Colorado Innovators: Tony Fagan, CEO of VideoAMP

[00:00:00] We're in a competitive environment. We see opportunities to not just be successful with what we're building today, but to move into entirely different business models and areas that are adjacent to what we do because we could build software to do it that much faster.

[00:00:19] In this episode of The Bear Roars, Dan Caruso sits down with Tony Fagan, CEO of VideoAMP and one of the most technically sophisticated leaders navigating the AI transformation sweeping through the software industry for a wide-ranging conversation on what it truly takes to lead a company through an AI-driven world.

[00:00:38] Tony shares the hard-won lessons from restructuring engineering organizations to replacing expensive data infrastructure with new AI-built systems and why every CEO, not just tech leaders, needs to treat this moment as a platform shift on par with the birth of the internet. Let's poke the bear to kick off the conversation.

[00:01:00] Well, Tony, I'm going to enjoy this because you and I have just recently met and you were telling me kind of the story you're going through now, your backstory too, but the story you're going through right now, which is really why I'm so interested in having you here because I think it's so relevant to what so many companies are facing and you personally are particularly well equipped of how to deal with this.

[00:01:21] And what we're going to talk about is the effect of AI on, I would say, more established companies in this case, meaning more established from the perspective of what were tech startups not too long ago,

[00:01:36] have done well looking backwards in time and then are now faced with this new chapter caused by AI and requiring the urgent and intense adoption of AI in order to not just thrive going forward, but in many cases, even to survive.

[00:01:58] And you are particularly well equipped because you're entering the COC from the vantage point of a CTO and with deep skill sets and how to quickly and confidently spearhead kind of this type of transformation. And I think so many companies are facing this right now and most of them aren't equipped with the type of technical talent that you represent from a leadership standpoint.

[00:02:21] And I think the story of how you are tackling what you're tackling, how you're progressing through it is going to be so relevant. So that's why I'm excited. And that's why I'm not the least bit worried about filling up this time slot, which we were joking right beforehand that he was comparing to some past guests that says, I don't know what we're going to talk about for the better part of an hour and a half, but I do. And that's what it is. Outstanding. Intense is the right word.

[00:02:51] Yes. To describe what's going on. Despite your big smile, it's a pretty intense period for you. Well, I mean, you got to roll with it, right? Everyone does. So they're going to be left behind. There's really no option. You can call it an existential threat. You can call it an enormous revenue opportunity. But either way, it's such a massive, you know, I think I refer to it as a platform shift. Yep.

[00:03:17] You know, the way when apps came out and the iPhone came out or when the web came out. And the bigger the company, the more entrenched the company and the less the leadership is represented by people who really understand AI technical transformation. Those are the ones that are going to struggle the most. And I think that's really what you represent here because you're not a big company. You're not a very entrenched company.

[00:03:46] And you are kind of a tech first, AI first leader. And I think it's going to be eye opening for a lot of those who are trying to figure out how do I navigate this to see and practice what you're going through. Yeah, no, I'm, you know, it's a wide range of things we could cover on this topic. And we will. You know. Let's start first with who are you. Oh, sure. What's your background? Why is it you know a little bit about the topic of AI?

[00:04:16] And then we'll get into the company itself that you're in and its history and your history with it. And then we'll get into the the meat and bones of the transformation kind of approach you're taking and how to generalize it relative to what most companies are going to have to really quickly get their heads around if they haven't already. Yeah. And I have a certain appreciation for it or I thought I did, you know, given my background.

[00:04:42] But I'm continuously surprised and not on like a six month or year basis on like literally a month by month, week by week basis and how fast it's changing and how disruptive it really is. And when I think I sort of get an understanding of, OK, in your head, this is what it's going to look like. And then it changes again and you realize you weren't thinking big enough or different enough.

[00:05:09] And it's just it's sort of unsettling, I think, for a lot of people, including me. Yeah. And we're we're. Crucible Ventures is tiny by by standard, but we have a ton of data that is really important to how we prosecute what we're doing.

[00:05:25] And you are conversation of a couple of weeks ago have already inspired us to read got what we're doing and put all of our data in a in a different and much more accessible, much more controllable and frankly, much less expensive format to work then with with in our case, Claude and our other tools. And so you've already inspired us to move really aggressively on that front. OK, we should definitely talk about that. Yes. Not here afterwards. All right.

[00:05:52] Because that's super interesting to me of how all organizations, big, small in the middle are going to change. But OK, you asked about you first. Like what what's your journey and how did you develop kind of these capabilities and how did that lead you to where you are right now? Yeah.

[00:06:11] So, you know, I went to school for engineering and and studied a lot of math and computer sciences as part of that in undergrad many, many years ago and came out west. What university? University of Vermont. And then I came out west. I was going to ski in Park City, but I stopped in Boulder for a night and never left. Wow. At least I did a stint in California.

[00:06:36] But at that point in my life, I was here and I was like, wow, why would you want to go anywhere else? Yep. You know, and I ended up going to grad school out here and I started my career in environmental science and engineering. And that quickly led into software development because at the time we just had a lot of data. The EPA and the states were all regulating all these industrial facilities across the country.

[00:07:06] And they were required to collect all this data and submit it to show that they were in compliance to these new environmental standards. And nobody knew how to do that. And so there was all this electronic data. And so we were, you know, I'm embarrassed to say the language I was using was called Clipper, which is a compiler language for D-Base, which I'm sure nobody remembers. But nonetheless, we were just writing everything from scratch because it just didn't exist.

[00:07:36] And we had at the time what we thought was large amounts of data. And then I moved on from that career and started data startup here in Boulder with a few other folks locally. And that was super interesting process through the venture capital process and whatnot.

[00:08:01] And then as part of raising money, you know, we brought in, you know, an outside CEO. The company went a different direction than some of us expected. We left. I went to California. So I did live there in Silicon Valley for four years. I spent, you know, almost 15 years at Google, which is where, you know, I learned how to build software. Yep.

[00:08:31] And what parts of Google, because in 15 years, I'm sure it's multiple parts. Yeah. I spent a lot of time in the ads business. I would say the vast majority was working on ads, probably mostly YouTube ads, but also some search ads and stuff. But really the focus for most of that time was on YouTube ads. Do you still have some connections at YouTube? Because YouTube just shut us down on our... Yeah.

[00:08:59] We were just discussing that actually before. So, yeah, my connections are on the ad side. Although the CEO of YouTube was a former boss of mine. So, you know. So I guess if you get really desperate, we could just, we could call him. Yes. I don't know if he'll answer, but we can call. So contact him.

[00:09:20] But, and, you know, we were really at the time, you know, YouTube, it was costing a lot of money to store and serve all that user-generated content. And it was just exploding. And the question was, well, is it, is YouTube ever going to make money? Now, I know that seems absurd today, but way back then it was a real issue for Google. And so we focused a lot on the ad system.

[00:09:48] And because it was a very expensive infrastructure. I mean, tons and tons and tons of videos being produced. Yeah. Well, and given your background, you know, we should do, after this, we should do a little homework on the infrastructure, the hardware that was actually built to serve the ads. Because at the time, you know, Google was sort of known for innovating on software with some of these technologies for processing large amounts of data like MapReduce, for example. Mm-hmm.

