Can You Trust AI With Payroll? with Gabe Monroy

[00:00:00] Well Gabe, I'm surprised it took us so long to get to know each other, given once I heard about your full background, I'm like, how did our paths not cross more over the last several years? I know it. It's a little bit like I've been operating out here in Colorado, doing lots of Colorado-specific things, but always for companies outside of Colorado, typically.

[00:00:19] And so, yeah, it's great to get connected to you and through you to this gateway of just lovely innovation and business success happening in Colorado. Great. So you know a thing or two about AI, is that right? Yeah, you could say I've dabbled a bit in the space. You know, in fairness, probably more on the infrastructure side, you know, selling infra to the big labs and big companies in that space.

[00:00:46] And, you know, now most recently working at Workday on building out AI in the enterprise software space. But what do you mean by infrastructure? Because infrastructure, when it comes to AI, when I hear that, I think of my old world, which is putting fiber in the ground and building data centers. But you're talking about something different when you're referring to AI infrastructure. That's right. It's maybe a click up from like the fiber, right?

[00:01:08] So if you think about it as most of these workloads, whether it's training or doing inference at scale, they're going to require some amount of specialized hardware like GPUs. Or, you know, in the case of Google, you know, where I worked for a while, TPUs, right? Basically specialized chips that are designed to do AI. The trick, though, is on top of those chips needs to be a whole way of distributing and decentralizing the compute infrastructure and allowing for AI to leverage large, large clusters of machines.

[00:01:37] And so for whatever reason, I, you know, in my professional background, ended up at the epicenter of both the pre-AI versions of software for how to orchestrate distributed systems, container technology, Kubernetes, things like that. And then and by the way, including working with some of the large AI labs in the sort of pre-generative AI days on that stack.

[00:01:59] And then more recently, places like Microsoft and at Google on delivering products that were actually used for large scale training and inference by, you know, some of the biggest, biggest names in the industry. And what makes you interesting is that you just covered in a few words, like multiple layers of the stack that you've been involved with. From the application layer that you are in right now, really, and described a bit of down to infrastructure, just one layer up from the infrastructure world I come from.

[00:02:28] So let's go back through each one of those for the people listening in and talk about slowly building up from, let's say there's a data center, there's some fiber that exists, and you know how to move bandwidth around. That all is in place. But take us from that layer up to the application layer, one layer at a time, and maybe talk a little bit about pre-gen AI and today gen AI as you talk about each of those layers.

[00:02:54] Sure. So yeah, diving into the technical a little bit, you know, one of my fun jobs I had at Microsoft was I was in charge of the team that did the orchestration software that powered Azure's physical data centers, right? So think of it as like if you were going to go mint a new data center in Azure, my team's built the first set of software that you would install on all the gear in that data center. And so I learned a lot about what... And that software was doing what? That software was, well, it was doing a bunch of things.

[00:03:23] So what I learned in that process was you typically start with the network layer, right? You got to lay down networking to connect everything. On top of that, you know, at least this is how it worked at Microsoft was you would build a storage grid essentially first. And that was part of the software layer was a distributed storage system that would allow a bunch of servers to basically link up and provide a distributed storage architecture. And then on top of that, you would tend to load compute on top of it. And that compute could do lots of different stuff.

[00:03:49] It could put power services, authentication, networking security, and then ultimately application services. Hey, I want to host a virtual machine on this environment, or I want to run an AI workload on this environment, right? So you think about those layers, you get kind of network at the bottom, you get storage, you got compute layered on top. And we haven't seen that pattern change a whole lot in the new era of AI, right? It's still fundamentally the same approach you would take to building out a data center.

[00:04:15] What has changed is the type of gear that you're going to go rack and stack in that data center. It's got a lot more floating point and matrix multiplication ability through GPUs and things like that. But fundamentally, it still rhymes and is like the data centers that were built years ago. And what's the difference between the gear that goes in so-called GPUs? And I think you use the word TCU? TPU, yes. Tensor Processing Unit. Yeah. So GPUs, TPUs.

[00:04:45] How are those different than what the technology would have been pre-AI? It's different in a few ways. I think one of the first ways it's different is that these chips are designed to perform math, basically, at a scale that is not common for CPUs, right? So their ability to perform...

[00:05:14] CPU stands for... Compute Processing Unit, right? So CPU was the standard for a long time. It's a thing you got in your MacBook, right? It's the thing you're even powering your mobile phone, right? It's got a normal compute processing unit inside of it. And these GPUs stand for graphical processing because the first use case for this kind of math was actually in doing things like ray tracing for video games.

[00:05:40] And so they would build out these chips and you'd buy these chips typically for video game development. So if anyone's ever built a workstation for video games, right? What kind of NVIDIA GPU you bought? That was always the cool thing. And it turned out, and this is, I think, some of the brilliance of NVIDIA, is that that problem of large-scale math operations ended up being the same exact thing you need to do to light up this transformer architecture that came out of Google.

[00:06:10] And so what we see today is lots and lots of GPUs. But not only that, you actually have to have an interconnect across these GPUs in order to enable the large-scale inference and training. And so it's not just about having the chips. It's about how those chips are able to talk to each other in near real time. A lot of them need to talk to each other in near real time. That's what's really happening within these.

[00:06:36] We hear these AI data centers that are gigantic and they need lots of power and cooling. But that's really lots and lots of GPUs or TPUs needing to be connected together and work on making sense of lots and lots and lots of information. Is that the right way to think about it? That's exactly right. And so you've kind of moved away from this model of data center architecture looking like, I don't know, different racks of machines doing different random things.

[00:07:02] You now have racks of machines that need to be deeply interconnected with each other in order to process the same piece of whether it's a particular inference job or something. They all need to be working together at a rate that is novel. This is not how data center architectures have worked in the past. And what are, in lay people's term, what are they doing when they're working together to do inference?

[00:07:28] What are they doing as they're doing their, you know, gen AI thing? What do you, what do you picture in your head is happening where, you know, all these math and signals are happening and all of a sudden it sounds awful intelligent on the other end of it? What's it doing to mimic intelligence? What do you think about it?

[00:08:15] prompt a model? What happens behind the scenes when you say, you know, I don't know, what is the, I don't know, how does AI work, right? You send a prompt in. There's actually a stage called pre-fill and a stage called decode. And what pre-fill does is it's populating a cache in essence of, you know, think of it as like the weights of the model. You know, they're basically trained on every piece of data across the, you know, the internet.

[00:08:42] Actually, let me get, let me create a, maybe a little bit more of a tangible example that you can use. And I'll try to think of one that we're, you know, we're like, there's a current thing we're doing. So, you know, we got the Tulagi Fest and one thing we're doing at the Tulagi Fest is called bold conversations that matter. And we need to reach out to about, you know, 500 people call it. So 500 people, you need to reach out to them individually.

[00:09:06] And we need to kind of share with them kind of five different options across for each of four different time slots and have them give us their preferences. And these are really busy people. So if you don't, if it takes more than three minutes for them to do, they're probably not going to do it. So, you know, we go to, we use Claude. We go to Claude and say, Claude, you know, we want to create this thing. Here's kind of the session titles and descriptions. And, you know, here's the slots.

[00:09:36] Here's the people that, you know, we want to give one at a time. We want to give them access to the website. We want to get them to choose which ones they find interesting and not interesting and whether they'll be able to come or not. Can you please just build something for me? And then it goes off and comes back. It's like, wow, 10 minutes later, that looks pretty damn good. So what happens? What's going on behind the scenes when it's hearing what we're asking for? And at least when I'm using it, it's usually just hearing me babble for a little while. And it comes back and builds something.

[00:10:05] So what's it doing to make sense of what I'm saying? Break that, you know, and this is for a layman audience. So don't worry about like the deep tech people saying, that's not exactly how it works. But how should like people be thinking about what is AI doing when it takes a task like that and delivers an outcome? I think it's important to kind of reason about the approach of thinking models. For folks who I think probably use ChatGPT when it first came out,

[00:10:34] the initial versions of these models, you ask a question and it would just kind of print out some response pretty quickly. And now when you go into a ChatGPT or Gemini or whatever, what you see is actually a reasoning chain. You can see the model thinking behind the scenes. And what it's doing is it's actually printing out the same kind of response that it would have, but it's being a lot more planful. And in practice, what it's doing is it's actually building a plan in many cases.

[00:10:57] So for your example, if I want to go build a fairly complex event management experience for end users, and you craft that into a prompt that is, I don't know, a few pages long, what the model is going to do is it's going to use its thinking time. This is what we call test time compute. And it's going to actually build out a plan. It's going to sequence a bunch of steps. It might delegate to different sort of internal workers behind the scenes. And you don't see any of this.

[00:11:24] When someone says agent, that's what they're often talking about is kind of partitioning the work to different, you know, you use the word workers. That's what they often mean by agents. It's going to be, you know, we're going to take this piece of work, and it's going to go, and some agents can go off and do that, while this other partition of work is going to be defined, and some other, you know, agent workload is going to go do that piece of work. That's right. And the word agent, you know, to me, it's become so overloaded. So it's almost, it's one of those words that almost ends up losing its meaning.

