applications out there are really Boring. They are just the, the, just the right chunk of data that was needed to get a certain thing done. Think of like, think of like a Redfin. You know, what's going on there? Well, they've got millions of records where each one of those records is a listing or, you know, bringing them in, it's like in Google Drive in a folder, you know, you're bringing them all into the one place or a copy of them, you know, and then saying, all right, let's go, John, do you.

Do any like devop?

And then off from that listing are other tables that handle all the pictures and all the maps and all the whatever, all the, you know, allowable values for stuff.

Are you not on that side of things?

I mean, we all have to do some things there, but yeah, it's not my.

That. That's sort of like one of the things I've learned recently too. That's a kind of its own specialty, right?

Yeah. Or how many bedrooms there can be or, you know, every little bit of it. So that's just all the data and then you've got, well, what does that look like to the user? And something gets put in front of them. So when AI came along, the latest kind of AI, you know, the open AI and the chat GPT and all this kind of stuff, this is the first new thing to happen in 30, 40 years. That's really interesting and new. Yeah, there were other things, you know.

The cloud probably should be thrown in there, right? The shift from on prem to the cloud.

Well, even when you got up on the cloud, you were still, what were you sticking on the cloud? A bunch of CRUD databases.

Right.

So you really, you know, you changed how it got. Right.

If you're like, there are people and they are, they're a DevOps guy, that's what they do.

Yeah, like, so they're in a way kind of enabling developers to do what they do because they're keeping the whole cloud and the environment set up the way it needs to be and this server getting copied over to there and so forth and then as you're going and taking code that a bunch of developers maintained and where it got put and so forth. But what, but was there real innovation in what people were actually getting, you know, what they were able to work with? Not that much. Then you had, throughout the 2010s, you had the first kind of wave of AI with all of, you know, predictive analytics and learning stuff, machine learning, prediction machines. And there it was like, you know, now that database we're Talking about is still there, but some of the fields in it are now getting filled in in a smart way. Printer working on your kind of. And Cody's going to kill us for getting so technical. But you're. You kind of have to get that stuff out to the customer. There's all this stuff that they go and do there. So like they're the. They're the programmers programmer.

They're sort of the, I guess the middleman between like the builders, the true builders and the end product and the.

User predicted by the machine. But now, I mean the ChatGPT and all the LLM stuff, large language model stuff has just changed things absolutely completely. It's the difference between, you know, orange juice and rocket ship. It's like there's nothing to do with each other.

So do an LLM can very accurately tell you that orange juice and rocket ship don't usually appear next to each other in a sentence.

Well, they'll find out that the making. I would, I would put the product manager in that space and I would put the DevOps really thing they have in common is Tang, which was an orange drink that the astronauts had. And you know, it'll go find that out. But like things to do with just the world that we live in and being able to, you know, you set aside for a minute, well is it really smart and all this kind of stuff and certainly knows as in between the. The code and the iron.

Oh, okay, I see code and the cloud. I see. Okay, okay. The other direction. Okay. Yeah, interesting. Yeah, I saw that they. And we're getting long here. But I did see the other day somebody. One of the main. Was Amazon or like one of the main tech.

Not conscious and who cares but like it can really hit on everything in the world that we go and deal with. So it's a. It breaks it wide open.

So to go back to the database thing, this is a little technical folks, but I do. I think it has big investing implications because there are companies like Mongo Database and those kinds of things. So tell me now with the AI stuff, how has it changed how you use the database? So the company said that they're now having like coding as part of their product manager interviews, which they didn't used to because like they're like if you're a product manager and you have these tools now, you can come up with products, you can prototype, you can come up with functionality on your own. Yeah. You actually don't have to know how to code. But like that is like why would.

You as a Product manager, not, you know, about hallucination. So these language models, they can be really smart, but when they're stupid, they're going to sound really smart. And that doesn't work for businesses that rely on accurate information. So by themselves they have not solved that problem. They still have they, that, you know, they'll, they'll keep pushing on the frontier of it to be like, that's not making as many of those kinds of mistakes.

Yeah.

And also the information that's in the language model is like the entire be.

Using some of these new tools to help that side of things.