[00:10:17] But it, and they had to come up with entirely new hardware systems to be able to meet the latency requirements of users. Sure. With all this video. And it just, you know, I wasn't involved in that. But as I understand it, a lot of that just didn't exist. Well, a lot of it, a lot of that was built by people who were our colleagues, you know, worked for us at both Level 3 and then ZEO.

[00:10:45] I mean, if you look at, to this day, a lot of the key team members who built out their network to support, you know, all of Google traffic are former Level 3 and former ZEO folks. And some of them are still here in the Boulder area. Yeah. That's very cool. So, yeah, we, you know, we would staff up teams to support them. And then they'd figure out which ones of the people on the team they liked the most. And then they would offer them double the money. And those people would leave.

[00:11:14] And then we'd have to go train other people. And not only would they have the expertise because of what they were doing with us, but they had all the inside knowledge about how to get our fiber assets in particular from us at commercially unfair terms on behalf of Google because they were using our insiders to, you know, to jump ship. And it wasn't just Google to this day. It's still Netflix was that way. Amazon has a lot of former ZEO, former Level 3 folks.

[00:11:43] Facebook, back when it was called Facebook, probably still to this day, you know, again, was at an overabundance of former Level 3 and ZEO folks. All of the big tech companies would staff their particular fiber groups and data centers as well from kind of the talent pool that we were creating at Level 3 and ZEO. Well, it worked. As far as I know, it worked.

[00:12:10] And it was between basically reducing the cost of through the infrastructure and ads. that YouTube became profitable. And as they say, what were some of the breakthroughs that you remember from your days there when you were trying to really monetize the YouTube video through the ad process?

[00:12:32] What what were the big moments that you guys were able to unleash the value with? Yeah, it's a good question. And, you know, at the time at Google, search ads were were so ingrained as what they called then direct response.

[00:12:57] But performance advertising, like it was close to the whatever the conversion event was. So you kind of understood the ROI better. And so the question was, well, what what would ads on YouTube look like? And and so that the thinking was, well, it has to be branding, right? That that YouTube would operate at the so-called top of the funnel and drive awareness and and intent purchase intent and all these kinds of things.

[00:13:27] These branding activities. And then people would then go search for things to find what they were looking for. And then they would, you know, do buy something, you know, do a conversion of some kind. And so it that that was the thinking. And but the challenge was, you like you go into, you know, an ad agency because that agencies controlled the large brand budgets.

[00:13:52] And you you'd go in to an ad agency and you'd make some sort of the sales team would make some kind of pitch for YouTube. And they hadn't say we would listen and then turn around and say, well, so let me get this straight. Like you don't you can't tell me what I'm good. You can't give me a discount for buying in bulk. You actually can't even tell me what the price is because it's done through this auction thing.

[00:14:17] And if a user, you know, doesn't like the ad after like, you know, 20 seconds, they can just like 15 seconds. They can skip it. And it's like this isn't at all the kind of advertising we want to do. Like we like TV ads, you know, and this is not TV ads. And so that became the challenge was like, well, how do you get brand advertisers that were used to buying TV to want to buy YouTube ads?

[00:14:41] And we ended up collecting a lot of data on and it became sort of measurement. The idea was that, well, maybe an ad campaign should if you're focusing on reach and frequency rather than just counting impressions, then YouTube could be helpful. And if you just took like one. What does that mean? Reach and frequency. Yeah, I'm sorry. Reach. So reach is some sort of fraction of the population that your ad was shown to saw your ad.

[00:15:12] Right. You know, and if you're doing some kind of, you know, CPG, a product you're selling, you know, detergent, well, then you might want a very large reach. You might want 60, 70, 80 percent of the population of the United States to see your ad.

[00:15:31] And so the idea was, well, if the pitch became, we used all this data and we measured the sort of reach and the frequency. The frequency is how often somebody sees your ad in a given period of time. So we got all this data on reach and frequency across TV ads and YouTube ads together across thousands and thousands of campaigns.

[00:16:01] You know, over 100,000 campaigns across many different verticals of, you know, whether it's autos, campaigns for cars, ad campaigns for cars, or, you know, for sports or anything else. And we looked at all the data and basically what we found is that if you moved, you know, a small percentage of your TV budget, like 1% even, over to YouTube,

[00:16:30] then you could increase your reach, your overall total reach of the campaign by a significant amount, by maybe like 5% to 15% for the same amount of money. Like it wouldn't cost you any extra money to do that. And that's because I think YouTube at the time, you know, had a sort of younger audience that was spending less time on TV and were harder to reach by just spending more on TV ads. And so the storyline became, you know, better together.

[00:17:00] Whenever you buy TV, just buy, yeah, marketing came up with that. So, you know, whenever you buy TV ads, just buy a little bit of YouTube at the same time. And that's kind of, I think, what finally drove revenue growth from the ads on YouTube. And there were certain types of ads that were particularly important. For example, new product introduction. You mentioned kind of consumer where it's really big versus a very narrow audience.

[00:17:31] Was it sports related? What types of ads was particularly, you know, younger audience might be it? Yeah, and it's a great question. And I think the ad content makes a big difference. Because at first people were just taking their TV ads and sort of splicing it and sticking them on YouTube. And I don't think that worked very well. At least back then. Because the audience was different. The people that were on YouTube and spending a lot of time on YouTube.

[00:17:59] And so you needed to redo the creative. Because you didn't have as much time. I mean, it's one thing if you have a, say, you know, a 30-second ad on primetime. Or a minute ad on primetime or something. You can tell a whole story in that time frame. But on a YouTube ad that was maybe a user could skip after, you know, 10 seconds or something. 5, 10, 15 seconds.

[00:18:28] Then you had to get your message out right away. And they also, I don't think, had the attention span. And so, and also if it was a little, you know, at the time YouTube content tended to be a little more alternative, if you will. And so, if you made it a little edgier or something, then you might pick up a little bit of viralness. And so, I think there was just very different strategies for making ads work on YouTube at the time. But people had to learn all that.

[00:18:57] It wasn't obvious at the beginning. And now, of course, it's so mainstream, it's all changed again. But these were the, you know, this was 15 years ago. What you were doing, you know, when you were at that stage, what was your role? What was your focus? Yeah. So, I managed a team of data science, engineering, and machine learning. And so, we would work on large-scale data.

[00:19:24] And once we would do some of the R&D, improve some of these ideas out like this. And then, we would build systems to scale them. So, both for planning ads, like based on reach and frequency. And, you know, improving the ad campaign sort of mid-flight, the ad serving process. And then, of course, the post-campaign measurement.

[00:19:48] And the idea was to bring that to market through the sales team for YouTube. And then, to do that in as many countries as possible. I mean, we did it in over 200 countries. So, you know, once you got it working in the U.S., you would scale it globally. Gotcha. Okay. So, let's move forward a bit. And you're spending 15 years. Any other kind of highlights during that time at Google? Well, there are a lot of stories from that time.

[00:20:18] You're in Google. But, you know, but for me, that was really what I focused on. Gotcha. But, yeah, we can talk about some stories separately. Yeah. Well, we'll maybe draw on some of those stories when we get to the heart of this, which is kind of the transformation. But kind of curious forward on how you navigated from Google to end up to the company you're with now. Yeah. Well, you know, like everything, you know, it was a good run.