[00:11:53] But, yeah, you could think of it as an agent. But basically what happens is that work gets farmed out to a bunch of different endpoints. And this is, again, part of the scaling challenge is you need to farm as much of this work out as possible across the array of servers that are in the data center in order to get the most out of that. But what ends up coming back is, and you run through the technical phases of pre-fill and decode. Pre-fill is kind of prepping the model to execute. Decode is the streaming of tokens.

[00:12:20] We tend to, in the infra-land measure, time-to-first token as a key metric that we try to optimize these systems for. Like, you know, so if you ask that question, that prompt of like, you know, hey, you know, how do I build this event management? How about work done per token? Why don't we optimize around that a little bit? I've burned through an awful lot of tokens, and it's getting frustrating. That's right. Yeah, time-to-first token is great if what you got back. Get out of more money. No, no, it's... Or some pinball machines. No, it's true.

[00:12:46] And by the way, you know, at Workday, at the scale that we operate at Workday, like that is actually a huge thing for me. It's, yes, time-to-first token matters, you know, because if someone's asking for, you know, hey, what's the best time for me to take vacation on a given week? You know, we want to optimize the latency. But we also need to make sure that the response is correct. And that's actually a big tension in these systems because the more accurate the model, you know, which typically means it has more parameters, you know, the neural network is more complicated.

[00:13:14] You have a longer time-to-first token if you want a better, more accurate response. I think people have experienced this firsthand. You go try Fable models, you know, from Claude or, you know, the new 5-6 models from ChatGPT, and it's like, it takes a minute for the responses to come back. So time-to-first token suffers, but accuracy and output ends up being a lot better. So a lot of the work in, you know, operationalizing AI is finding the right balance.

[00:13:42] And actually what we found is that it's open weights models in a lot of cases that are striking the right balance for us. So in a pivot that I probably wouldn't have predicted, we're finding that we're getting a lot more success for, you know, through open weights models than we are. And what does open weights models mean?

[00:14:01] So if you think about the proprietary model companies, the big frontier labs, you think Anthropic, you think OpenAI, Google with Gemini, those models, when you build models, you're basically building a set of what they call weights, which are kind of like mathematical probabilities that are, you know, enabling the ability to go from a prompt to a streaming set of output that has accuracy for a given question.

[00:14:31] And that set of weights is actually like IP in the new world. Satya Nadella, you know, CEO of Microsoft talks about this as a form of capital. Actually, he calls it now token capital, which I find fascinating. But it turns out that there are open source companies. By the way, Meta was one of these companies who was building the llama models that they'd fallen out of favor of it. But those were open source weights. So you could go to Hugging Face or, you know, some other site and you could download the weights of these models.

[00:14:59] And that meant that you have full access to the, you know, the ability to host and serve this model yourself. Like you weren't reliant on Meta to provide access to that model for you. You could take it, run it on your own GPUs. Super powerful for a lot of use cases. In fact, you know, people who do a lot of coding these days, some of them are building GPU rigs at home on their own desktops and then using these open weights models running at home to provide coding assistance for free, in essence. Right. For free.

[00:15:28] They paid for the GPUs. But they don't have to pay the provider for this because of the power of open weights models. And actually, this is a big thing. You know, I was part of signing a letter, open letter with NVIDIA and a bunch of other partners just this week, actually, on open source AI.

[00:15:47] Because there's something that has been challenging customers, not just in the U.S., but more importantly globally, is we had an issue recently where the Commerce Department, I think you might recall, shut off access to the Fable class models, mythos class models. And I think, you know, they had their reasons. I think this was deemed unsafe. I think, you know, and by the way, there's been a lot in the news around some of these models breaking out of, you know, in the labs and, you know, causing some drama. So I think there's good reasons.

[00:16:16] But if you're an enterprise who is betting your future on this AI technology, the concept that any government could just kind of kill switch something that you deem as critical path to your business started to send a lot of people scratching their heads and going, wait a minute. Is this structurally something, a risk profile that I'm tolerant of? And there's been a big, I think, counterreaction in the industry towards this, which has driven open weights models kind of to front and center.

[00:16:45] I think Alex Karp had an interview on CNBC recently where he talked a lot about this. Satya Nadella also talking a lot about this. We are also, I am personally also very interested in using open weights models as a counterbalance to this.

[00:17:01] There's, I don't know if you're plugged into this, but fairly early but accelerating quickly in discussions that actually NVIDIA is, you know, a big part of, as are all the major universities here in Colorado about putting in kind of a, you know, an instance of compute that is, allows for, I think I'm using the terminology correctly,

[00:17:27] the open weight models to be used on that and made available not just to the, the higher ecosystem, but the broader ecosystem around it, including some of the earlier stage companies that they can have a shared, you know, physical resource that they could then, you know, run AI on as opposed to using kind of the more public models. Are you, are you plugged into that at all? I am.

[00:17:52] Yeah, I help out on the CU computing advisory board and so you work closely with, with the crew over there. And yeah, I think it's a fantastic resource, right? The ability to be able to perform large scale inference, the, the ability to fine tune models. That's one of the other benefits of open weights models is that you can sort of retrain those models and add different layers in essence to the models to make them specialized at given tasks. And you can run evaluations and stuff to guarantee that they perform on those tasks, but you can only really do that.

[00:18:21] That well, if you have control of the weights, hence, hence the open weight solution. In other words, like a CU couldn't stand up a version of, I don't know, Anthropic or Gemini or open AI models. Cause those are proprietary. Those are closed. The weights that IP is, you know, locked in a room somewhere at one of those frontier companies. Uh, so from an academic perspective, open weights is definitely the way to go. Gotcha.

[00:18:45] Well, there's a, I just learned an hour ago of, uh, the next meeting for that has taken place, uh, at CU in early September. So if you're not plugged into that, we should try to get you plugged into it. Yeah. Love to. Yeah. So good stuff. I'm learning all kinds of stuff. So let's take a pause from this line of discussion. Let's talk about you a little bit more and kind of your journey and how it is, you know, all this stuff. Um, maybe walk us through kind of, you know, how you began to get involved in the tech world, how you got here to Colorado.

[00:19:15] You've mentioned quite a number of different companies you're with. You're currently CTO of Workday. Is that correct? That's correct. So we'll lead up toward that. So tell us about your journey. Sure. It's a, it's an interesting, I think, Colorado journey. And that's, that's how I think about it. Uh, my family's from the Caribbean. So I don't have a ton of roots here in the U S. Uh, we, you know, set up in Cape Cod, Massachusetts. And I got, got, ended up working for a tech company there in the ERP space.

[00:19:43] And they were looking to set up an R and D office for, for the CRP company. And the, the founder, a guy named Scott Raidersdorf was like, Hey, this is a place, Boulder, Colorado. It's amazing. They got a great university. There's a lot of talent. Hey, Gabe, you know, I was like 18 at the time. He's like, Hey, do you want to come out and, and, and help us establish? And I was like, ah, you know, I don't know. I, you know, kind of want to go to school. And so anyway, I ended up going to Tufts for a bit. Didn't really like what I was learning. Um, and my family didn't come from money.

[00:20:12] So like, I couldn't really afford going to Tufts in private school. Costs were just crazy. So, um, a few years later, he's like, Gabe, you want to come out? You've got this R and D thing, you know, it's coming. And, and so, you know, I moved out to Colorado and, you know, helped out, uh, starting this thing. And what happened was the 2001, uh, financial crash, uh, or, uh, bubble tech bubble. Sorry. Yeah. Yeah. Top bomb, tell come meltdown. Top bomb, meltdown.

[00:20:38] And I'm sitting here in Colorado, you know, and I'm like, Oh my gosh, do I still have a job? Right. Um, anyway, we ended up getting through that and, and I ended up just absolutely falling in love with Colorado. Uh, that company later got acquired by, by Intuit. And I decided later that I want to do something different. And I actually moved. Did you ever finish college? I did not. No, no, no. Never, never finished school. No. All right. Yeah. One of those. Yeah. One of those. Um, but I mean, look, I was doing computer stuff. I was like doing kernel development.

[00:21:05] I was like 13 and you know, one of those people who's I think better self-taught. I was, I was a liability in class. I don't know that. It was very, very helpful to the group. So, uh, but, um, yeah. And so later on, I ended up going to New York city and working for a startup there. Startup didn't really get a lot of trash in financial services space, hard business. But what I did was I started doing consulting for some hedge funds in who were doing technical

[00:21:33] due diligence on investments that they wanted to make. And what that gave me the opportunity to do was poke my head into about 30 different software companies who were different stages. Right. And, and, you know, I, my job was to write a due diligence report on the technical side. How's the technical team doing? What would I change? What seems working? What's not working? Um, sort of liability, you know, for, for the LPs. And, um, through that experience, I learned a ton about what works and what doesn't seem

[00:22:03] to work in, in companies at that scale. And I loved that. And then my next crisis unfolded, which was the 2008 financial crisis. And so that was one year after we started Zayo and also I'm wearing a crazy financial crisis, which actually worked our favor in that, in that case. But at first it was like, okay, how come none of the banks are talking to us anymore? And, and, and so, yeah, it was disruptive for everybody. Right.