And I went the other way. Right. I was a product manager and an Internet but in compressed form where it can't decompress it back out to get it word for word of what it was. So it's not like knowing everything. It's having good sort of VP of product for ten or a dozen years and. But I kept my kind of technical chops and I made the move to, I wanted to go to the technical side and to be actually building stuff. And that's what I do now.

Interesting. Well, cool. John.

That was a common sense about everything. But when you want to get back to actual information, think about lawyers, you know, and they want to be able to say, I want to actually say what was the case that I can bring to use on, on this thing? Well then you better have access to the actual full data, which the language model does not have. Yeah, it used it.

I learned a lot from these calls, as always. And your expert, I love getting into the technical stuff with people that do it every day because as good as AI is, I can't, I can't understand like the human side of it from AI. It's not the same as talking to you and it's like, oh no, this is really how we humans do, you know, I mean, I could have they, I trying to explain it to me, but it's not the same as full.

Data to create itself, but it doesn't have it in it. So now when you get into the, this world of AI applications, they have to bring those pieces together, the language model or the generative AI together with the real data, which is in the form of all the laws.

Hearing it from a human that's been doing it for 20 or 30 years or whatever it is. And oh no, no, this is really how it all ties together. Like that makes so much sense to me.

So I appreciate it for sure anyway.

But it has folks, it has like big investing implications because this like we're, I'M focused a lot on software right now and at the hedge fund because I think there's huge opportunities because a lot of these software stocks are at like 52 week lows like Salesforce and.

Adobe and work in, in, in the world or in the country or so forth or all the data in my company's database and everything like that. So for it to bring that stuff together well you actually need a really good database to work with the LLM and what our company has found is that graph database and not just ours but lots of them have found that the graph database is a great way to go for that day.

And so many like not just those infrastructure companies like confluent all this. So there's. There's sort of like this big divergence in the world of software and you have a handful that are like trading a huge premiums. You have the Palantirs and Oracles of the world which are trading, trading at valuations they've never traded at before and you have like the Salesforce which has never been this cheap before. And so trying to figure all this out. I think there, there's some major opportunity.

And understanding and so what's been called RAG retrieval augmented generation that goes and.

Gets some what's happening in the world of software right now. So John, last thoughts. We're gonna wrap it up.

Just pleasure talking to you again.

There's just a couple comments in. He's talking about DevOps. The. They're the maintenance.

Yeah, the maintenance folks.

Right.

They're the human infrastructure.

Yeah. And then he just asked if we have any plans to invest in Adobe. I don't. We don't right now.

We've chunks of text that it can pull in and use for the LLM. Now more popular is graph rag and taking a graph database and being able to use it to get the right stuff back into that LLM and make it as small smart as it can be.

So does do you know if. What, what type of database Mongo. We've looked at Adobe a lot. Adobe actually is one of the software applications that I do think is probably at risk of being disruptive and that's what the market's pricing in and it's not crazy cheap. It's, you know and the growth is probably going to slow down. I mean if you're doing photo editing and that kind of stuff I just. The AI is so freaking good at that now. You know DB does. I think they have a. No, I'm just looking at that. It's a no SQL NoSQL yeah, it's.

Called NoSQL database and it's a property value database. So it's another kind and it gets used a lot. Where they're really strong is in project that.

I mean it's not like it's so much like Adobe is such a different scenario than Salesforce to me. You know like Salesforce is an app. Yes. And you could, you could build your own version of a CRM right now. But like why would you that need.

To go ramp up very quickly and need to put some kind of storage on their back end.

It's good for do that and that why like why would you have a go through all the effort to rebuild what Salesforce has already built? I mean save money obviously that's a potential but the risk is that you screw it up and your CRM system doesn't.

It's documented oriented data.

That's one of the things that they're supposed to. I think Mongo database does this well is that if you're dealing with unstructured data that's not neatly in columns and rows. If you just have a random invoice or a picture or things like that then you use a no SQL database to store that kind of unstable work. And if you're an enterprise with millions of customers or hundreds of thousands of customers even there's just like it's too large of a risk you're not going to cancel your Salesforce subscription. And not to mention that they have all your data for years and years and years about your customers. Right. Know everything about your customers. But Adobe like as far as I can tell Adobe does not have that sort of stickiness with the data and with the like. If you're in the photo editing design world like there are pure Play eight I've competitors that are better, better products than Adobe.