[00:20:48] And at some point, I was ready to go. When did you come back to Boulder? So, I spent about four years in Silicon Valley and then came back here and then worked out of the offices here in Boulder. Although I traveled a lot. So, you're working in the Google offices? Yeah. Here. So, the Google offices here in Boulder are, you know, not much more than five, seven minutes from where we're sitting right now. Big, beautiful office complex that they built out.

[00:21:17] But they were here long before that. Long before that. In some more beat up buildings right around that. Yeah. Google had acquired a company called SketchUp. And that basically was the first office here. And it just grew from there. Yep. But I was happy to get back here. Sure. As you can imagine.

[00:21:40] And so, anyway, when I was ready to leave, I was looking at some random things. And I was going to take some time off. And, anyway, I had met this company, VideoAmp, that I'm at now through the sort of industry work. Because my last year at Google, I did much more industry outreach work in the ads community.

[00:22:07] And open source, worked with open source software and stuff like this. And so, I got to know them. And, you know, they wanted some advice and technical help. And then they offered me the CTO job. And I was like, well. And what was VideoAmp at the time? So, VideoAmp was, I think, a very technical ad agency. So, it's in the ads business. They were growing very fast.

[00:22:34] And more of a professional services firm. Yeah, absolutely. A managed service, as we call it. Yep. And they would, you know, brands and agencies would give them budgets. And then they would use data to try to identify, actually similar to the YouTube story, which is what, first, I was interested in.

[00:22:56] And they would try to target people online based on what they had viewed through TV, through linear TV, as you'd call it now. So, cable and broadcast TV. And they were trying to make those online ad campaigns more effective by conditioning that on all this linear TV, viewership data, you know, what people watch on cable.

[00:23:25] And did they have their own platform at that point? Or was it more about advisory? Well, they had bought a company that was processing set-top box data and smart TV data. So, they had this notion of data at the time. And they had a small engineering team. And they were starting to work with ad agencies on sort of planning campaigns.

[00:23:48] But, you know, 95% of the revenue was all, you know, from brands and agencies giving them a budget, an ad budget. And then they would take a slice of that budget as a commission and buy the media on their behalf. So, that's the ad agency model. So, you know, back then the mantra was if you wanted more revenue, just hire more people. So, it was definitely a services organization.

[00:24:17] And then, you know, when I came on as CTO, I... Which is roughly what year? It's about four and a half years ago. And so, I got asked by the founder and CEO to sort of transform the company into a software company. And that's where we are today. Now, of course, being a software company is a very interesting, you might say, precarious position to be in at the moment.

[00:24:46] Which makes it super, super exciting. So, before we get there... As to where it goes next. Yeah. So, before we get there, the transformation took place over the first, called two, three years that you joined. And what was that journey? Where did you end up? Yeah. Where were you at two years ago, put that way? Exhausting, I would say. If I had known it would be that difficult, I never would have signed up to do it, quite honestly. But it was hard.

[00:25:16] I mean, you know, especially coming from a place like Google, which had a very particular way of doing engineering. It was highly structured. It was a very strong philosophy of this, you do it this way and not that way. Plus, there's even beyond philosophy, there's underlying kind of culture of technology. Everyone's deep technology people looking to solve things. That's right. From a deep technology approach.

[00:25:42] And the company you went probably thought they were a technology company, but, you know, they were really probably on the fringes of technology. They were really a head agency first who was sort of dabbling in certain type of technology. Probably didn't feel like to them they were dabbling, but by the standards of Google, they weren't a technology company first, I would imagine. No, they weren't even close. Yes.

[00:26:09] And, you know, Google has a particular definition of software engineering of what that means. And, yes, VideoAmp did not fit that criteria in the slightest bit. And so that was the real challenge. And it was everything across the board. Because the other thing was, you know, Google developed a lot of their own technologies. And so when you were there, it was fantastic.

[00:26:35] But when you got out in the real world, you were like, oh, you didn't have any of that stuff anymore. And they were generally speaking. I mean, again, Google invented a lot of the big data stuff. But going back, you know, 15, 20 years, 25 years now. But when I left, it was still ahead of the market, I would say, by five years. So that gap had closed because there was so much innovation that went on outside of Google. Once those, Google would publish a paper.

[00:27:04] And then immediately all these startups would then just launch a product they're offering that would be sort of an outside version of what Google had internally. And so part of why I wanted to leave was to actually see what was out there in the world. Because that gap was closing. And so when I got to VideoAmp, I was like, wow, you know, you're using this, you're using that. And these technologies that at first were kind of foreign to me.

[00:27:33] But like what was even crazier was the way the team was organized, the culture, how the software was actually developed. The software development life cycle, as it calls, which we should come back to when we get to AI. But that's that sort of process of defining traditional way of like you define requirements. And then you design the system. And then you build the system. And then you test it and release it, right? And that just didn't exist.

[00:28:03] Like anyone could write any code at any moment for any reason and just launch it to customers. Or launch it internally. And there was no understanding of whether it was right, whether it conflicted with what anyone else was doing. And I tried to bring some structure to it. And it was not well received. I would say that. You know, I remember one person said to me is pulled me aside who I didn't actually know very well at all.

[00:28:30] And said to me, I don't think you understand video amp. Like we are cowboy coders, which is a common old expression in engineering. And we are here because we can be cowboy coders. And if you take that away, we're all going to quit. And I was like, well, okay, this is going to take a while. It's going to be hard. And, you know, part of it is... Did you think about leaving at the time? Uh, I did. Sure. Yeah.

[00:29:01] And, but there's some things. The company went through a pretty big inflection point. And ironically, I felt like at that point, I couldn't easily leave. I needed to help it through that inflection point. And it gives us an idea of the size of the company. Because this isn't... We're not talking about a startup or something. Yeah. It's just past the startup phase. It was a pretty established company. Yeah.

[00:29:29] So, so we're about 360 employees now and maybe another 40 contractors. So, um, so, you know, late stage. Which was similar to what it probably was when, you know, in the 2022 timeframe as well. Right. It's not like it grew since then. It was... Yeah. It was that size when you pretty much got there. Yeah. It's been, it's been a bit of a roller coaster.

[00:29:54] Um, but, but yes, it's, it's, um, you know, it's gone through these transformations and... And that's why I think this conversation is going to be so interesting when we get to AI is, uh, you're talking about a company that's had a longer history that had, that is sizable in terms of number of employees. You know, it has a certain way of doing things and there's so many companies out there that were funded, you know, 10 years ago, eight years ago, 12 years ago that fit that profile that have had whatever journey they had.

[00:30:23] But now they're not a real big company, but they're far from being a small company and that they have a long entrenched history, including, you know, tech stack and, uh, and an organizational, you know, construct, uh, that, uh, you know, that's totally being upended by AI. And so that's the environment, you know, you're describing.

[00:30:46] Yeah, absolutely. And video, um, had a very strong culture, um, in a really good way in the sense of, in the startup kind of way, like, you know, break through walls, make anything happen. Um, we need something done. If a customer wants to do something, we can say yes and get it done 24 hours.

[00:31:05] So that part of it was really good. But when you start to try to scale something or you need sort of reliability in your systems, you want to become a system of record or you want to be a part of, you know, what I refer to as essential infrastructure at, for a large publisher, um, or, or an ad agency, you gotta, you gotta build differently.