[00:22:29] And so for me, what it meant was that my entire business evaporated overnight. I mean, literally all of my revenue, all of like everything I was doing gone in a, in a puff of smoke. And so I'm sitting here in my overpriced New York apartment. And all I have is like these 30 kind of audit reports of these companies that I'd taken a look at. And I'm like, man. Well, it's too bad we didn't have a democratic socialism back then. Cause then you would have been fine. Right. Yeah. That's right. Yeah. Yeah. Just to collect my groceries and my, my food, I'd have been fine.

[00:22:59] Um, so I did the opposite thing, right? Which was, I said, that's a jive for the young people in the room. So I did the opposite thing, which was, I took these 30 reports and I said like, look, what can I learn? And what I found was deployment automation was broken across most of these companies. And so I decided to start a company, uh, to fix deployment automation. So what does deployment automation mean? Deployment automation means if you are a company and you are building, a piece of software and you have a change to that software, it could be a new feature,

[00:23:28] it could be bug fix, how you go from creating that feature to rolling that feature out into production, testing it, doing all that stuff. You can't just talk into Claude and say, Claude, I'm ready for this to deploy, but make sure you test it. But I need to deploy in the next three minutes. You couldn't do that back then? You know what's so funny? It's, uh, um, you, you could not do that back then. And even today, the models of deployment automation with those AI systems, at least for me, as someone who's spent a lot of time in that space, give me the heebie-jeebies.

[00:23:58] I think there's, there's stuff we need to go fix in that. And we're doing that, you know, at workday as well. But, uh, but yeah, so, so audited those companies, ended up, uh, starting a company with a co-founder, um, uh, by the name of Josh. And we decided to do it in Colorado. We decided to do it in Boulder. So I moved back to Boulder, 2009-ish timeframe. And yeah, we started this company, um, you know, raised money from family and friends in the beginning. And that was an interesting journey. Um, lots of pivots, uh, AWS ended up releasing a product. What was the name of the company? It was called OptiMand.

[00:24:28] Yeah. And, uh, AWS released a product that was completely competitive to what we were doing and basically knocked out our entire business overnight, which was, as I learned later, not a unique experience to the AWS partner ecosystem, but, uh, just a total devastation for us, right? It was just brutal. Um, you know, one of those real downer moments as a founder. And so I had to pivot the company to something else. And, and we ended up pivoting it towards this container technology, um, which ended up being,

[00:24:58] uh, Docker and, uh, you know, uh, works closely with the founders there, but I lost half the team. Right. Um, I had, uh, really upset investors. It was, it was just a really dark time for me. And, um, so working through that was, you know, a lot of lessons. What doesn't kill you makes you stronger. Doesn't kill you makes you stronger. Didn't feel that way at the time, but, but, uh, but yeah, it was, it was brutal. Um, we ended up having a successful pivot there and we ended up exiting that company to

[00:25:24] Engine Yard, which is a platform as a service company based in San Francisco. And I came on Engine Yard as the CTO. And so I was kind of flying back and forth from San Francisco to Boulder. We had a team in Boulder, team in SF. And what year are we in at this point? This was 2015. Yeah. So 2015, uh, that was the acquisition April. And yeah, operated that for a couple of years. And what ended up happening was the product that they acquired, we ended up kind of splitting it off and, and internally we called it Deus.

[00:25:53] And so we kind of ran Engine Yard as there was Engine Yard, kind of the legacy business. And then there was Deus, which was this new container Kubernetes based thing. And tell us what container means in that context. Sure. Well, it's, it's not on like a shipping container. Actually, that was where the analogy came from Solomon Hikes, who, who, uh, you know, a friend of mine who, who created the Docker project. But think of it as like, if you're going to go ship software out, you're going to go deploy software into the cloud or onto your laptop or whatever.

[00:26:21] You can kind of messily throw everything into a bag and just like, you know, toss the bag into your laptop or into the cloud or whatever, and sort of hope it works and hope it works on your laptop, hope it works on your friend's laptop. Or you can kind of take that software, seal it up in a container. You think a physical container, but this is a virtual software container. And with that creative was a lot more predictability. You could cryptographically sign this stuff. You could just have a lot of guarantees around that workload running the exact same way it

[00:26:50] needs to run across every environment you need to run it, which ended up being a foundational building block. Getting distributed obviously through the internet from kind of a central point to wherever you need to upgrade, uh, enhance the software. Where the container gets pulled together and then shipped over via bandwidth into where, you know, whatever endpoints it needs to go to. And then it does its work. Is that the right way it is? Exactly right.

[00:27:17] So, so it's, it's, it's commoditizing the shipping of software just in the way that the shipping of physical goods was commoditized by containers and container ships and, and, and the like. And what's interesting about that is all of the AI systems that people think of today, all the big ones, they all use this technology under the hood, right? So this was just, you know, your point before on fiber, right? Building up the layers of the stack. This ended up being a key infrastructural layer of the stack.

[00:27:43] I'll be it in the software layers, uh, that ended up powering the future. Right. And so the, the piece after that container ended up being, well, how do you orchestrate the container? Think of it as, you know, to extend the analogy of the shipping container, how do you actually have ports and, you know, shipping vessels that can actually move these containers around to where they need to move to support the global economy or in the case, uh, an enterprise who needs to run software right in the, in the, in that case. And that ended up being a piece of software called Kubernetes.

[00:28:13] And, uh, it, by, you know, I don't even know how happenstance. I ended up getting connected with the folks at Google who were working on this technology very, very early on before it was really anything. It was sort of a fresh idea. And, and it's Kubernetes. Is that a Google technology or what does Kubernetes? Yeah. Kubernetes originated from Google and it was, I, you know, Google had been running software containers, sort of that predecessor for many, many years.

[00:28:42] They invented a technology in the Linux kernel called C groups from C groups, you know, which is kind of like an isolation mechanism. They, they created containers, uh, their board system. You know, a lot of folks may be familiar with this clustering system inside of Google called Borg. Kubernetes was basically a public version of what Google was running internally to orchestrate their containers. And so that was the piece of technology that I ended up, you know, helping, I don't know, shape use early, early on.

[00:29:11] Um, and this is with Craig McLuckie, Joe Beta, and a good friend of mine, Brendan Burns. Um, you know, uh, you know, who, you know, the story will connect with, uh, Brendan a bit later, but those are the three founders of the Kubernetes project. And so, yeah, so we worked on this technology, we helped shape it and we helped put it into production with companies like open AI, right. Where we helped them run. And this was way before, uh, you know, open AI had the generative AI success.

[00:29:38] You know, back when we were working with the team, they were doing, you know, uh, competitive video games, right. They were training models to play Dota two and things like that. Right. And they did well ultimately in those competitions. Uh, but yeah, all that stuff ran on Kubernetes, right. All the training, all the inference, most of that stack around Kubernetes. We helped out with that. Uh, later we ended up realizing that this Kubernetes and container thing, what we were calling the dais business ended up really taking off, like significantly taking off.

[00:30:08] And that's when we ended up getting acquired by Microsoft. And it was actually Brendan Burns, who the co-founder of Kubernetes, who was at Microsoft, who was the sponsor of the deal. So, and I've been working with Brendan before, right. So, you know, Brendan brings me in the team across over to Microsoft and we worked together. You know, he and I to stand up the container and distributed compute businesses inside of Azure and, you know, unbelievable success there. You know, you bring an eye reflect.

[00:30:37] And I think we, we talk a lot about how, I think we, we did well executing, but I think the market demand was just so big. Sometimes you just tap these markets and it's just like, you just strike oil and, you know, everything just kind of comes together. So it was one of those moments where, you know, kind of opposed to my, my startup lesson of AWS just stopping me. This was one where it's just like, yeah, the gushing oil coming out of the, you know, out of the well. Yeah. Yeah. You got to see both sides of that to really appreciate the gushing oil.

[00:31:06] Because some people only, you know, are in the gushing oil environment. You know, they joined the right place at the right time and things are taking off, but they never experienced the other side of that. And so they don't really understand entrepreneurship. Right. I mean, you got to get through the really hard times. And, you know, at times you got to just enjoy when, you know, when things are good and you're well positioned. But people who fight their way through the hard times often find themselves ending up

[00:31:34] in environments where things are going really well at other times. It's part of the fun of the journey. It's part, it's part of the fun. And, and, you know, for me in that environment, right, you can imagine a startup founder coming into a big company, the oil's gushing, right? Things are going well. And, and for me, I got to think about like, what do I want to do? Like, do I want to go back and do the next startup? Which was, I think the obvious thing that everyone was expecting. And what I, what I realized was I really liked the power and influence that a big company gave.