Figma.

Yeah. Figma came public last week or two weeks ago. Did you see that John?

You think so? Yeah.

Structured data is that.

Well yeah they kind of ended up getting NoSQL to mean something like Mongo when I've always taken it to mean not SQL. So like I would put the graph database in as NoSQL as well. It's not a SQL database but yes they're, they are document oriented and they were going to get acquired by Adobe.

We were going to buy them.

For a bit there you had kind of this. They'd call it a vector database but there's not much database there. You're kind of just having chunks of text and they're getting indexed in a way that's probably too technical to talk about, but that technical way of indexing them or like semantic based indexing is now being used across the board. I'm sure Mongo's using vectors and.

Yeah, no, so FIGMA is. Became public and I think it's like a $15 billion valuation or something like that. And the stock immediately tripled. Designers love it.

Yeah, I've been using it for years too.

Yeah, I did a lot of work on the product and it makes so much sense to me. Basically it's almost like a collaborative environment.

Where the graph database is using vectors. I'll just say it's like picture again, the spreadsheet, but where a single term, anything, pogo stick, you know, any, any term of it.

Your designers, your UX people, your product managers, everybody can get in there and.

Work on anything, whatever is represented by like a thousand numbers.

Multiple designers can work at the same time on. Instead of it being in like little disparate, you know. Yes. This guy's working on a logo, this guy's working on the whatever. You know, it's all happening in Figma. Everybody can see what's going on just.

For that one concept. And those numbers, there's math that can be done on it so that jump and pogo stick end up with numbers that are like close to each other.

So I think it's a great, it seems to me like a great platform, I think.

Yeah.

You know, if the price will come down, we'll. We'll pick up some Figma.

Anyway, good, good seeing you.

And okay, so sort of the way that LLMs work, right. And it's just kind of that same.

Vector, relational vectors, like in the vectors, like in. Maybe we'll see you next week, huh?

Yeah, sounds good. Thanks everybody. Appreciate all the questions too. Okay, bye.

Take care. Language models in LLMs.

You know what? Interesting, John? I watched the interview with the anthropic CEO and I gotta say, I didn't really like him going into it, but after I watched his interview, he came off very likable, knows his business. He seems like just kind of a quiet confidence, like he's not really worried about all these things that everyone else seems to be worried about all the time with AI. Yeah. One of the things he was talking about was like mama Day, huh?

Dario Amade.

Yeah, Gosh, where was I going with that? It has to do with the vectors. Gosh dang, I can't remember what I was saying.

So you saw him, you saw an interview with him and was he talking.

About the vectors yeah, it had to do with what you're saying. Anyways, it was good. It was a good interview. And like, I, I don't know, it makes me think like the, like, I think about Oracle even and like that product has been around forever. Right. The Oracle database was in the 70s, I don't know, 80s?

Yeah, yeah. I think of 80s, but have been as early as.

We still use a lot of Oracle databases. Right. I mean, I don't know. That's a rapidly growing part of the market, but that's still a huge business that's out there. And just because like these new things are coming along doesn't necessarily mean that the old way of doing things goes away either. Right.

I think of it as inertia.

Yeah.

You know, I mean, it's not going to go away quickly. They have a ton of customers and they. And it's another reason connected back to what we were saying earlier about having the installed base is so important. That's one of their biggest assets is their installed base.

Yeah. And I do think that, like, when you're talking about. Because they're really focused on working with enterprises, I think is where I was going. Claude, he's taking anthropic, like almost not. I don't know that they're yielding like the consumer space or like giving that to open. I still think they would love to compete in the consumer space, but clearly chat. ChatGPT as an app is the consumer app for AI and it's, you know, orders of magnitude larger than Claude's app or Grox app, despite Elon's threatened loss of Apple. And he's just talking about, like, when you're dealt, when you're dealing with enterprises. He's like, okay, I'll go meet with the CEO. And they're like, yes, I'm in. We would love to do a deal with you. Okay, let's make it happen. But then he's like, then I have to deal with like 100,000 employees at this enterprise and somehow get them to use Claude and do it in a way that is complimentary to their existing workflows and stuff. And like, that's challenging. Right. There's like, there's a lot, like they have, they have old Oracle databases, they have some MongoDB, they have data pipelines, they have on prem, they have cloud, you know, like, very few companies can just like, okay, flip a switch. Now we use Claude in every workflow at the, at the company. Right. Yeah.