[00:31:29] And plus it's a company that whose history was felt more like an ad agency, ad placement agency, then it felt like a, you know, technical platform. So on top of all that, you're also transitioning from a company that looking backwards in time, you know, was more of a professional services, uh, type of company. Right. Uh, but wanted to be, and was becoming one of more of a technical platform and you've got that dimension as well. You're dealing with.

[00:31:57] Um, yeah. And, and, and we've now flipped the model completely where before 95% was services. And now it's, it's, you know, this year it'll be like less than 5%. It's all software revenue, which is great. Right. And we have a, uh, uh, uh, uh, sort of established product and engineering and, and professional services organization, um, that can support scalable high quality software. Right.

[00:32:26] Which is great, but it is been, it's taken, you know, four years to get here. Yep. Okay. So let's, let's do one more thing. And then we're going to start talking about AI and that is the services that you offer today, the revenue that comes in, the customers who are providing you that revenue.

[00:32:44] Tell us what that is. You know, what are you doing today as a, uh, as a, uh, recurring revenue software focused company, uh, versus what was an ad agency before? Or what are you doing and who are you doing it for and what makes you guys unique in doing it?

[00:33:02] Sure. So we, um, we focus on large scale data, digital data, streaming data in particular, that allows you to do some different things with advertising than, than you used to be able to do.

[00:33:19] Um, you know, whenever content is, uh, streamed on a device, you get a footprint of, you know, what was shown, when it was shown, how long it was shown for, what ads were run. You have census level, as we call it, census level data. Um, you have census level data. And when you get that kind of data, then you can start, you can do a few really interesting things with it.

[00:33:42] You can start to, um, you can advertise based on very customized audiences you're trying to reach. Like you might want to reach car enthusiasts or you might want to reach people that, you know, watched, uh, a sports event or something. Um, all sorts of super interesting, creative things, um, that you couldn't do before. And, um, there's large scale what we call outcome data that's available.

[00:34:11] So this is the actions that people took after being exposed to ads. You know, did they go to a website? Did they do a search? Did they, um, visit a store? Did they buy something in that store? These kinds of things. And when you, when you bring all this data together at sort of a household level, um, this large scale sort of, um, what media they're consuming, the content they're watching, what ads they're exposed to, what actions they took.

[00:34:41] Now you actually, and, and the fidelity of that data and the accuracy of it, you are actually have the ability now to run incredibly sophisticated ad campaigns that you couldn't run before.

[00:34:54] Um, and if you can do this at scale across publishers, across a YouTube and Amazon, a Meta and Netflix, et cetera, Disney, et cetera, Paramount Plus, you know, on and on and on and on NBCU, um, um, um, then, then you have something that, that is, we think is really unique and new.

[00:35:18] Now, you bring in some AI technologies, which we could talk about how those work in our products, but you bring in machine learning and AI and you start to have the ability to sort of optimize these ad campaigns mid-flight. Now, if you're really good at that, then you maybe can start guaranteeing even what the, on those outcomes. You might go to an advertiser and you might say, look, we will guarantee you a certain amount of sales.

[00:35:47] And that changes the whole picture for advertisers who are, who are trying to figure out how to spend money on advertising in an efficient way and have to justify the ROI of that to their CFO and their boards and everyone else. And, and we're trying to give them a system to do that, um, basically. So, uh, if I was listening to this casually right now, I would be saying, well, that sounds interesting.

[00:36:16] You got there four years ago, you led a transformation. It was harder than you thought, but you got through it. Uh, now you're well positioned with all these additional things you do with AI. And so AI is going to be even additive. So it sounds like you guys have really taken off in the last few years and are now just accelerating kind of your overall success story. But that's really not the story right now.

[00:36:40] The story right now is, you know, is AI is going to enable all these things, but AI is also very disruptive to, uh, to, uh, you know, what you do as a business and is giving you the opportunity, but also requiring you to take advantage of the opportunity to really rethink your, your tech stack, but also the cost structure that's associated with the tech stack.

[00:37:06] So what happened such that it's not all hunky dory, uh, you know, in the last year or two at video amp? Yeah. I mean, it's, it's a bigger change than when I got to video amp and, and trying to go from, um, you know, an ad agency to a software company trying to now go from a software company to sort of an AI based platform or performance, uh, platform.

[00:37:33] And, and our products, but then even just organizationally internally, um, it's such a massive change in, in, in how everybody does their job.

[00:37:45] Um, and whether it's the roles, the responsibilities, whether it's the hierarchy that is common in companies today is going to change, um, whether it's, um, and in particular, it's, it's, it's probably most acute in, in software development and product engineering. But why, why can't you do that gradually? Why can't you, uh, say, well, we're an established company. We've got customers. We've got a nice product.

[00:38:15] Uh, yeah. AI is the new thing and we're going to have to change, but we don't have to change so much so quickly. Why can't you take a, uh, an approach for us over the next three or four years? We'll kind of adjust the way we do things. Why is it so intense? Why is it so abrupt? What's the force forcing factor behind that? Yeah. I mean, I think, I think there's two, two, two drivers of it, right?

[00:38:38] Um, one is the sort of incentive side where there is such an opportunity to build software so much faster and software that you just couldn't have built before yourself. Or maybe couldn't have imagined before. Couldn't have imagined. Absolutely. Um, and we can get into some discussions about that, um, if you want.

[00:39:01] Um, but it's just like, there's never been a bigger opportunity in my view for a software company since the internet. And for, for, you know, and our future is by no means, obviously it's by no means guaranteed. And so we're in a competitive environment.

[00:39:17] We see opportunities to not just be successful with what we're building today, but to move into entirely different business models and areas that are adjacent to what we do because we could build software to do it that much faster. So like, if we take advantage of this stuff and others don't, it's a massive competitive advantage.

[00:39:43] And all of a sudden you start to look at the whole industry different and you start to look at the scope of our company differently. Like, absolutely. I want to deliver on our current mission, but like you quickly start to see that, um, you could do a lot more. Yep. So smaller, newer companies who are AI native have huge advantages.

[00:40:08] They don't, uh, you know, they're choosing to be, first of all, the people who are there are choosing to be there. Uh, they're not in an environment that wasn't AI native that they were already in and now is trying to become AI native. And, you know, you don't have that dynamic.

[00:40:24] You have people are choosing and are getting funded to be AI native companies, uh, and a really small number of people can do what very quickly, what very large numbers of people could have done in a pre AI era.

[00:40:39] So part of what you guys are probably dealing with is, is, uh, are there companies who are getting formed now AI native that are going to be able to do what you do, uh, and get there more quickly and more profoundly if you don't move quickly yourself? How much of that is on your mind or do you feel like that's not really a factor for you guys? No, no, it is. And, and that's like the second part of this.

[00:41:05] The first part is, again, is the motivation to be able to do more, to, to accelerate your product roadmaps and all those kinds of things and, and just deliver better products in the market that are, that are just like stuff no one's seen before. But the other half of it is, you know, you have to do it because if you don't, somebody else is going to do it. And like, I think you're right.

[00:41:32] The big advantage that the startups have that are AI native, um, is really the cost leverage they have. And so you can no longer operate a saw, in my opinion, you can no longer operate a software company at the same cost structure that you had last year. You just can't. So you've either got to double revenue or half your cost. And you gotta do it fast. Well, let's talk more about that. First of all, why is that?