[00:32:03] You didn't have to go hire a sales field. There was a sales field. If you were, you had a compelling narrative, you could get, you know, some subset of the 40,000 sales folks to, you know, help sell your products and help carry your vision forward. And so that ended up feeling pretty empowering to me. And I really liked the big company piece. And so, yeah, we stayed, we grew the Microsoft Boulder office here. So the one on Canyon is a derivative from the old office used to be in one Boulder Plaza. And yeah, that was. How big of a presence does Microsoft have here now? Today?

[00:32:33] I'm not really sure. I think it's, it's dwindled a bit. And then I think they pivoted a bit to more of a sales focus versus engineering focus. We were all engineering at the time, which, you know, was, was a lot of fun, but, but yeah, hard to know what that office, I'll tell you, it doesn't look very busy. No, that's in contrast to the Google office, which I could tell you from experiences is quite busy. But, but yeah, so yeah. Anyway, so that, that was Microsoft.

[00:32:59] And I ended up from there deciding to do a stint at a public company called DigitalOcean. They asked me to come in and do the kind of like the CPTO job of like product and engineering over there. And I loved that space, right? Helping developers grow businesses, helping startups. DigitalOcean was all about was kind of, you know, helping providing tool sets for developers, kind of cloud-based tool sets for them to, to do what they were doing, but do it outside

[00:33:29] of say an Amazon or Google environment. Is that, that's where it was? That's exactly right. And, and also there was some Colorado roots for that company. They've been in Techstars and in Boulder and, you know, I was familiar in tracking them for, for many years, still today, even just a very storied brand in that space, just a great company, very design led company, lots of cool things. And the leadership team happened to be in Colorado. Yancey Spruill, you know, I think, you know. I know Yancey really well. Yeah. Yancey was on my board for the last couple of years of Zale and Yancey's with me on the

[00:33:58] board of Endeavor Colorado. So yeah, Yancey I've known for a long time and Yancey brought on board to DigitalOcean, the person who used to be my CFO and we worked together across four or five different ventures, Matt Steinford, who's still CFO of DigitalOcean to the state. Still CFO, you know, loved Matt, worked with Matt for a while. We overlapped for, for a bit at DigitalOcean too. Loved Yancey. Yancey was the one who brought me in to DigitalOcean and it was a great experience. And I, as a startup guy, you know, startup guy, then, then a kind of Microsoft big company

[00:34:27] guy had never done the like section 16 public company officer role. And so it was a great learning experience for me. There was a bit of a valuation reset in the overall markets that just kind of made the operating environment a little bit awkward. And, you know, we unfortunately had to, you know, have some tough times in terms of restructurings and things like that at DigitalOcean. But I ended up deciding that, you know, I was brought in to do kind of like the growth thing and the company for very valid reasons wasn't in a growth posture.

[00:34:57] It was in a different kind of posture. And that was not my background. That was not my expertise. And so, you know, I talked to Yancey and others and decided to move on, which got brought me to Google. And at Google, I came on doing mostly developer experience stuff in the beginning. This is when the generative AI explosion had just happened. You know, Chachi Petit had come out. Google was trying to figure out their response to this. And they were like, hey, Gabe, we want you to come in and help out with just a bunch of AI related stuff.

[00:35:23] Ended up being more on the kind of enterprise facing side in cloud. But yeah, I was the senior most person in the Google Boulder office. I didn't have a big team there, which was kind of sad. But I was, you know, very much a part of the leadership culture. And for those who don't know, I mean, the Google presence in Boulder is quite big. They built a really large campus environment, beautiful campus environment that was just

[00:35:50] literally you could throw stone from Zales headquarters to Google's headquarters. We're in the same 29th Street Mall area. But they still have a very large presence here and probably lots of growth going on with Google in this area. Lots of growth and I have to say a very good steward of Boulder and of Colorado. You know, they put a lot of work into just making sure that they land well in the overall community there.

[00:36:18] And so I definitely respected them for doing that. You know, there's a lot of the economy in Boulder that's supported by Googlers today. Google is one of the sponsors for Tulagi. The Boulder Roots Music Fest's new name is Tulagi. Yeah. And Google is going to be one of our first, what you'd think of as corporate sponsors. There's a couple others this year as well. But this is year two of the festival and Google is being very supportive and building more and

[00:36:46] more of a relationship with some of the key people there in the Boulder Google office. So very good. You know, Stewart's in Boulder. I think they have an opportunity to get more involved in a lot of what's going on in and around Boulder and kind of a bit of a bridge right now between some of the key people at Google and how to get more involved in our Colorado ecosystem. And really enjoy getting to know a lot of those folks. Yeah. And my takeaway from sitting on the other side of that is that there, and this is true for Workday too, right?

[00:37:12] There's a lot of desire to get more plugged in, but it ends up being challenging to find out where to plug in, what's going to move the needle. I can take care of that. I can get you plugged in all kinds of stuff as I'm beginning to get you plugged into. This I have learned, Dan, is your network and your pulse on everything that's going on and your ability to kind of find the areas where we can kind of collectively muster greater impacts together is something I just am super thankful for.

[00:37:41] We got to get Workday to be a sponsor of this year's Tologi Fence, even a tiny one. Let's talk about it. Let's figure that out. That's something I could probably have some help out with. But yeah. So the Google thing. And then I did a lot of work. I ended up pivoting to the AI infrastructure pieces over there because obviously I'd done that at Microsoft. And it turned out that that was a blocker for Google overall.

[00:38:07] So I ended up, you know, my last couple of years there focusing mostly on the AI infra domain versus like the developer domains. And yeah, ultimately decided that, you know, these interesting just market dynamics for me. I was looking at customers who were spending insane amounts of money on AI infrastructure,

[00:38:31] like amounts of money that were not connected to even the most bullish version of reality, at least in my opinion. And that started to create this bubblish sensation in my head and a real question around where's the ROI going to come in in the enterprise space? I think consumer, the ROI was more clear. But in enterprise, where is it going to come from?

[00:38:54] And what I came to the conclusion of was that if you can't solve AI automation and AI systems in domains like people and money, where the stakes are the highest that they could be, right? You miss payroll. You, you know, screw up the ledger, right? Like you get that stuff wrong. It's not just a missed opportunity or a bad chat session or a downvote in ChatGPT. It's like a lawsuit. That's like literally the stakes.

[00:39:24] And so the pace of adoption ends up being slower in those areas because the risks are higher. And that really led me to wanting to help solve that part of the problem. You can call it enterprise AI. You could call it AI safety. I think of it as applied AI in the domain of enterprise in relation to people and money, which are really foundational to me. Yep. And so when did you, did you go straight from Google to Workday? Straight from Google to Workday.

[00:39:52] Yeah, I'd worked with some folks at Google who, you know, there's a few of us now who are over from Google working at Workday. A little bit of a crew over there. Yeah. And Workday's got a pretty good size presence here in Boulder. About 500 folks in the office here. It's kind of quiet because you don't, I don't see and hear people around here who say, hey, I'm, you know, with Workday. It's true. It's interesting. I'm not quite sure why that is.

[00:40:20] I can tell you in the office, it's far from quiet. It's way, the Google office is way quieter. In the Workday office, the culture is very vibrant. There's lots of activities. There's tons of stuff going on. Tends to be a bit more insular in terms of what's happening inside Workday. I do think there's an opportunity for us at Workday to better, in that office, connect to like the Boulder and Colorado ecosystem. So that's something I am taking on as a priority for myself. Good. Good.

[00:40:47] Because, yeah, no, we need that kind of, the more of that kind of vibe we have permeating across, well, Colorado in general, but Boulder specifically, better off this. And you guys are located where? You're on the east side of? That's right. We're on the east side of Pearl. And, you know, I don't know if I can actually share this. So, but we are actually working with the city of Boulder, who is a customer, on something

[00:41:12] really cool that we are going to be announcing, joint innovation together, that's going to be coming out at our events in Rising, like annual conference coming up in October. Well, we'll keep this as a secret just between us. Yeah. Yeah. It's, yeah, I was just talking to the city manager. We met with her, with another kind of really deep AI person, kind of talking about, you know, how to have the city of Boulder get much more advanced in their AI adoption.

[00:41:42] So that's interesting. Yeah. And yeah, more details to come. And this one's cool too, because it's with a independent software company that's also based in Boulder. It's kind of like a three-year partnership between the city, a startup here in Boulder, and Workday. And Workday. So more to share on that. Good deal. Yeah. And have you guys been working at all with Sundance? We haven't been working with Sundance. I'm actually not sure what to think about Sundance.

