Nobody tells you AI is mostly plumbing

Thank you so much. It's great to be here.

Sure. Kevin has a long and winding road through entrepreneurship, I will say. I started my first company in business school after I'd done a lot of stuff before then. first company didn't go particularly well, that happens a lot. but I was introduced to the right people, and those people turned out to be angel investors and venture capitalists. And I ended up operating as a portfolio company executive for a number of years where I was running companies. I was basically the first credentialed, like MBA type who would come into these companies. They'd raised a bunch of money, they weren't really doing so well, and they'd hire me to try and figure out whether or not this is a company that could be turned around, or unfortunately, whether or not it's a company that needed to be shut down. So Honestly, it was a really privileged position to have because I got to see all of the ways that ways that companies messed up. And I thought to myself, man, there's just gotta be a better way to do this. Like I don't have oodles and oodles of money. So I want to figure out how to start my own companies. And I need to do so in a way where I'm not gonna lose my house and my marriage. Like that was sort of like the rule in my house is, you know, I could be an entrepreneur, but you know, we can't put the house on the line. So what I hit on was digital marketing. And I know that you have a digital marketing background. So this should this should resonate a little bit with you that with the rise of social media, you know, long time ago now, I realized that I could really rapidly test ideas. without necessarily putting up a bunch of capital and basically slap up a landing page, have a hypothesis, drive some traffic to it. If that hypothesis worked, then I'd pony up the dollars and invest in whatever the idea is. And I executed on that by buying US patents, so utility patents. And I would go out, I would find patents that would that that that people had built, like really smart engineers had built something, but they couldn't market their way out of a box. And I would test the idea. If the idea worked, then I would license it or buy it. And I got pretty lucky with some of those. we built several different patent portfolios for different products and then brought those to market initially through channels like Amazon. but also through direct advertising in Facebook and Google. And then we got very lucky and ended up selling through big box retail like Walmart and Costco and places like that. You ask, what does this have to do with AI? This is just digital marketing stuff, right? So the AI comes in where it became very, very difficult to be a digital marketer back in 2017. So you know, you weren't probably not really on the scene yet, but when our friends at Cupertino at Apple made tracking really hard, like pre-2017, like you could do basically anything, a little less so in Europe, but in the US, you could really have a very good idea of who was visiting your properties, how to retarget them, et cetera. Well, as attribution narrowed, we lost our visibility into advertising.

And we were spending a lot of money on advertising, like eight figures on eight on advertising. So out of necessity, I leaned into machine learning, which at the time nobody really called AI, by the way. But machine learning is the, you know, the basis for a lot of modern artificial intelligence. And I ended up developing attribution structures based on billions and billions and millions of points of data. That informed how I did my advertising spend. And then I exited those companies what feels like five internet minutes before Mr. Sam Altman dropped all of this on us without any instructions. And I was like, my gosh, like this is absolutely what I've got to do next. So Ascend was born in December of 2022, like literally like two days after OpenAI dropped.

Yeah, sure. I would say that that that there are at particularly in two thousand twenty six, there are a lot of consultants out there who can advise on strategy. These are very smart people. They have an understanding of how the technology works, where its strengths are and where its weaknesses are, and they can come in and they can advise. Okay, great. And then there are people on the other side of that coin who have an implementation bench. These are the people who are waiting for the information from the first side for somebody to have this really great idea of how they want to do marketing automation or whatever it is. And they say, Great, I can do that for you. And they bill a bunch of engineering hours and you get your toy or whatever it is. The third piece of the puzzle that is not particularly well covered by either of those. Is what I would call process engineering. So it's understanding what is the right type of technology that needs to be applied to the right type of business process in order to accomplish whatever the goals are. And we realized that you need to have all three of those to be really effective. And I and and it's also just much more interesting. Like I don't want to just do like, This long list of, hey, these are really cool ideas. Go do them. And I don't really want to receive bad ideas from other people and then be in a position where I'm trying to execute on sort of poorly formed ideas. So we we came up with this idea of the AI office. And you hear a lot about forward deployed engineers. this comes from Palantir. You've got Anthropic and OpenAI are now deploying these forward-deployed engineers. And they get into the organization, they start to understand the organization, and they solve the problem. Well, that's basically what we do, but we do it at a larger scale, as in our whole team can come in and operate. So it might be myself or my partner who are operating on that strategic layer. And then we embed ourselves and our people with the the functional leaders within the organization to understand what they need. And then we do the development as well. So a little bit different from being a consultant. at the end of the day, am I a consultant? Sure, in that I'm selling yeah, I'm selling a an output and then I'm executing on that output at the other end. But I think the model is a little differentiated.