[00:42:00] Uh, and let's focus on the, uh, the doubling revenue is great, but let's focus on how do you totally change your cost structure in a positive way because of AI? Like what kinds of things are you doing? Can you do? Uh, and actually, as we get into that, maybe let's do one pause. You're not CTO of this company. You're CEO. And this is the first time you've been CEO.

[00:42:26] And they, they chose you to be CEO because of your deep CTO background, your deep software engineering background. And the need to kind of lead this type of AI. I'm not gonna, I'm gonna go use a different word in transformation because it's really the AI opportunity. Transformation is part of it, but it's really how to fully exploit the opportunity of AI.

[00:42:52] To do that from the, from the vantage point of not being a CTO supporting some other CEO, but by being in the CO seat itself. So what led to you ending up in the CO seat? Why, why did they turn to someone without CO experience, but, you know, with deep, deep, deep software engineering background? Yeah. I mean, I think an AI was a large part of it. You know, a lot, it was really two things.

[00:43:19] A lot of what we've built in the last, you know, four years has been what I consider infrastructure and for advertising. And, and so a lot of our customers are using that infrastructure. They're, they're building on top of our APIs, for example, as a way to leverage our infrastructure.

[00:43:46] So part of it is we're very technical focused company. But the other part of it, as you said, is that with this AI transformation, and I think the board felt that they needed somebody who was, was positioned to understand and, and, you know, and be aggressive in this area.

[00:44:08] And if we've learned anything from the prior platform shifts, we know that if you get there first or you get there early, you benefit disproportionately. And if you get there late, you get penalized disproportionately. Yeah. Disapportionally, in a, in a, in a, you know, in all capital letters, I mean, the, the vulnerability right now, if you're slow to, you know, to figure this out is, you know, you could be out of business.

[00:44:40] Yeah, exactly. And that's that existential threat that I think AI is for software. And so, you know, we're going to get there early. That's my job. For software, I would say, I wouldn't say for software, I would say for most businesses, there's very few business models that are protected from the need to adopt AI quickly.

[00:45:02] There certainly are some, but, but I would say the vast majority of business models, even lower tech businesses, if they don't figure out how to be a, you know, at maybe not the bleeding edge, but certainly some level of the leading edge of AI, you know, they're going to find themselves extremely vulnerable within, inside of two years. I agree with you.

[00:45:25] I try to frame it in software, just because I feel like I know more about software than these other businesses, but, but no, it's, it's, it's changing. It's hard to think of a business that shouldn't be making some very significant changes as fast as they possibly can. And I think you said this, the larger the organization, the harder it is to make this change because it's not just a matter of, you know, workflows and tools, which is sort of obvious.

[00:45:54] Like, oh, everyone's going to start using, you know, an AI tool. Because you talked about organizational structure. I want you to talk more about that. You talked about leadership structure. You mentioned that. And you also talked about culture, like, you know, the difference between some people really wanting to be in this type of environment and others who can be very valuable looking backwards in time,

[00:46:17] who are a lot less willing and maybe a lot less able to want to, you know, transition to this new world. So let's talk about organization, leadership and culture aspects of it. And then we'll get into some, maybe some more of the technical specifics of the transformation that is underway. Yeah, sure.

[00:46:39] And, you know, I think it starts by, by just, you know, owning the situation, like admitting to yourself, the leaders and everyone in the company and being very sort of upfront and vocal. And I think it's very powerful that this is a massive change and this will change every single person's job in this company. It certainly changed mine a lot.

[00:47:05] And we have to, you have to move now. You can't, we can't wait. And so I think that the message, the messaging just has to be that we have to figure this out as fast as we possibly can. And it's going to be disruptive and it's going to make people uncomfortable because it's changing people's jobs.

[00:47:29] And anytime you change someone's jobs, especially a job, like you said, they've been very good at for many years. And now it requires a different set of skills, a different set of activities. It requires you be good in a different kind of way. And that's really unsettling to people. And you were, when we talked before, one example you're using is, is, is management layers, how management's being structured.

[00:47:58] And you also talked in terms of how the, the engineering organization, instead of it being more specialized, it's more about, you know, I would say functional, you know, around business outcome and having kind of a technical leader on the end to end, you know, technical stack, I don't know if that's the right way to put it, for getting a certain, certain outcome, you know, to the finish line.

[00:48:27] So less reliant on passing kind of between different type of functional specialties and more about people acting more autonomously across a broader scope of what needs to get done relative to a specific outcome. Yep. Yep. And, um. So talk more about what, what that means. And I probably used the wrong words for that. Well, and, and, you know, within the tech organization, as we call it, you know, which includes product and, and engineering.

[00:48:57] And, and professional services. Um, you know. Well, you just said something important there. You said within the tech organization, then you described kind of a range of functions that most people wouldn't think of as the tech organization. Yeah. Like you said, product and you said professional services. So I think there's something real important in that too, that all of a sudden, you know, someone who's in charge of something is in charge, but across a broad scope of, of outcome. That's what I mean by the business outcome.

[00:49:23] Um, it's not, oh, we got product people, then we have professional services, then we have this type of, you know, engineering and that type of engineering. It's, you're, you're giving responsibility to individuals who are looking across a broad spectrum. Well, and those three organizations for us are tightly coupled together because I think they need to really, uh, to function well. They need to be just tied at the hip. I mean, I joined it, so to speak.

[00:49:48] Um, and, um, but the, the roles are blurring, you know, like when I got there, we, I set up the swim lanes, like as an example between product and engineering. And it's very clear what the two responsibilities were. And what I saw a lot of was that product managers were trying to do the job of an engineer and vice versa.

[00:50:15] And it was just causing a lot of, um, unhappiness on both sides. And so we separated those swim lanes and made it very clear what product managers are responsible for these things and engineers are responsible for these things. And, um, and it's very simple, very easy to talk about. And it's like, you know, oftentimes when there were problems with collaboration or in projects, it was because of people crossing these boundaries.

[00:50:41] But you look today and, you know, we have product managers now that are writing code and submitting it to production for release. That is not something we would have allowed, you know, even a year ago. You know, and you have engineers now that have enough domain expertise that are writing product requirements. We wouldn't have let that happen last year.

[00:51:03] So this is all very recent where the role definitions are now fuzzy, um, and, and are, are changing. And, and, and it becomes more about, you know, having the domain expertise. And you start to think about the traditional structures in an engineering organization. You know, one of the problems in, in engineering organizations is that, you know, you anticipate a certain amount of work.

[00:51:32] And so you hire people onto those teams. And you're always wrong. You're, so a team always has too much work or another team has too little work. But they become quite rigid organizationally. And they get hard to reassign people even temporarily across teams. So it's sort of an inefficient allocation of resources to accomplish a set of goals. But in this world now, because software. And you're talking about, you know, database administration, software engineering. Like everything.

[00:52:02] What are the different functions that, when you say different roles, what, what give us just examples? Yeah. So it, it, it could be, uh, what we call backend engineering. So it could be the data systems. Um, it could be developing new features that get released, uh, product features that get released to customers. So it could be anywhere in the stack, um, from the front end user interfaces or APIs to all the way in the backend.