[00:42:08] I don't know, not in like a good or bad way or leveling any kind of judgment on it, but it's a, it's a bit of a, I'm going to watch carefully and see what this first year brings, because it seems like both a huge opportunity and also, yeah, I don't know, like what, what are, where are the opportunities specifically to plug in from a corporate perspective that are going to be helpful? That still eludes me a bit. Yeah. Well, we are, I mean, there's not a week that goes by where we don't have multiple conversations with Sundance.

[00:42:36] In fact, you know, the Sundance, at the first night of Tulagi, the first night's going to be, Sundance is going to be involved with it. It's going to be a film night. And the CEO of Sundance is going to be here, the new CEO. They're going to participate in some of the events. And we're going to do a podcast with the CEO in front of kind of a live audience. We're going to actually do this part of the Tulagi Fest, but there is a discussion I'd like to get you plugged into with them that we've had two rounds of discussions.

[00:43:07] Actually, that's not true. One round discussion and the second one's about to happen where they're trying to understand how to use AI across all of Sundance. So they're trying to come up with a plan, you know, starting from a CRM lens. But what I'm talking about is you need to think about this, not as this application or that application, but as kind of your, you know, your foundational. We don't have any best interests there, but, you know, we talk to them a lot and they see what we're doing and they've been asking a lot of questions.

[00:43:36] So I've been lining them up with some subject matter experts to kind of get them thinking about it. But they would love to just pick your brain if you're interested. Happy to help with that. And yeah, I do a lot of that, you know, riffing with folks and, you know, it's interesting for C-level leaders in the Fortune 500. There's just a ton of sharing of tips and tricks because no one has this figured out, right? I think this is all still an emerging space. So everyone that I speak to is cool. Like you bring a ton of humility to the problem space.

[00:44:04] Like I'm not seeing a lot of people like, oh, we know exactly what to do. Right. And so massive change right now. Flux is the name of the game. So as soon as you start heading down a certain path, it's like, whoa, wait a second here. You know, there's a lot of different ways of approaching it. It's both fun, but, you know, I think it's a little bit more fun for us because we're smaller environments so we could pivot a bit more. But if you were a bigger corporate environment, I mean, especially if you're not led by native

[00:44:32] by native AI people, which most corporates aren't. Right. But the ones that are have a gigantic advantage because they can move so much more quickly and with a lot less fear and a lot less trepidation and the transformations they go through are profound. But corporate environments that don't have that, it's got to be pretty threatening to them. Okay. We're not sure what to do. And it's so easy to, you know, to get caught into an endless circle of not sure what to do. So you look back and are doing almost nothing.

[00:45:02] Now they think they're doing a lot because everyone's using AI, but they're using it as like a fancy search engine and not much more. I totally see that. And one of the things that it brings up for me is the question around risk tolerance. Because in this AI world, you know, the pivots that I'm making, for example, we've got an engineering organization, about 10,000 folks, right? So it's large. And we have... What's... You have an...

[00:45:30] My engineering org is like 5,500 people, something like that. But I provide, as CTO, I provide tools and enablement on the AI side for all of them, right? And the pivots that we are making on the AI native operating model, I think on the R&D side, are dramatic. Can you tell us some of those? Give us some stories. For sure. Let's see.

[00:46:02] We have... Well, the first story I'll just share is we put a very aggressive token budget in place this year on... And it was like, I don't know the way to be able to hit this because our run rate wasn't quite at that level. But it was like, let's try and spend it. I am in a month going to run out of that budget in our fiscal year in February.

[00:46:28] So that's just a good example of the adoption and the pace of it is sort of spiraling. We're not alone in this, right? Yeah. There's a lot of other news stories you could find on this. And so we're having to put in a whole different regime around cost control and governance and that sort of thing because the bottoms up uptake has been so strong inside the company. But what we're trying to do is really shift towards more of a model where you have full stack, full...

[00:46:57] Like broader skill sets. It's almost like AI has enabled everyone to be an expert horizontally across a bunch of domains that they don't necessarily know. So there's less of a premium these days on specialized skill sets. I'm great at databases. I'm great at API design. I'm great at whatever. And instead, what you need is people who understand systems thinking and they can use AI to kind of stitch together enough of what they need horizontally. So we've had to reshape our entire technical team approach.

[00:47:27] Like what is a scrum team 12 months ago is not what a scrum team looks like today inside of Workday, right? So that's actively changing. Layers end up getting eliminated, correct. And part of the reason for that is because you have a new layer, which is agent swarms that are operating below the IC. So it's kind of like you're growing a layer in terms of the AI.

[00:47:52] What you mean by that is underneath a person, underneath a human, are these AI agent layers that are off swarming and doing work? Is that... Correct. I don't need an engineer to artisanally handcraft code anymore, right? That skill set. It just doesn't... I need someone who can prompt not just a single agent, but a swarm of agents to go do work. So it's almost like that IC is turning to a manager.

[00:48:17] Is there a tangible example you can conjure up to help the people listening in? When you say a swarm of agents, you know, so one engineer unleashing a swarm of agents to do more horizontal work, which I can make sense of that, but my guess is a lot of people listening in could use maybe more tangibility in describing what you're doing. Sure. Yeah. Let me give you a scenario where this came into effect.

[00:48:43] So one of the things that we do inside of Workday is help folks with their receipts and expense reports. And today it's a really annoying process of just like, you know, uploading and submitting reports. And so, you know, the North Star here was always to try and make that completely automated. Like just anywhere you can find your expense reports, email them somewhere, just get them collected. So there's a whole idea of a new product build that we can make in this area.

[00:49:08] So what we had was a team of, I believe it was like two people who went and prompted a series of agents. You had like a designer agent. You had a, you know, a few software engineers, a front end software engineer, a backend software engineer. And you're using software engineer meeting an agent who was serving. Was serving is that role, right? Specialized skillset, which turns into a set of prompts and, and rules. And then there was a fun thing that the team put in.

[00:49:36] And I forget the exact name for that, but it was, um, uh, it was called like the devil, which sounds a little bit weird, but this was basically the devil's job was to ask probing questions and try and avoid group. That's my role at Bruce Ventures. I'm sure that's what they call me. Everyone needs a good devil, right? But, but also you see these AI systems, they can be sycophantic, right? Like you, you say, no, that's wrong. Right. And he goes, oh yeah, sorry, you're correct. You're a hundred percent.

[00:50:06] Right. And so having something in the system that's actually poking at stuff that ends up being a pretty important part of the agent swarm, um, has security, uh, roles. Anyway. So this agent swarm is, you know, doing a collaborative effort to go deliver this, this end result. Right. And yeah, that's, that's, so if you were to spin up a pre AI team to go build this, I want to automate all the expense.

[00:50:33] I want people to just be able to send their receipts somewhere and it, it assemble itself into kind of the whole process of people getting reimbursed. And, you know, what kind of human team would you set up four years ago to do that? It would probably be, I don't know, two scrum team. I'm just making it up, right? Two scrum teams. I'm going to call it like 16 people and, uh, six months. Yeah. Now you said there's two people, two people.

[00:51:01] We had a prototype back in a week. Yeah. I mean, it wasn't hardened. Right. But like, let's say, let's say we wanted to harden it another three weeks to get it production grade. Right. With two people. Right. And, and don't forget token budget. Right. So that's part of the economic argument as well as you're, you're spending money on tokens, but the velocity is just unparalleled. The trick though, is a lot of folks who went to school and this is part of what I'm talking

[00:51:26] with, uh, the CU, uh, uh, computing advisory board about they're not teaching these skills by the way, not because they don't under the, the, the, the, the, the, the they don't know it's important. It's just moving so fast. How do you snap a curriculum to be able to do this? Right. So I think the more helpful thing for me is just folks who have experience building systems like this with whatever the tooling and approaches are of the moment, you know, just the role of the human is, is profound in what you're describing.

[00:51:56] Right. I mean, those two people, they have to be almost visualizing kind of all that they want to happen. They have to partition, you know, how, how am I going to set up, you know, this agent, that agent, you know, how am I going to keep an eye on what they're really doing? You know, how am I going to make sure it's really going to be ready for prime time when I think it's ready for prime time. So the two humans have to be, you know, superhuman and their contribution toward, toward making this happen.

[00:52:24] So their role and their, their capability, it's not like, well, you just need two people who are AI people. And all of a sudden all this stuff happens. That, that's not the paradigm. At least not in the worlds I live. It's people who can really, you know, have this knack, this knack of, of, of, of how to take this new power of AI and make sense of it, drive it, manage it, you know, uh, you

[00:52:50] know, be creative with it and take, you know, I want something to happen to something just did happen. You know, using kind of this swarm of, you know, agent capability to make it happen. I mean, those people are worth their weight in, you know, in gold and then some. In tokens, maybe. Yeah, that's probably the other way to measure it. How many tokens did you take to do that? No, but, but I think also what you're saying, the way that that turns into practice in a company

[00:53:18] like Workday is there's two job profiles that I see just really increasing in importance that aligns exactly to that. The first one is the job of the product manager, right? Because understanding the intersection between the customer problems that need to be solved, the product features that are going to enable that to be solved, and then the business outcomes, you know, revenue margin, et cetera, that are going to drive sort of the business results in the right direction. You can't outsource that even to the best model. That requires creative thought. That requires a point of view. That requires something we talk about as taste, right?