So it's becoming increasingly clear. organizations that are s that don't have a strong technical infrastructure are a sweet spot for us. So we work with luxury hospitality, for example. So these are like boutique, really high-end type luxury properties. they don't really have a tech function. They don't have a CIO or a CTO. They might have like a web developer or they might have you know, an IT infrastructure type person, but there's no like technical person on the team that they can lean into. So that allows us to slot in and take over that whole capability for them and help them build that without necessarily having to invest into the into the tech infrastructure. It always gets a little bit hairy when it's when you're coming into a highly tech forward organization. Frequently they realize they don't have these specific skills. So they're finding somebody like me, but the CTO's office still has a lot of like agenda. They just don't have the ability to like like do it. And then you're trying to integrate with a lot of their existing practice and and approach. and it it just creates a lot of friction. so we're happy to do it. But and in fact, sometimes we'll we'll take on engagements where we're basically using our FDEs, our forward deployed engineers, to train the internal teams such that they can then execute on their own. And then I can kind of back off and be like, all right, okay, now you guys know what you're doing. We're here to support. When you come up with a big project, then let's talk.

one, you better walk the walk. So you need to be building yourself. if you're if you have no clients at all, like your first goal is to get a client, right? So the first automations that you're probably leaning into are web build development. They're probably SEO, search engine optimization, AEO, answer engine optimization, it might be advertising, etc. But You should not be doing any of that the traditional way. If you want to walk the walk as like somebody who is an implementer, then you better be trying to solve your problems, not through hiring an agency, not through hiring somebody on Upwork or some virtual assistant in the Philippines. Like try and automate it yourself. So that doesn't take very much. Okay. So we actually have a product that we call the AI Build Kit. And This is funny because it it's it's it's it's applicable to this. But the the idea is you go into a client who's like, we want to vibe code, we want to build stuff, and we're not developers, we're like marketers and we wanna do this. And you know, a a traditional CTO, their head like starts melting because like there's a lot of risk in this stuff, right? But what we teach them is the same thing that I would advise like a solo operator is you have to have the right sort of foundation. And yes, you can watch a bunch of YouTube, but the funniest part, what people like don't get about AI, is it solves its own problems. Like the first thing that you should build with AI is a cloud or an open AI project within the chatbot that understands who you are and building, and then you can bounce ideas off of it about what you're actually trying to solve. And then you start to build your capacity. And you know, a basic stack is going to be. Cloud Code or Codex. I I they're both up on my screen on right now, as a matter of fact, doing different things for me. You need to have a back end like Superbase. You need to have something like Vercel, or you could be using Cloudflare instead of Vercell and Superbase and all for the low, low price of about 20 bucks a month. Like this is not cost very much to get going. Just don't convince yourself that it's overly complicated. So use the tools that already exist, solve real problems for yourself. Start with something minor, like you know, spinning up a website that actually has like a Google Tag Manager implementation in it or something like that, that that like you it's something tangible that you can you can look at. And then think about the way that you spend your time. am I spending all of my time running around, you know? Like trying to get on podcasts. How can I automate getting on more podcasts? Or how can I automate putting more content out the other direction? find those things and just do them until you're good at them.

so it touches on the transcript. I would say that the thing that everybody is after, and listen, budding entrepreneurs or people who are out there, like there's a big opportunity here. I haven't figured out how to execute on it yet, but maybe somebody listening will call me. it has to do with what we call the enterprise knowledge graph or the enterprise context graph. And this is about pulling in all of the relevant information of what's going on in the organization. This starts with transcripts, and even all of these are challenges, by the way. Like you you you can pull in, you know, we have hundreds of calls that happen in my organization every week, and that's a lot of information. And even though we're using RAG to then chunk the information and make it more accessible, it's still like sort of expensive and pretty complicated to manage, and who has access to what and whatever. So all of the transcripts, all of the direct messages. All of the emails, all of the client documents, all being able to be put together into something that is that is usable as a core piece of information within the organization. And right now it's really hard to like figure out where all those data, all that data is sitting and being able to configure it and whatnot. but it is so cool once it comes together. and then tying it in via an MCP to Claude or to GPT, such that when I'm operating in my daily mono thread of, you know, I that's the way that I operate is I open Claude in the morning, I have a morning brief, it comes through, but now my Claude MCP has access to the transcript, to the Slack, the finance, all of the different pieces such that I can pull all of those together. And then as the owner or the CEO, I have a really high overview of all of these projects that I'm not touching. So how is this client progressing? So the project management tool is tied in. And I spend more of my time not like looking under rocks for things that are wrong in the organization, but those things come to me first thing in the morning such that the rest of my day is focused on solving the right problems as opposed to looking for problems.