[00:52:30] So you'd have these engineering groups that specialize in these different areas. And, and they would, you know, different companies do this differently, but we would have centralized teams. So we would have a data science team, which would be sort of the idea of these centers of excellence. And people would get better career paths and they would collaborate on, they were subject matter experts in a particular area. So you might have a backend engineering team that was responsible for bringing all the data in from outside the company.

[00:52:58] Processing that data, making it available in some sort of data warehouse for pro, for other teams to sort of use, consume and build products against. Right. Um, and so, you know, these are the traditional structures of a software organization.

[00:53:16] And, but when you look at it now and they, they become a barrier in a lot of ways that they add friction because when you can increase the velocity of what, of developing software using AI, it, it, it actually becomes all about velocity. You want to move as fast as you possibly can. And you quickly realize the organizational structures and the roles that you've had in place for, for decades, uh, um,

[00:53:46] that were there for good reasons so that you could build high quality, reliable software on time at a reasonable cost are now actually, uh, a barrier to moving fast and, and, um, developing software agentically as it's saying using agents to write the software.

[00:54:11] Um, and so you've got to rethink the entire, um, setup. And we, we, we are experimenting with this idea of project teams that were people from, from will come together for a period of time that may be, you know, less than three months to ship some sort of product end to end all the way out to the market.

[00:54:36] Um, and the, and the, and I imagine those, those teams, um, are a very small number of people versus if you look backwards, even two years, it would have involved a lot more people working across, as you say, more friction because they're working across orders, different boundaries, but there's also a lot more people would have to have touched different aspects to do it.

[00:54:59] But now a, you know, I'm putting words in your mouth a little bit, but, uh, you're, I envision that when you say we get this team together, it's probably three people, not 10 people. Yeah. Yeah. No, that's, that's, that's exactly right. I mean, building software is like building a house. The older it is, you start to regret some of the design decisions you made early on. And, and at some point you just want to tear it all down and build something new because it became,

[00:55:26] it becomes too cumbersome to try to keep modifying it over and over and over again. And, and, and software is the same way. And so, you know, we are currently, uh, rebuilding some of our most difficult components of our system that process our, our large scale data, uh, from scratch with six people agentically. We're not writing any of the code. The agents are writing all the code.

[00:55:56] And putting that in perspective, I know we had that earlier conversation. You're talking about removing systems that are, uh, expensive, kind of have brand names associated with them, at least in the software world, uh, and multiple of them that you're heavily reliant on. And you're basically gutting out of those systems, which have a lot of recurring costs.

[00:56:20] And if we would have thought about this a few years ago, would have been a massive forklift to move even one of those systems into something else. Now you're talking about a very small team of people who are building kind of from the ground up, kind of data structure to house a lot of your data and saving a whole bunch of money in the process, but also having, I would imagine more efficient access to that data itself.

[00:56:44] Meaning that your AI tools are able to get access to that data far easier and in a much more straightforward way than the systems you're, you know, you're moving off of. Well, that's exactly right. So it's, um, you know, it's, it's such a heavy lift, what you just described, that you just don't do it. Didn't do it. Didn't do it. We didn't do it. It was such a heavy lift that you just wouldn't have done it. Just wouldn't have done it.

[00:57:13] It was unthinkable even for us a year ago. And from the lens of the- Even six months ago. Yeah, and from the lens of the companies that, you know, had the chokehold on the housing of the data and the processes that were built within their software, it looked like their franchises would last for another 10, 20 years because most of their customers, you know, once you get deeply embedded in a system like Salesforce,

[00:57:43] you know, you would, the thought of moving out of that was just daunting. In fact, you really didn't even think about moving out of those kind of systems two years ago unless you had a really compelling reason to do so. And what, you know, really triggered me when we had that discussion is the, not just the ability to move out of those systems,

[00:58:03] but the massive financial benefits just staggering for companies that are certainly at your stage, but also for companies that are far larger. Now that, you know, I want to get back to the fact that you're able to drive this in a way that you're driving it because of your background. It'd be very difficult for, you know, a CEO of a similar size company with a similar size of both opportunity,

[00:58:32] you know, and need and requirement that they change for them to have the ability to pull that off because, you know, you have deep knowledge of what to do and not to do in order to kind of orchestrate that. But you're talking about getting yourself out of these deeply embedded, relatively recent, you know, systems, but deeply embedded, deeply entrenched, and just building from scratch with a small team. What tools are you building that in?

[00:59:00] Yeah, and what you're describing certainly applies to enterprise software, but it also applies to the own software we built. So what I was describing, of course, was our own architecture decisions and how we built our own products. And that's what we're tearing down and redoing from scratch that we never would have thought. But again, it wouldn't even have been a conversation even six months ago.

[00:59:29] And even three months ago, it was like, well, we're going to try this. But yeah, you know, it's maybe a long shot just because it's such an, I mean, again, it's such a massive change. What's creating the urgency to do it that quickly? Like why? Because you can. Like, again, you start to, it's a weird thing about using AI technologies. The more you use them, the more you start to understand them.

[00:59:54] Like I've trained, I use Cloud a lot, and both Cloud Code and then Cloud Desktop. And you can define projects and skills and all these sorts of things. And a lot of things that I would typically do that would be very time consuming for me, I now do using Cloud.

[01:00:16] And the strange thing about it has, for me personally, has been the more I use it, the more it understands what I want to do. And it could be like, you know, writing a document or something or, you know, it could be an all-hands meeting. It could be responding to an email.

[01:00:39] And it's starting to actually understand my behavior, my logic, the way I reason. And it's learning to do it the way I want to do it. So it's not like, you know, could I get an AI just to do something for me? Of course. Will it do it the way I would do it? Probably not. But it is now actually doing that for me. Which is just, you start to see the possibilities differently when that happens.

[01:01:08] And when it starts to reason better than you, then it's like, okay, now this is something I wasn't expecting. It's one thing if it can mimic me. But when it actually starts to out-reason me on a topic that I think I know a lot about, now, of course, you have to have the right conversation with it. Yeah. But then it's like, okay, now we're going to really start to do things differently.

[01:01:35] Because you get that confidence where now you can like, well, I'm going to just like, wait, I'm not thinking big enough. I'm going to change everything. And that's how it is with us redeveloping our internal software. It's like we start small and it's like, well, what if we started processing data this way instead of the way we used to? And it's like, well, that worked. What if we then did something different and even like got more ambitious?

[01:02:03] And we're using CLAW to write a lot of software. We're asking it to write in Rust, which is a modern, relatively new software language that has a whole lot of benefits beyond what we, the languages of the past, at least in my opinion. We're using some different technologies that we could not have switched to for processing large scale data.

[01:02:29] So obviously the workhorse of large scale data for a lot of people has either been Spark or SQL for years now. But there are more modern technologies that can handle things like polars and stuff that we're introducing. And we're getting enormous performance gains. We're reducing costs to like 20% of what it would have cost.

[01:02:55] What kind of costs are you reducing to 20%? What's the 80%? Are compute costs. And talk more about what you mean by compute costs. So, you know, engineering costs, you know, show up in a few different ways. You typically have, you know, the people you pay to develop and support the software. You often license, in our case, we license third-party data.