I wonder if we're skipping a step and you know, and it's a little bit too much to expect to have something like Anthropic Claude which is just basically the other, the other chat GPT. You know, to have something like that just show up into a large company and all those people, it's like you do you have a ton of workflows and how do you actually get in there? How do you actually change the, the culture despite this thing is great. I use it every day. You know, how do you get the culture to change at a large place like that?

Right. Like, and I think there's a lot of like companies like Anthropic and others and even like I mean that's honestly that's probably why IBM has done what been able to do what it's been able to do all these years that like they are experts at getting in and helping you know make, make, make, make art out of a pile of paint. You know what I mean? And like you have all the paint and all this stuff and like IBM comes here and like holds your hand and shows you how to, how to.

Organizational transformation.

Yeah. You know and so there's, there's, there's clearly a place for that. I've always been like sort of down on IBM, but I totally get the issue there is that you need these big firms that can get in their hands into this monstrosity of a corporation and actually change it and bring, bring new technology, you know.

Yeah, we used to be like a McKinsey too would be.

Yeah.

To get so, so yeah there's something missing there because those companies Oracle, like let's say Oracle not as a technology company in this case, but more as a technology user. You know there's huge advantage they are going to get as they allow these things. Does they figure out how to get these things in there and they just, they got to figure out how to do it. But that's true across the board with different, you know, companies out there for their AI companies. They're doing AI for the non AI companies. They have to get into AI by finding the right investments to make for bringing it into their business.

Yeah. And it's funny like despite what we might think, not every single employee at every corporation is focused on AI all the time. Like we are, you know, like if you work in tech, which we are, I would say we're in finance but we invest in tech. So this is, you know, spend all our time thinking and talking about it. But like the average, the average business user is not just going to go out there and be like oh there's this I heard about this new technology. I'm going to like try to bring it into work. Like they're, they're just going to go do their job and if, unless their job requires them to do that, they're probably not going to just natively start using Claude in the business somehow, you know, foreign.

It's, it's not easy. But I think the, the more it's, it's, it's being talked about, it's kind of, you can't miss it. You know, every day more and more is being discussed around this. It can't be avoided. So they're the early comers and they're the ones who hold back and the ones that watch others do it, then the ones who realize we can't wait any longer.

Right. And I guess that just kind of goes back to the beginning of this conversation. Whether we're in a bubble or not. It just, when you think of it like that, it still feels like we're so early, you know, like if, if AI, if the models themselves stopped improving. There's still so much work to be done getting this technology into the right places and into the hands of the right people and it can make productivity like it can improve productivity as is if it's, if it's implemented correctly. There's a couple comments here from Craig. The risk of under investing is maybe what's adding to the hype, therefore the bubble talk. The FOMO seems to be the fuel, definitely. I mean, I would agree with that, John.

Yeah, I guess it comes down to like how you jump in. You know, it's like there, there will be parts, there'll be little mini pops that happen. I don't see a, you know, dominoes tumbling, everything over.

Yeah, Epic has been around forever and should have implementation down to science in the healthcare world. But even their implementation isn't seamless. I'm not familiar with epic, are you, John?

No.

Maybe you want to tell us, Craig, what company that is. I can't imagine these programs just up and changing to one of the new coding tools. I do think that in the, in the, if you're, if you're tech adjacent, you know, like those clearly are the first adopters of this stuff. And if you're in the software business, you're using AI.

Oh, you're absolutely. There's not, it's, it's not that they're, you know, off all using something like replit necessarily. You know, there's a lot of those tools are kind of, I think being used more for smaller things and more for stuff that's starting from scratch prototype, but for existing code houses like where I work. Absolutely. Are we using AI and whether it's Cursor or Windsurf, you know, or Claude code, you know, we're using AI to get things done quicker.

