How CEOs Build with AI - Sean Devine of XBE
Joining us today is Founder, CEO, and AI early-adopter Sean Devine. Sean is the mastermind founder of XBE. XBE is a bootstrapped company providing cutting-edge software to horizontal construction companies (think roads, bridges, tunnels, and airports – big wide stuff).
The company is interesting, but today we focus on how Sean is learning, leading, and building his company differently in this wave of AI Innovations.
Sean is using AI to transform XBE – the product and the company. He also shows the AI tactics he applies as a CEO and individual contributor.
AI is set to be the next big thing, and Sean has been thinking about it for a long time. In this episode, you'll get a glimpse into the possibilities AI provides. So buckle up and get ready for a conversation that might change the way you think about the future of AI and company building.
Links to Platforms:
Here’s what I learned from the episode:
Sean’s story about founding XBE may hint at where business building is headed in the next few years and the leverage that AI can provide.
Sean is a self-taught programmer (in his 30s!)
Being a generalist founder with experience in multiple fields, later in his career helped him start a company without outside capital.
Sean uses ChatGPT as a constant collaborator, like a partner sitting next to him.
Working with AI is counterintuitive – Most people don’t ask it to generate a lot of ideas. The norm is rather to just ask for one answer, like you would a person. This is not the way to do it. You want to it to generate as much as possible, then you filter for the best.
Here are three main ways to achieve a goal with ChatGPT. 1: Be clear about your goal. 2: Provide it all the context your have. 3: Have GPT explain its thinking and break down the problem before it does more work.
#1 secret to problem solving: knowing what problem you are solving.
#2 secret: break the problem up into smaller bits.
With new AI tools, Sean thinks XBE can double the business with the same size team.
Sean predicts the AI wave will impact the top of organizations first and then ripple down.
AI changes hiring: Expertise doesn’t matter as much. Hire for problem-solving and mental stamina.
This episode is sponsored by Bread.
Bread is not your typical dev shop. They're like a technical cofounder team that you can add a whole product team as a pod. So, if you are non-technical founders or you want to spin up a new project, a new "swim lane" in your company quickly with very talented people, talk to Bread.
If you reach out to them, please tell them Eric sent you. It's madebybread.com. Check that page out. It's very cool. It's very well designed and will give you a sense of what they can do.
Learn more about Sean Devine:
Additional episodes if you enjoyed:
Episode Transcript:
Eric Jorgenson: Hello again and welcome. I'm Eric Jorgenson. I don't know much, but I have some very smart friends. And if you listen to this podcast, then no matter who, where, or when you are, you do too. Together, we explore how technology, investing, and entrepreneurship will create a brighter, more abundant future. Today, my guest is Sean Devine. Sean is the founder and CEO of XBE, who provides mission critical software to horizontal construction companies. I didn't know what that was either. Horizontal construction companies build things like roads, bridges, tunnels, airports, stuff that's really heavy on concrete and asphalt and is usually funded by the government, huge niche. Sean has fascinating stories about founding and building this company. But that's all for another day. Today, we focus almost entirely on how he is applying AI tools inside his company right now, today, as we speak, as we listen. And he's doing that on a few levels, as an individual contributor. He's the CEO, but he wears a lot of different hats. We talk about applying AI inside his product in AI features and how they use AI to build the product, and then how it's changing things in his hiring and his team structure. It's really a very thoughtful approach. And Sean’s been thinking about it about as long as any operator I know. Every time we've hung out recently, this topic has dominated our conversations. And since so many of us out there are hearing about or even playing around with some of this new AI technology, but kind of yet to make the leap to do the work to apply it to our problems in our companies today. It's increasingly clear that artificial intelligence will be a sea change that's on the order of mobile or maybe even computers themselves. And we're all working to kind of wrap our heads around it and figure out what to do. And I hope this conversation helps. This podcast is one of a few projects I work on. To read my book, get the newsletter, or invest alongside us in early stage tech companies, please visit ejorgenson.com. We invest in everything from b2b SaaS to nuclear startups, all looking for funding those 100x returns. The fund lets you invest your money right alongside ours into 15 to 20 exciting startups every year. We keep our fund minimum low so anyone can get exposure to high tech startups with minimal time and effort. I'm honored that many listeners, including today's guest, have already joined the fund as co-investors. Learn more at rolling.fun, which is linked in the show notes below. Accredited investors can invest with us through AngelList today. Our conversation starts shortly. Until then, here is this one episodes- this episode's one sponsor. And if you're pulling out your phone to skip this, that's a great opportunity to leave a quick review for the show in your podcast player; it really helps the show out. Thank you very much.