[01:03:23] So that's a fixed data cost, as we call it. And then there's compute costs, which is... And by third-party data, tell us what that means. So, like, we will license data from, you know, set-top box manufacturers that collect data on what was viewed in a home. Gotcha. As an example. And so we have that. That's a sort of fixed data cost.

[01:03:49] And then you have what's called the third big cost bucket is processing. Processing that data. Storage and processing. Now, typically, storage is much, you know, in order of magnitude less expensive. So you tend to worry about the processing of large-scale data. So once you assemble all this large-scale data, because we have massive, massive data sets. But there's also, like, third-party providers of software platforms. You know, some people might use...

[01:04:19] Absolutely. And we use them all. Yeah. I mean, we use many of them. And we're talking about, like, it could be Salesforce. It could be ServiceNow. It could be a Workday. It could be a whole, you know, grouping of those types of companies, depending on which firm you are. But those all cost a lot of money, you know, on a per-seat basis, on a usage basis. So that's, you know, what I would say. And that's different than the processing, I imagine. The processing, I think, of that might be AWS. Yeah.

[01:04:47] And I'm thinking more in terms of the products we develop that use so much data as opposed to the internal IT systems like you're describing. Okay. So, like, when we develop the products, I mean, when you think about what our product does, we basically bring in data around the content that was watched, the ads that were served to households, identity information as to identifying a household. So, like, postal address, you know.

[01:05:16] And then what we call outcome data, purchase data, other actions that people take out there in the world. And that's basically what we pull in. But it turns out it's a massive amount of data. So we use a lot of data processing tools, whether it's, you know, like cloud infrastructure, as an example, or whether it's data processing software that's from a third party.

[01:05:43] What would be examples, whether you guys use them or not, what would be the examples? Yeah. So for data processing, well, for cloud, it would be like Amazon, AWS, or Google, which is called GCP. You know, for data processing, common ones are things like Snowflake or Databricks, these kinds of things.

[01:06:05] And so those are, so you use basically third party tools that help you because that way you don't have to write everything yourself. But again, it comes, the trade off is that you don't have as much control to fine tune how you process data. Right?

[01:06:33] And so developing things agentically from scratch gives you that ability to tailor exactly what you want to build your products in the way you want to build them. And potentially massively reduce costs, reduce time to market of your products.

[01:06:59] And, of course, to do the same amount of work with a lot fewer people. And deliver more value to your customers. Absolutely. When you're talking about, you said reducing the costs to 20%, but you were talking about a certain area of your cost structure.

[01:07:15] If you look at across a firm as a whole, and do they have an opportunity to reduce their costs, you know, similarly situated companies, by 10%, 30%, 50%, you know, to, you know, what is the level of opportunity to, you know, if you were just thinking about from a,

[01:07:36] hey, step number one is we need to get our cost structure, you know, to where it needs to be because if we don't, you know, the attack of others, you know, we're not going to be ready for. You know, how much cost do you think, you know, companies of your, tech companies of your scale need to be thinking about, okay, this isn't about reducing our costs by 5% or 10%. We need to look across our entire cost structure and end up with, you know, 30% less cost, 50% less cost.

[01:08:06] What do you think that order of magnitude is? Yeah, it's a good question. And it's probably, obviously, very specific to the company. But, you know, and the other challenge with answering that question is, you know, you have some sort of vision of how far you can see out based on these technologies. And then all of a sudden it changes, like the next week, which we could talk about a bit.

[01:08:31] How, what we're seeing in terms of, because we're being incredibly productive with Claude, but we're also starting to build our own LLMs on our own hardware. Yeah, because this isn't a game of reducing costs. It's a game of getting. It's productivity.

[01:08:49] Yeah, it's a game of, like, getting to a whole different paradigm that is both efficient for what you're doing, but more importantly is unleashing a whole other set of opportunity. And really being on the positive side, the breakaway, or delivering more value, more services, versus those who aren't able to do that. And all of a sudden they're going to be, you know, like we said, they're going to be looking at, like, I don't know where to go from here.

[01:09:17] So if you don't, you know, so this isn't, boy, all of a sudden we could become more, we could spend less, more money and get the same amount of revenue. And it's like, first of all, you just got to change your whole structure. You got to change how you do everything. So that you could be much stronger from a going forward standpoint, from a going market standpoint. Right. And then you have optionality. You have that, you know, what they, you know, investors call operating leverage. Right.

[01:09:43] And, but I would say if you didn't grow revenue at all faster than whatever you were doing before for a midsize software company, you've got to look to reduce, you know, costs in the 10 to 30% range. Yeah. You know, in the next, in the next year. I mean, it's, you can't wait. You got to go.

[01:10:03] Now, of course, the hope is you, you use these technologies to increase productivity, increase the velocity of your product and services, and you grow revenue. To your point. Because it's really not an option. It's really not an option to think we're going to get at the same revenue and same growth rate. We're just going to spend less money. That's really not how this is going to play out. No, everybody wants higher growth rates of revenue. Right.

[01:10:31] And, and better products in the market. So that's where it starts. Right. It doesn't start with cost reductions. And if you don't do this, you're going to end up with shrinking growth rates or even more directly negative growth rates. So it's hard because it's not really well. Yeah, I guess we could do that, but we don't need to really be 10% or 20% more profitable. Our profits. Okay. And we're comfortable. Our growth rates. It's like the world's not going to stand still for you.

[01:11:00] And it's not going to stand still for two years, no matter what your company is. If you don't go on the offensive, you know, other people, other people are, including those who don't exist as a competitor to you. They're going to be able to, you know, the old term of leapfrogging. They're going to so leapfrog those who have innovators kind of dilemma problems. You know. That's right. The world's not going to stand still. And the AI native companies aren't going to look like. The companies today.

[01:11:30] So back to the whole organizational structure. How is it changing it? So one of the things you quickly realize when you start to think about these technologies at scale. So what happens when every single employee is using some sort of AI tool on a daily basis to get their job done? In a meaningful way. Not as a replacement for the search engine. Yeah, yeah. Sure. But like really proficient users of it on a. That's right. On an all day basis.

[01:12:00] They're doing their job better. They're doing the job differently. They're significantly more productive. All of a sudden, like the whole information flow changes. And the structure of the company itself changes. Because what you realize is, number one, you start to want to record every decision that is made. All the information.

[01:12:26] I mean, if you think about what a traditional work structure does for a company is, it essentially retains institutional knowledge. Right? So that you can make decisions. You can allocate resources. And you can grow the company. But when all that information is actually recorded digitally. Right? Every sort of the knowledge that people have. Their workflows on a day-to-day basis.

[01:12:56] The decisions that are made. If those decisions get recorded in a way. You know, in engineering, they're called architectural decision records. You know. But the idea is to say, an important decision was made. We should record what the decision was. Why it was made. What were the alternatives that were considered? And this is like half a piece of paper. Like not a big document here. But when you start recording all of that stuff.

[01:13:24] And that information is available through a chat interface to every employee. You don't need to ask anyone anymore. There's no more status emails that get sent out. There's no more like, you know, team meetings and one-on-ones to tell everyone and communicate what's going on. The idea of like communicating the vision of what the leadership is thinking. You don't need to do that anymore. Because all that stuff is available in real time. And all they have to do is ask for it.