What's your main ide, John? Where do you spend your time if you're actually drafting and editing code?

Well, my IDE is Visual Studio Vs Code but that's just the editor. But my go to language model for code is Claude, anthropic Claude. And I also have used a little bit Cursor, a little bit windsurf as IDEs that are actual AI IDEs. And I'm just starting to play with Claude code which is a command line level interface for doing this kind of stuff.

Yeah, I've tried to play around with. Because we spent so much time on reletric, I've tried to play around with, with Windsurf and Cursor and it's. That's over my head. You. I think you've got to be a software developer to know. Yeah, it's for developers to know how to do that. And like what I would. When I was playing with Windsurf, all I was into, all I ended up doing was talking to an LLM, asking it how do you to do whatever Windsurf was telling me what to do, you know, like down, you know the, like, you know you're doing pipe on and pip install.

Got to know what you're getting, you got to know what you're trying to make happen.

But if you're working, it's like, it's sort of like doing like almost like building like maintenance on a building versus building a new building from scratch. Right. And if you're, if you're, if you have a existing 60 story skyscraper, you don't use the same tools to maintain that as you would if you were building a new house. Right.

A new two stair and it's a two story house.

Right.

60 story skyscraper. Because you're talking about zero code and then go make it versus hey AI, go reach into my 400,000 lines of code and see if you can understand what's going on. And it actually really can if you help it along.

Really?

Yeah.

How do you get all that context to the AI? Like how do you.

So that's the real. That's a million dollar question. So like Windsurf can read what it wants. You give it your whole, you expose it to your whole code base and it's doing little like text searches of parts and looking around like, until it's like, I found it. When I'm mostly doing it, I am finding the pieces where I know I was going to be making changes. And I'm doing a lot of copy paste and I'm providing code files or pieces of code files into the project area of a.

Describe the workflow. You have a repository somewhere like GitHub or something, and then you, okay, I want to make this change to the interface or something, or the database or whatever. Then you would go, maybe download a couple of those files.

Well, no, when you're. I mean, when you have GitHub, you've got already all the files locally as well as up on the server. So you're doing things get push, get pull kind of things that kind of keep it updated, but it's all right on your own kind of machine. And then you're reaching in and. Because you've been doing it manually and if you are a coder, you know, you're dealing with all of those. Those files. And I said, it's like, you know, whatever, four hundred half million lines of code or something. It's in the form of probably 500 files or 800 files or something of code. It's like, I know where to look, I know what the stuff is because I was going to go do it myself anyway. It's just going to take me a week and instead it can take me two hours.

Right. When you are talking to Claude code or whatever, you might tell it, hey, go into this, I'm working in this file or something like that, and point it in that direction.

Yeah, well, if I'm doing a Windsurf or Claude code or things like that, it's in a place where it can just. I can tell it about them. And it's like, thanks for telling me. I know all the stuff that's there. I'll go grab what I need. But if I'm working with more claude, which is like, more like GPT, then I'm having to provide those files sort of as attachments or, you know, bringing them in. It's like in Google Drive in a folder, you know, you're bringing them all into the one place or a copy of them, you know, and then saying, all right, let's go.

John, do you do any, like, DevOps stuff? Are you not on that side of things?

I mean, we all have to do some things there, but yeah, it's not my.

That. That's sort of like. Sort of things I've learned recently too. That's a kind of its own specialty, right?

Yeah, right.

If you're like there are people and they are DEP They're a DevOps guy and that's what they do.

Yeah, like so they're in a way kind of enabling developers to do what they do because they're keeping the whole cloud and the environment set up the way it needs to be and this server getting copied over to there and so forth. And then as you're going and taking code that a bunch of developers are working on you're kind of, and Cody's going to kill us for getting so technical but you're, you kind of have to get that stuff out to the customer. There's all this stuff that they go and do there. So they're the programmers, programmer.

They're sort of I guess the middlemen between the builders, the true builders and the end product and the user.

I guess I would put the product manager in. That's.