Our sponsor today is Bread, a new sponsor, This is a company founded by very good friends of mine. And you can think of them as your technical co-founder to launching your company. They will help you design roadmaps for your product, pick the right tech stack, they’ll help you build MVPs, they’ll help you do product research, wireframes, help you even recruit and onboard technical team members, even co-founders. So if you are a non-technical founder, and you'd like help getting your software product off the ground, talk to them. This is not your typical dev shop. These are previous experienced founders. They've been building companies together for a long time. They've been through the early stages before. They know what it takes to get it done and get it done well. And not only that, I know and trust that they have founders’ best interests at heart. I think a lot of agencies know that they can get the cash and run. And these guys really want to help build successful companies. They have a framework and a roadmap for creating long term success for your product and hiring, which I think is such a critical way to think about this. They're not trying to work with you forever. They're not trying to make you dependent on them. They just want to help you get off the ground and get flying on your own. I have a good friend, a repeat founder with a very successful exit, who just signed a deal with them to build the first version of their product for the next company. And so if you have a startup or company that needs a very talented team of technical folks spun up today, please check out madebybread.com. If you reach out to them, please let them know that I, Eric, sent you. If you have any questions about them, email me, I'll happily answer them or personally introduce you. Now with both ears and everything in between, please enjoy this conversation arriving in three, two, one.
I can't wait to do this because I am surrounded- I'm digitally surrounded by Twitter threads of people who just like have been punishing GPT and coming up with like incredible, quote unquote, use cases and building giant threads about how you could use it. And I've talked to very few people who are actually digging in in real operating businesses and applying AI today and changing how they think about their company like immediately. And Sean and I have already had a few conversations. So I'm incredibly excited to like bring this to the world and invite some other people to sort of catch up with what you've been doing. I think there's a lot of people operating businesses just overwhelmed with like the new possibilities and not sure where to start. So I want to have you back to tell your whole story. But today, let's just deep dive into how you're thinking about all these new AI tools, what's going on, what you're changing. And if you don't mind, can we like set the context for the company that you're managing and size, scale, industry, that sort of thing?
Sean Devine: Sure. So well, thanks for having me on. This topic has really lit a fire for me, both not just in what I talk about, but what I do every day, how I work with people, the kind of work I do, etc. But a little bit about me, so my name is Sean Devine, CEO of XBE. I founded it in the beginning of 2016. XBE is a software company that provides horizontal construction companies with operations management solutions. We have one giant platform that goes all the way from scheduling and planning to dispatch and monitoring and analysis and continuous improvement in safety and administration. So kind of the entire physical operation of industrial contractors that build roads and manufacture ready mix concrete and mine aggregate, all the heavy construction that Biden's infrastructure bill is meant to support.
Eric Jorgenson: So business is good for your customers right now.
Sean Devine: Business is good. Now, the 2022 inflation issues were not great. Because the huge increase in funding was absorbed by a similar amount of actual volume because the cost of natural gas went up two and a half times, the cost of liquid asphalt cement went up two times, and so on and so forth, diesel went up twice. So, that was a little bit of a dent in things. But most of those have settled down now. And at least for the next three or four years, things should be pretty good. So anyways, that's what we do. The software platform is fantastic. It's super wide, super deep, has been seven years non stop of one step at a time building it out. We work with the industry leaders across the country. They run their whole operation on us. So we're pretty critical to getting the job done every day. And we take that pretty seriously. So that's a little bit about me.
Eric Jorgenson: The team size roughly?
Sean Devine: 21 people. We are half in the US, half an India. It's always been that way since day one. So I was number one. Milen Alvarez in India was number two. And it's just been eeny meeny miny moe since then. Yeah, interestingly, I think we've doubled in the last 18 months or so, and the team size has stayed about the same. So, as a believer in leverage also, it's fun to be on the podcast.
Eric Jorgenson: Yeah, you love to see it. And this is a bootstrapped company also?
Sean Devine: That's right.
Eric Jorgenson: Or minimally funded, not the typical kind of like venture backed tons of capital situation.
Sean Devine: No, we had- I don't mind saying what the number is. So the amount of capital ever put into the company was $150,000.
Eric Jorgenson: Wow. This is an incredible example of like my thesis that like software investing and venture investing are slowly diverging, like the capital necessary, including now especially with AI, to like build a software company and grow it is just so much different than like the traditional multi round, huge venture scale stuff. And I think, I mean, the main topic is like how we think about getting maximum leverage out of the team size that we have now and continuing to grow that. So, I forget the exact number you said. You more than doubled and the team size stayed the same over the last 18 months. Amazing.