[01:13:54] You know. And so all of a sudden, the role of a manager changes. You know. You want to have mentoring. You want to develop people. So you got to figure out a way to do that. But like the role of a traditional manager of being like explaining to their team what's going on every week, every day. And guiding them in their workflow. You don't need to. And allocating resources to get tasks done.

[01:14:23] You don't need to do that anymore. So you're going to end up with flatter organizations. The roles of leaders are going to change dramatically away from managing per se. And communicating towards making decisions that are important for the company. The agentic infrastructure and workflow are going to actively look for conflicts in decision making and surface them to resolve them. We're already actually seeing this.

[01:14:52] Because this goes back to this notion of a software development lifecycle. So we're running a few software projects right now in this fashion. And you can ask it what's going on at any moment. You can just go to Claude and ask it for this project. Tell me what's going on. And if there was a meeting yesterday where something happened, decisions were made, there's an update. I can now see. I can now know what happened literally yesterday.

[01:15:23] And as you do this, we're doing this across multiple projects. And then it starts to point out, well, hey, be careful. You made this decision here. But you made some other decision on this other project that could conflict. And so the system starts to become the institutional knowledge, the intelligence, if you will, that humans then interface with to make decisions and move. And that's a pretty different world than what we have today. Wow. Wow.

[01:15:53] I thought we were making a lot of progress. And now I realize that the fun part of what I just heard is we're about to discover a bunch of stuff that we haven't even realized is right in front of us. But the scariest part is we're not there yet. So talk more. We only have a little bit of time left. But I want you to go back to the you're starting to build your own LLMs. What triggered that transformation?

[01:16:22] How different does that feel now that you're entering that realm? Yeah. So back to the compute cost as being the cost of processing information and data at the company. That's always a real concern, particularly for data heavy companies like ours, software companies that rely on massive amounts of data. It's a problem.

[01:16:48] And so thinking forward, you know, hard to say how far forward, but maybe a year or longer. We start to worry about the use of AI in doing everything we do and what that's going to cost.

[01:17:05] And so what, you know, I don't know exactly how many years ago, maybe five years ago, there was a researcher and these these AI models that are called LLMs came out with this concept of Laura, which is, you know, these how to take a model.

[01:17:31] Like maybe there's an open source version of a large model that's sort of like a cloud. You know, they call these frontier models. Right. And you could adapt it, right? Using this Laura approach. If you want all the technical details, we can get into it. But it's basically low rank adaptation. It's this approach from matrix and linear algebra. Anyway, so you have this frontier model.

[01:18:01] And then the idea is to adapt it for your own purpose. Like, so let's take writing code. OK. Claude can write code exceptionally well. Now, it turns out our code inside VideoAmp for our product development, we really want to be writing Rust, a language called Go that Google came up with, and SQL.

[01:18:27] And that's, let's just, and then some, a few other things here and there. But let's just focus on those three. The question is, could we adapt an open source LLM using this Laura approach to do as good a job as Claude can do for just specifically writing code in Rust or Go or SQL just for that task? Let's just take SQL, which is so commonly used.

[01:18:56] Could we develop our own LLM, again, adapting an existing LLM out there to write SQL as well as Claude can write SQL? And if we can, then why is that interesting? Because, number one, the model is open source, so it's free, the frontier model.

[01:19:21] The adapter model that we developed, this Laura model, we developed ourself. So that doesn't, we don't have to pay anyone for that. So now, all of a sudden, we've reduced our cost for that particular task. Now, it might be terrible at everything else. You know, the last thing you want to do is ask it the weather. But it could write SQL, which is what we needed to do.

[01:19:44] And so if we can do that in a lower-cost fashion, then we could run that on cloud infrastructure. Or if it's small enough, because it turns out these Laura models are quite small in relation to orders of magnitude smaller than the frontier model. So you don't need the hardware. I don't need a massive data center necessarily to run that Laura model.

[01:20:11] And so we could potentially put this on our own hardware. Now, we might want it for, certainly for anything with our products, we want it on cloud infrastructure for reliability. Right? But for internal development of writing code, maybe we don't need the reliability of cloud infrastructure, and we could run it on our own server. Okay? So now you're doing a hybrid, a little bit of on-prem, as they call it, versus cloud for certain tasks only.

[01:20:41] And now think about extending that. So let's say we had a Laura model for SQL. We had a Laura model for Go language. We had a Laura model for... We now have, say, 100 Laura models to do 100 different specialized tasks. Well, then what you do is, when a user is trying to use cloud, you need basically a fancy router that then hot swaps these models depending on what the ask is.

[01:21:06] So if I'm using cloud and I say, write me a SQL statement, or you wouldn't say write me a SQL, you'd say, get me this data. Right? And it's going to translate what I wrote in English to SQL. It will then go through a router and call in real time and call the Laura model that was specific to SQL. So that's the kind of stuff we're working on. Gotcha.

[01:21:29] And with that comes innovation, like go-to-market products, services, differentiation, hopefully a nice revenue growth rate. That's really what you're aiming for. It's all about revenue growth. Because cost is no longer your biggest problem anymore. Yeah. I mean... But having people who are able to adapt to the environment you're describing, you know, we kind of said all the employees are going to operate in this mode.

[01:21:57] But in reality, it's certain people are going to be very comfortable and able to do that shift. And many others aren't going to be either comfortable or able to. And that becomes... We're seeing that already. One of my challenges and things that get escalated to me are people that have made...

[01:22:17] That have sort of crossed the Rubicon and are resentful or are complaining about the people who haven't, that those folks are slowing them down. Yeah. So the cultural impact of this is real and it's messy and it's... Well, I hope everyone at Crucible Ventures listens to this because, you know, this... We're certainly going through this type of cultural and operational kind of transformation.

[01:22:47] And it's... You know, I have a huge urgency that I, you know, express both in words and actions with the team that this isn't optional. This isn't something we could take our time and it's not going to be easy for everyone. But everyone who wants to be part of what we're doing has to do that or else, you know, this is going to be the wrong environment for them very quickly.

[01:23:11] And we're already seeing kind of both the vast opportunity because of that, but also the, you know, the stress that comes along with it. And I keep telling them this isn't because of the environment we're in. This is the environment that is happening in many places. And where it's not happening, that's where you should be concerned because we're...

[01:23:35] The environments where what you just described isn't taking place in a meaningful way in real time, they are falling behind so much more rapidly than they have any idea because those who are, are moving ahead. I mean, the stuff you just described is you're probably in the top 1% of companies who are, you know, at that mode already. But if you don't get there, you know, you got a problem. And our fear is we're not in the 1%.

[01:24:04] We're not moving fast enough. Yeah. Well, you got to come at it with that type of fear because it just requires it. Well, we're out of time and you were all concerned we weren't going to have a lot to talk about. We could have gone on for another hour is my guess and maybe we will. We covered a lot of ground. All right. Well, good luck with what you're going through. Good luck to your entire team. And I'm glad we built a relationship. We live only a mile apart from one another. And it's great to connect. Thank you for having me. Really appreciate it. Good deal. Thanks.

[01:24:34] Thank you for listening to this episode of The Bear Roars. Check out Stretch, the new song from Dan Caruso, with music by Jason Mendelsohn. Available now on all streaming services. This podcast was produced by Loud Bear Productions and edited by Natanya Chatfield with support from Kendall Weinberg, Gibson Siegert and Alex Kim.