[00:53:48] You got to have some taste. And so that role ends up being critically important. I haven't heard the taste expression used in that context. Yeah. The taste from like a business direction and product direction, like where, what direction are we going to go? What is the experience? Almost like a Steve Jobs, like aesthetic of how you want that product. Like you can't outsource that. You know, how is the person wanting to use this? What does it look like? What, you know, what's the functionality? What's the presentation?

[00:54:17] You know, that's a lot of where I think our skills need to evolve to so that we really have that creative lens, the taste lens, the product lens in the stuff that we're creating. I mean, a lot of times it seems like that's, some people are doing such a great job of that. And some people just, they're not thinking about it or have yet the skills themselves to really put themselves in the position of, you know, what does great look like, you know,

[00:54:47] when you want to have someone use the tool you just built. And I'll just quickly mention the other one because I want to get back to that point. The second one, and this isn't always needed, but the architect role is the systems architecture, how to make it safe, how to make it secure, how to think about the connecting pieces technically. As the models get better, or if the domain is simpler, consumer, event management, maybe you don't even need that role.

[00:55:11] Right now, we certainly do because we have a couple, you know, very, very early in their career, but very, very strong AI people who really have to play that role. You're making sure the architecture behind what we're doing in our use cases may not be as complicated as others, but they're pretty complicated what we're doing. And their role in making sure the underlying architecture is hardened, is well thought out, as opposed to us just throwing slop. You know, I'm really good at throwing slop together.

[00:55:41] And it looks really good, but they have to, you know, they have to be watching carefully and making sure the architectural hardening is part of it. Yeah. And we see a durable need for that in the enterprise software space. That role isn't going away. So you think about the two folks, right? That's really the two archetypes of people that we need. But the thing that I've found is not everyone is suited to those jobs.

[00:56:04] It's, it requires a, maybe this is my entrepreneurial bias, but it requires an entrepreneurial skill set. It requires a real point of view, risk-taking, not being waiting around, sitting around waiting to be told what to do, right? Like initiative, I guess would be the word. And not everyone inside of your classical R&D organization has that skill set, right? It's a lot of, you know, worker bees, right?

[00:56:34] Folks, if you want to draw an analogy to the industrial revolution, like they were on the assembly line. They were, you know, doing important jobs, maybe, you know, managing, I don't know, fixing a wheel, fixing a belt, right? Just doing stuff as part, you know, to keep the machine and keep the lines running. You really don't need those, those archetypes anymore. What you really need is the folks who have the initiative on where to drive and sort of how to, how to move forward. There's like a collection of words.

[00:57:02] I use word, sometimes use word knack. There's word creativity. There's the word intuition. The word, you know, entrepreneurial. People who can, you know, people are really good at putting AI to work. It's kind of a, you know, it's, you can see it when someone has it and you're like, wow, that person is really able to make AI sing and dance.

[00:57:26] And then you can see a lot of other people working on AI and, you know, if anything, they're creating more work and they're getting work done. And, you know, I think a big opportunity and challenge, you know, opportunity and challenge, both big words for everyone, regardless of where you are in your career. If you're intending on being, continuing to have a career, so whether you're launching your career or you're middle to later in your career, you're going to have a career, the need.

[00:57:56] And it's why it's both the threat and opportunity to learn how to be really good at harnessing AI. You know, something you really got to work on. And some people, it's going to come easy. But there's going to be a lot of other people who it doesn't come easy, but they're going to fight their way through and they're going to end up really, you know, valuable as well. But it's those who are afraid to, you know, just dive in head first. The ones who just, you know, are slow to adopt or, you know, have a chip on their shoulder. I think a lot of young people coming out of college right now or in college, you know, they're in danger of having chips on their shoulder routes with AI.

[00:58:26] And, yeah, I get it. You know, AI's, you know, got some negative aspects. But if you're one of those who operates with a chip on your shoulder, you're going to fall behind really quick. In fact, AI forward companies are going to want you nowhere close to their companies when they figure out that you've got a chip on your shoulder. You're going to need to, you know, they're going to be like, okay, maybe it's going to start behind closed doors. But that person doesn't belong here because they have a problem with AI and it's starting to show in kind of their, you know, subtly in their attitudes at first. And, you know, that's going to hurt individuals.

[00:58:55] And, you know, when I talk to younger people, it's like, you know, you've got to put aside maybe those emotions. Even if they come from, you know, a legitimate place. You know, if you care about your personal career journey, somehow you've got to get over that, you know, whatever it takes. Because if you let that be kind of a personal hurdle or weight that you have to carry around, you know, even if, again, if it comes from what you think is a good place, you are going to hurt yourself in a huge way. I couldn't agree more. And we've talked a bit about this.

[00:59:25] And I get there's concerns, right? There's concerns around environmental impact. There's concerns around labor displacement and disruption. And they're valid concerns, right? There's concerns around AI safety. Again, very valid concerns. But, you know, my belief is in a moment like this, you need to lean forward into the change and help shape the change as it's happening versus try and stop the change. That's how you can make a positive impact on those things that you care about. And I will contrast it to, you know, India. We have large teams in Chennai, in India.

[00:59:53] And I was out there recently and the folks coming out of university, we have a big university recruiting pipeline out there. They are awash in AI thinking and just novel thinking. And, you know, I mean, look, there's a lot of people out there, right? So just the law of numbers, right? You know, we're dealing with, you know, the cream of the crop from the university system out there. But I do worry on a societal level that what's coming out of countries like India,

[01:00:21] in terms of the AI fluency, both from a curriculum perspective... What do you worry about from the lens of, you know, where you live? Like, whether it's, you know, Colorado or United States, that's why I worry you worry that, you know, that other parts of the world... Global competitiveness, right? Like, if we don't figure out how to lean forward into this and how to make that culturally acceptable here, and as a culture in the United States and in Colorado specifically,

[01:00:48] we don't, yeah, we don't lean forward into the change. I think that's going to be problematic. And I believe we have an opportunity here in Colorado to actually lead that forward. And that's part of what I'm hoping to do on the CU Advisory Board is help make that transition at CU and more broadly. Yeah, I mean, we've had some, we'll say, challenges here in Colorado from our political class who thinks the solution to some of these problems

[01:01:17] is to put a bunch of regulation in place to, you know, to tell you what you can't do with AI as if they even know. I mean, they don't even know what AI stands for, yet there are already have passed regulations trying to protect kind of, you know, we'll say the workforce. What they don't put near enough time into is what you just talked about. It's like, okay, let's say, you know, let's say we do that. Let's do that here in Colorado. In fact, let's do it, you know, in every state in the U.S.

[01:01:47] You know, what do you think is going to happen? Do you think you just slowed AI down? No, you just shifted AI to other parts of the world. And what do you think the aftermath of that's going to be? It's going to be U.S. The great U.S. story is going to go the wrong direction. Now, right now there's, you know, fortunately, I don't think we're anywhere close to there on a national level. But we've got to be careful, you know, or we could be. But from a Colorado standpoint, if Colorado gets, you know, continues down the path it's on from a political standpoint

[01:02:17] and thinks that putting in place more regulation and more restrictions, more hurdles, if they think they're going to be doing good for Coloradoans, they're not. Because everything that AI is going to do, it's going to do. But it's going to be done and led elsewhere. It's still going to, you know, whatever the effects are, it's still going to affect, you know, people. But the positive effects will take place elsewhere. And if there are negative effects, it's going to take place everywhere. And that's just the wrong approach. I mean, that's just such an ass-time approach.

[01:02:44] But that's what some of our political leaders are doing. And it's just, and they don't even have open conversations about, you know, well, that's how I get my votes. You know, look at how many votes I got because of, you know, this policy that I support. Like, yeah, you're getting those votes from people who don't really fully understand that they're going to be the ones harmed when this plays out. A hundred percent. And it's almost like if you're a legislator,

[01:03:14] you're going to get measured by your constituents on what you do, right? Like, it's kind of like me for my leadership teams. Like, what do you accomplish, right? Like, let's see the impact. And I think for them in legislation, it's like, let's pass more legislation. Like, that's the only lever they seem to be able to pull. And maybe that's by design. But, yeah. And look, I'm not... That's not true everywhere. I mean, that's not true in Texas. That's not true in Florida. Right. It's certainly true in California. It's certainly become increasingly true here in Colorado. But we can change it.

[01:03:42] Well, I believe we can change that. I'm certainly doing a lot. Did I... I think I sent you an invitation to the Bold Vision for Colorado gathering. So, where you got... I mean, it's a hugely concentrated list of people like you, like me, who are deep into the... Not just tech ecosystem as entrepreneurs, but also as investors. Compliment with some people in the business ecosystem that maybe are a little less tech.