Boy, it's amazing that we've been doing this for four four years with generative AI, or very close to four years now. And people still don't get that this isn't just like a tool, like, you know, Excel or whatever that they can just buy and like throw into their organization. Like it Does I do a lot of training still, a lot of like AI 101 training with executives and teams and and and things because you have to understand this technology as a technology more than other technologies. You do not have to understand how Excel works to use Excel. You just have to understand how numbers work. With AI, understanding its strengths and its limitations is a prerequisite. For being able to lever it appropriately in the organization. So having not just a bunch of tools that are out there, hey, look, we've got GPT teams or we've got anthropic teams and go for it, guys, and make stuff happen. Like that can sort of work, but people need a lot more help than that. Most people are not as technologically forward as us or your audience, and they need to be enculturated into the way they think about this. Okay, so that is that is definitely number one. then there is a there's a poor understanding about just how all of these different data sources work together. That's sort of the previous conversation that we just had. And there's there's sort of an assumption that this is just magic technology, which don't get me wrong, this is a magic technology, but like it's not perfect. Like it takes a lot of work, it takes a lot of iterations and it's a skill set. But The expectations are of basically instant gratification from things. And that gets built into what would ordinarily be a more disciplined like project plan. And then everybody's frustrated because it's like not working out of the box. And it's like, guys, come on, we got to iterate on this. This is going to evolve over time. I would say that the last, and this is sort of funny because this is yet another entrepreneurial opportunity. Is we spend a lot of unti time undoing the damage that others have caused by poor implementations when the technology was not as robust. And in fairness, I am sure there's somebody else out there who's undoing some of the damage that I caused, not in like a nefarious sort of way, but like it's just you would never build stuff the same way we did in 2023 today. So Now you're having to undo a lot of those processes and reintegrate them. And you have to do it because context isn't adequate the way you did it before. Token cost is probably really high the way you did it before. And that tape that like connected things together creates risk in the organization that isn't a great idea. So taking ripping off that tape and rebuilding it, we sort of jokingly call it AI plumbing. Like, you know, the guy with the wrench like over his shoulder. Like, we we gotta come in and we gotta fix the pipes because the pipes are all kinds of messed up. that's gonna be a thing because, you know, here you and I are like citizen developers and we're building all of these things, and inevitably, like something something's gonna go wrong. And it's always harder to fix the thing that's gone wrong than it is to build it right in the first place. So out comes the digital wrench. We gotta, you know, fix it.

Well, one, I would have bought some NVIDIA shares. that like even as close as I am to this, like I'm not a great like retail investor, but it was very obvious to me even back then, like the direction a lot of these valuations were gonna go. Anyway, but that's not here or there. the other thing is as an entrepreneur, you know, a lot of the opportunities i are in building. And it's not building a SAS because I'm I'm s I I've been around the block enough. That I wasn't really willing to like build some rapper. So I, you know, the term rapper, just rapping like, you know, people made millions and millions and millions of dollars doing like stupid rappers, but like I it it it just all felt so transitory to me that that wasn't really a thing. But acquisition, like if I could roll back the clock, I would probably just start snapping up like little marketing companies and things like that, such that I could be just making direct improvements to them. but as far as the send itself, it hasn't been easy. Like I think that's been a little bit of a surprise in that I know that we are very good at this. I know that we do a very good job of it of of working with the executives, but it is it n nobody really understood what the heck we were talking about. Even today, some t a lot of people don't understand what we're talking about. But 2023, 2024, my gosh. So hard to get that going. I d you know, I don't know if I could have done much differently that that would suit my own way of as a bootstrap type operator. you know, I did hire, I did some hiring in like a traditional sort of way that I should have eaten my own dog food as I was telling your listeners to do, and I should have tried to. build the solution instead of trying to hire the solution. Because what I realized was it was really hard to hire people who have the level of skill that they're going to need to build this. And I spent a lot of time and a lot of money on some very smart, very nice people that didn't work out at the end of the day because they didn't have the right builder mindset. And that time and money might have been better spent, either, you know, working internally to develop the automations we needed, or even, you know, contracting to somebody else, if I didn't have the resources to do it.

so the the site is I'm sure you'll put it in the show notes, but it's ascendlabs.ai. I also do run a weekly podcast that is called AI at work, and it's more nuts and bolts, hands on the keyboard type experimentation where I I like to think I'm fully transparent of you know the the sort of mistakes that I make. I spend every weekend like building. this weekend I had a lot of experimentation with GPT Astra and all of the amazing things that GPT Astra can do. and yeah, I I try and be as transparent as I can. I make expensive mistakes, I make stupid mistakes, and I share them all. Also LinkedIn. I'm easy to find on LinkedIn.

Great. Thanks for having me. It was super fun.