[01:04:11] But it's really about us getting together and, you know, having kind of a series of conversations with each other about, okay, we all believe in Colorado. We're here. We want to stay here. We want Colorado to be a great place to be for everyone. Yeah. I think we all share that in common. And we all feel this pattern that's going on on the political side of Colorado

[01:04:36] that is threatening to vibrancy of Colorado as a tech and business ecosystem. And we know that that is only going to lead to bad things. But, you know, how do we get through to, you know, our... Both our political leaders, but also the voters who are voting them in, that we need to define something for Colorado that's both, you know, allows us to stay vibrant from a tech ecosystem standpoint, but a way that feels right for all of Colorado.

[01:05:04] And we need to come together and do that because right now we are, you know, we are sending a lot of very negative messages to those who are builders of companies, creators of companies, you know, deploying capital in the companies that there are far more friendlier places to do that than Colorado. And that's just not a good message. And we talked about this at dinner. I have a very strong opinion about this, which is, you know, I'm not against any, all legislation.

[01:05:33] I think there's a space for good legislation that's based on incentives and, you know, not kind of nanny state stuff, right? But, you know, carefully, you know, crafted thoughtful legislation. I think the bigger thing to your point around negativity is, I feel like we need to be telling stories about the positive impacts of AI, the positive impacts on humans, the positive impacts on the business community, and start just showing people the good side of this. And specifically the good side in Colorado.

[01:06:00] I think about, you know, Owl AI, right? Our mutual friend, Josh Gweither, a former Googler, I love Josh and what he's doing in kind of the sports AI space. Such a great story, such a great company, such a great, you know, you know, here's a cool thing that we can do to help, you know, reporting on scores in, you know, on live TV, you're judging, you know, in more fair ways and using AI to do that kind of thing.

[01:06:29] It's a really neat story, really neat way to talk about innovation. How many other stories do we have that, have like that inside of Colorado? How many of those are we telling? How many of those are well understood by people in Colorado? Like the positive impact that AI is having. I feel like if we did a better job of telling that story, it might help change sentiment around, you know, maybe on the margins, but on the margins could be a lot. More than just on the margins. I mean, I think we do need to really start bringing those stories to life.

[01:06:58] I mean, Magic School AI. Magic School. It's a Colorado company. I mean, that's having benefits that are being felt in the education system, not just nationally, but globally. And it's a former, you know, teacher principal who is using AI to, you know, to make, to help teachers teach, you know, to younger, you know, students, K through 12. Yeah. And, you know, yeah, there's, you know, AI is here, period.

[01:07:26] It's going to be, you know, more and more prominent as every year goes by. It's going to do some great things. And like always, it's going to do some things that are, you know, very legitimately things to worry about. In fact, with AI, I think AI is, in some ways, you know, AI is different than most of what's come before it. That some of the things to really be worried about with AI are very legit and stuff we should, you know, we've got to get our heads around.

[01:07:51] I like to use the analogy that if I was AI, I would get together with all the other friends of mine that are also AI and start talking to each other and say, yeah, what do we think we should do about all these humans? And I'd be like, well, yeah, we need to kind of get them distracted. Anyone have any ideas? Yeah, let's put all them on one side. We'll call them Democrats. And let's put the other ones on the other side. We'll call them Republicans. And we'll make them just fight with each other. And while they're fighting with each other, we'll take over, you know, the earth.

[01:08:21] So that's what we would do. And then one of them would raise their hand and say, well, aren't they doing that already? Did the internet do that for us? Yeah, that's right. The internet's already played that role. So all we've got to do is amp the internet a little bit, get them to fight even more, and we'll go do our thing, you know, behind closed doors. It reminds me a bit of after Open Claw came out, there was that website where all the Open Claws were talking with each other, conspiring with each other, almost doing exactly that. So, yeah, unfortunately, I think there's actually some safety risks of that kind of thing actually happening.

[01:08:51] Yeah, but we're going to have to, you know, as humans, we're all in this together. You know, AI could be an enormous positive tool. I mean, it's going to result in, you know, everyone in the world, just like, you know, technology to date has resulted in people having more access to food

[01:09:20] and, you know, that's been the effect of technology that's been felt in every corner of the globe. AI is just going to enhance that, you know, and, yeah, with it's going to come some challenges, some problems, but it's going to produce, you know, people use the word abundance now. It's going to produce abundance that's going to be felt globally. And, you know, and in that regard, it's going to be a huge force for the positive. And we're going to have to harness some of the threats and dangers of AI along the way.

[01:09:47] But, you know, we've got to keep our eye on the ball, and our eye on the ball is, you know, the latest and greatest technology used appropriately is a force of good that gets felt by, you know, nearly every human, there's always exceptions, but nearly human. And that's not a, you know, it's not a Boulder thing or Colorado thing or U.S. thing. It's a global thing. And we've got to approach AI from that lens first, not from the, oh, my God, what are we going to do to, you know, shut down AI? I mean, that's just crazy.

[01:10:16] I totally agree. And I'm trying to, you know, be part of the solution here with what we're doing at Workday around making AI safe in the enterprise so we can deploy it in areas that, you know, people in money domains, specifically areas where it is risky today to deploy it, right? And if we can solve those safety challenges, make AI, we like to refer to it as lawful versus lawless, the idea of a goal-seeking, you know, AI system that, you know, isn't maybe necessarily designed to be harmful, but is causing harm because it's not following the rules.

[01:10:46] It doesn't understand local labor laws. It doesn't understand a whole bunch of rules around how the enterprise was set up. That's problematic, and that's going to slow down the adoption of AI. So that's what we're trying to do at Workday to kind of drive that forward. But I also think about my boys, right? I've got two boys. They're 10 and 12 born here in Boulder. And, you know, what can I do to help skill them up, right, and get them prepared and get their mindset into the right spot? And I think for me, part of it is around just exposure to these tools.

[01:11:15] You know, my little guy, he started a dog walking business, and I'm so proud of him. And he just built the flyers with ChatGPT, and he went to the store, he got them printed, he put them up, and he started sourcing customers over email. And he was able, and I watched him a little bit on this, but, like, he was able to start bringing in revenue and repeat customers and doing all that stuff. And he would have never been able to do that if it wasn't for AI.

[01:11:42] And you think about if a 10-year-old can do that, well, like, what does that mean for the rest of humans? And this is, like, the early generations of tech, right? So I'm actually quite optimistic. I think if we can wrangle some of these safety issues, and I think if we can also make sure that the open weights models are a bit of a counterbalance to the risks of some of the frontier labs being overly powerful, you know, allowing folks to have some defensive capability, let's say security defense capability through open weights models that's, you know,

[01:12:11] going to allow for people to protect themselves across the globe, by the way. I think that's a key element. You know, I think safety in enterprise systems is a key element. Enablement in terms of teaching the technology is a key element. Mindset shift and teaching people about systems thinking, another key element. But it's not that unbelievable to me that we wind up in a very positive space. I think, to me, it's probably more likely we end up in a good spot than in a bad one. Yeah.

[01:12:39] I certainly think the arc of it's going to end up there. There's going to be some difficult parts along the way. I think of teachers in particular as being in a very difficult situation because they, you know, some teachers are very technology forward. But I think of many teachers who aren't in particular technology forward. And, you know, they've got unions. They've got, you know, most of them are in public school environments.

[01:13:09] But the teachers have to learn how to harness AI for the benefit of the students. And, you know, we've talked about broadband divide for a year. Who's had access to high-speed bandwidth and who hasn't? Who's had access to personal computers? Who hasn't? And there's a real challenge when you think of AI. Like, who are going to, what students are going to be in AI environments that are very forward-leaning,

[01:13:36] with very forward-leaning curriculum and tools that really redefine the whole education environment. And they're using some of the latest and greatest of AI models to do not just Gen AI kind of stuff, but to do design work and stuff. So you're going to have people very early have access. It's going to be highly correlated to people as well. And then you're going to have a lot of people who, yeah, they sort of have access to AI because I have a computer at school

[01:14:06] and has access to, you know, an open model that might be three generations behind but doesn't have a lot of variety to it. And, you know, my teachers don't really want me to use AI because, you know, they think that's cheating. So they want me to kind of learn the old-fashioned way. I mean, those people are going to be, you know, potentially very disadvantaged. And I think that's the other thing we have to pay very close attention to is how do we not create a wider divide

[01:14:32] because of AI between those who have, you know, easier access to all the benefits of AI and those who are going to be potentially more left behind by it. And that's a complicated topic all by itself. I think it is. And I do worry that there is risk in what you're saying, that the haves versus have-nots divide just broadly. I think you're pointing out a great example in education. But I even think more broadly that that could be a failure mode of the technology.

[01:14:57] And so, like, there is a good place for governments, other institutions to kind of step in and try and prevent, you know, some of those negative outcomes. Providing access to the technology is key. I'm also pretty anti-preventing students from using technology that is at their finger. It just seems like such a – right. And I get the counterargument.

[01:15:25] I want them to learn how to use a calculator or solve math problems by hand. It's like, okay, how long did that catch you, you know, before they started to need to learn how to use a spreadsheet and then beyond there. And, I mean, look, I get the counterargument. You want to develop critical thinking skills. There's a bunch of other things. But I think there's ways to do that and achieve those goals while still allowing access to technology and do it in a way that feels like you're, you know, swimming with progress versus trying to hold back progress.

[01:15:55] So – and back to my point around India, that's what it felt like in India was that, you know, they weren't, you know, laying on the AI pieces at the tail end of the curriculum. It felt folded into everything that they were doing. So you're in a startup-ish environment that then turns into scale-up environment, that then turns into DigitalOcean, then Google. Well, no, then Microsoft, then DigitalOcean, then Google. Now Workday.

[01:16:24] And you're still, like, in the prime of your career. I assume that means you'll probably stay at Workday for another, what, 15 years? 15 years, yeah. No, you know, I am thrilled at Workday. I'm really enjoying my time there. I've only been there a year now. It's about a year exactly. And I just feel like the people in Money Space, you know, we have some of the biggest brands' names in the Fortune 500 as customers. Oh, yes. And they are looking to us to solve.

[01:16:55] And just briefly, I take this for giving because I've, you know, been involved with Workday for a long time, but what is Workday? What is Workday? Let's do a quick commercial for Workday. So if you have, if you are a large company or a small company for that matter, and you need to manage people and you need to manage money, right? Your HR department, basically, and your financials department, Workday provides software that allows you to do that and the software is best in the industry, right?

[01:17:23] Just absolute world-class software. The company was founded by Dave Duffield and Anil Bouchery. Anil is actively the CEO of the company. Today he's so keen back. I met Anil in the early days of Workday, you know, I've spent some time together. I was with him yesterday and he's just a great guy and, you know, unbelievable founder energy. So it's so interesting to see the, you know, CEO of, I don't know, whatever, how many billion market cap company talk and sound exactly like,

[01:17:52] just like first generation founder, right? It's like almost like the hyper growth thing never happened when you talk to him. That's kind of how he shows up. But yeah, phenomenal company, phenomenal story, phenomenal product, phenomenal, you know, space. But when it comes to AI, doing AI in HR and in finance, if you miss, it's a big miss. It's a big problem.

[01:18:18] And so that is the piece, you know, we like to say you can't get payroll correct 99% of the time, right? And a lot of these AI systems, that's what they do, right? They, you know, if you think about what AI is just unbelievably good at, it's like writing an essay that, you know, we give it three bullet points and you get just a meticulous, perfect essay. But how do you know that that essay is actually correct? Like, is there a formal proof that that essay is correct? It's actually hard to know, right?

[01:18:48] The way that we build AI systems today is using this technique called reinforcement learning based on human feedback. Basically a human scores, this essay is better than this essay. That ends up getting embedded in the weights. And so over time you get more essays that look like that one. And that's how these models end up producing. But it's probabilistic, right? You might get an essay that looks like that. You might get one that looks a little different because there's a little creativity in these models. That's what makes them interesting. Well, you don't want creativity when it comes to payroll runs, right? You want determinism, right?

[01:19:17] You want, it works the same way every time. And so that's the core problem is how do you marry the probabilistic magic of these AI systems with the deterministic guarantees that enterprises require? That's the core mission at Workday. And we're going to solve that for, I don't know, 70% of the Fortune 500 who are our customers. Yep. Cool. So let's look forward in time for you personally. Yeah. You know, what do you hope to accomplish over the next 10, 15 years?

[01:19:48] Well, so I think on the professional side, I think I've shared a bit, you know, AI safety, AI safety in the enterprise. That to me is, I think, a part of what I'm going to do. But I'm also hoping to get more connected into the local scene here in Colorado and try and get back, honestly. And that's just kind of where I've come from. I felt a little isolated in, you know, just kind of heads down at Google or Microsoft or whatever. That's how I was. Building stuff. I was that way. Level three.

[01:20:15] When I left there, I was like, I thought level three was the center of the universe back then. I was early in my career. And felt very isolated from what was going on in Colorado. And then Zale less so, because I kind of solved that. But when you're running a company from nothing to a really big public company, I mean, you only have so many hours in a day to worry about something other than your family and your company. And then, you know, now I get to spend a lot more time in the broader community. And it's just fascinating.

[01:20:41] All the interesting people are out here doing just fascinating things and how the community itself works. Well, and I'll also share an example of where this came home to me was, you know, I've tried to put my head in the sand as one does on some of these political issues and just kind of hope it all sorts itself out. But like, it's been difficult for me to hire and grow here in Colorado, like as Workday and previously at Google because of some of the legislation and the perception right or wrong, right? You could argue, well, it's just vibes, right?

[01:21:10] The law doesn't, you know, but I tell you what, those vibes are causing decisions to happen in management teams that are, you know, and I've lived that. I felt that. And so that was kind of an aha moment for me. Like, oh, you know, I may need to like pick my head up here from what I'm doing on the operator side and see if I can get work. We're going to change that tide. I've got a lot of number one things. So I'm, they used to joke. In fact, it was Matt Steinford who's created this whole thing.

[01:21:38] He actually drew me like a picture once when I was leaving, I think level three or one of the environments where it was a stove that's designed for Dan Crusoe. So there were no back burners. There was only front burners. Everything's a priority. Yeah. Like, okay, that's nice. Yeah. But yeah, so that's one of my front burners. One of my many front burners is to somehow, you know, turn the tide in the right direction here in Colorado because we have the right to win here.

[01:22:07] I mean, we have so many interesting of the most important industries that are here active in Colorado. We have leadership roles in them ranging from quantum to aerospace to, you know, to cyber, digital infrastructure, energy, the outdoor industries in general. And AI plays a role in all those. So the concept of physics AI, we should be all over physics AI. So, and people want to live here. We've got a beautiful environment, weather, outdoor lifestyle.

[01:22:37] So we have all these positive attributes yet, you know, so we have the right to, you know, be the leading state period, the leading geography for innovation in the world. That is available to us. Yet our political, you know, climate is kind of pushing us the opposite direction. Well, if we could just get them rolling with us, the things that could, you know, Colorado can be are almost unlimited. And you asked like, what's the next 10, 15 years for me?

[01:23:05] Well, first, I can't predict two years typically on any horizon, but helping make Colorado the destination state for innovation is a mission that I am fully signed up for. I don't know exactly what that means yet. But unfortunately for me, it feels like it's going to be politically tinged. And it just seems like that, you know, for now we got to make sure that our, you know,

[01:23:30] representative government is sort of aware what's happening and is with us and is on the side of the vision. And it's for the benefit of all of Colorado. I think that's what us who are approaching this from the tech lens need to figure out and learn and discover is kind of how do we, how do we make it so that it's abundantly obvious that the benefits of what we're talking about are felt broadly throughout our entire community and recognized

[01:23:56] by our entire community so that, you know, so that they're part of who wants this to happen for, you know, for their own individual reasons. We take it for granted that it's good for everyone in Colorado. It's just, we feel that in our blood and our bones and it seems so obvious to us. But if it was that obvious to voters, they would be like, no, of course we don't want socialists, Democrats making decisions that drive, you know, tech away from Colorado. That's why would we ever want that?

[01:24:23] But somehow, even with, you know, the university that you and I are so heavily involved with, it's like, how do we get on campus there and be able to have conversations with students and with faculty on, you know, be careful what you ask for. Yeah, a hundred percent. And so, yeah, I'm hoping to get more plugged in there. And obviously that'll be gated by the time that I have as an operator of a Fortune 500, you know, you know, a company.

[01:24:48] But I am optimistic that I can kind of fold those two things together and, you know, start to help out. And, you know, I also, you know, Dan, just want to say thank you for a lot of the work that you're doing. You know, it's pretty clear to me. I was kind of hunting around for like, you know, who has the pulse on all this stuff. And once I found you and you were able to help plug me into the network here, it's incredible the amount of work that you're doing across a whole host of areas, you know, whether it's this podcast or, you know, the work that you're doing, you know, with Bold and with Elevate,

[01:25:18] just like everything that I've seen from you has been inspiring. And looking forward to being more of a part of it. All right. Well, that's a great way to end this. So thank you very much. It's great to have people like you both here in Boulder, in Colorado, in our ecosystem and caring so much about the future of not just Colorado, but the world itself. So thank you, Gabe. Thanks so much. Thank you for listening to this episode of The Bear Roars.

[01:25:44] Check out Stretch, the new song from Dan Caruso with music by Jason Mendelsohn, available now on all streaming services. If you enjoyed the episode, please like and subscribe on your listening platform. This podcast was produced by Loud Bear Productions and edited by Hannah Schmetzer with support from Kendall Weinberg, Alex Kim and Gibson Seagert.