Sean Devine: I think that my story about starting XBE and the way that we've gotten so big and successful with a relatively small team, I think hints at what's going to happen over the next year or two. So, I can program also. I didn't tell much about my background, but I'm sort of a hybrid business guy that can program, also an entrepreneur. And I think a lot of my friends over the years have really questioned my commitment to being a good programmer. Because I didn't start programming until I was 32, I think, and was relatively successful career wise at that point. And my reason for learning how to program- You don't mind this digression, do you? Because I think it's relevant.
Eric Jorgenson: No, please. I'm very curious, actually.
Sean Devine: So my reason to learn how to program was that I had some very interesting jobs in my late 20s and into my early 30s, but was sort of forecasting the rest of my career and sort of seeing that my intensity, like the way that I worked, just didn't work great with everyone. And I didn't know that I was really interested or capable of changing that. And I said, what are the ways I could make maximum money? I need to find leverage. And either I'm going to find it through having leverage over people, which is what I had then – I was pretty high up in giant fortune 500 companies. But that caused- that required enough sort of political agility and comfort, and I saw that even if I was okay at that some days, it really wound or really sort of ground me down. And so, leverage with people seemed like a shaky proposition. I wasn't wealthy, so I didn't have leverage from capital. And I could get leverage from other people's capital potentially, but ideally, I wanted to figure out how to do things myself. And I thought, well, I mean, getting leverage through machines seems like, theoretically, the best strategy for me. And the only problem is I don't know how to program. This is me at 30. I said, well, I mean, I was smart enough to program. I was one of the best people at Excel that I knew. And I'm like, well, if I'm great at Excel, I probably could be a decent programmer too. I’m good at math, etc. And I said, well, ideally, I would have started at 14, but second best time is now, and I'm just going to learn how to program so that I can create leverage myself. And so that's sort of why I got into it. It was a little bit clinical, even at the time, and it certainly sounds pretty clinical in hindsight.
Eric Jorgenson: What was the path that you took to actually learn? Was it self taught? Did you bootcamp? Did you go back to school?
Sean Devine: Self taught. So I had a couple of problems that I tried to solve. In particular, at the time, I tried to solve the problem of sort of scraping a couple of websites for bill of ladings and bill of lading related documents for LTL carriers to build like a lead generation database. And it worked pretty well. And I mean, the code, I'm sure, was just God awful. But I had a real clear problem to solve that I knew money was attached to and said, well, if I can figure out how to do this, one, I'll come out the other side with more skill, which has leverage, and more money, which has leverage. And so, that's kind of how it got going. And then, I'd enter hackathons on the side, like on weekends. I did a number of those. I ended up in a sort of strange series of events, being the host of the Ruby on Rails podcast for a while. And it was before I was that good of a programmer, but I was very much in the center of that community, and interviewing all of the primary people in that community. And eventually, over that time, I became a decent programmer, but I didn’t really start that way.
Eric Jorgenson: It’s a great way to learn.
Sean Devine: Yeah, right. Well, I just reached out, I'm sure you can relate, I reached out to every person I was impressed with in the community and invited them on. And over the course of a couple years doing it, all of a sudden, I was one of those people. And so anyways, it’s a long way to say that I put all this work mid career into learning how to program. And I think at the beginning, I had no idea what kind of commitment that was going to be. And I got some skill and then my company Partage, the company before XBE, it didn't start off as a technology business entirely. It was a partial truckload brokerage that I started with a good friend named John Labrie. But pretty early on, we saw that the key to differentiating, mostly at least, was pricing sophistication. And there weren't off the shelf tools available to do exactly what we wanted. And so I said, well, I've kind of been hacking on some programming stuff. Let's see if I can put something together. And again, it was pretty amateur hour, I think, in the execution, but got from here to there, generated some money. And then we realized, well, the transportation management system that we're using, it's not really sufficient to have customers book shipments themselves with our new automated pricing system, how hard could it be to build something? Well, it turns out the answer is pretty hard. But I didn't entirely know that at the time or accept it. And so I just started going. And anyways, it’s a long way to say that eventually, after probably five years or so, I could program pretty well. And that brings me to XBE and sort of the bridge to AI, which is, at the time, again, I think a lot of friends, a lot of family members are like, you're nuts. You're spending all of your time learning how to do something, and you don't have to do this, your career is fine. And I said, well, here's the thing, I'm going to start a new business, what's now XBE, and I've already raised my Series A. What do you mean? I had no idea you were even raising money. Why didn't you come to me? And I said, no, no, no, I don't need it. I don't need the money. Because I already have the thing that I would hire. In other words, you need a product manager. I already am good at that. You need to someone to do the architecture. Well, I can do that. You need someone to rig up the initial front end. Me too. You need a CFO. Well, I went to business school. You need someone to do strategy. Well, I used to do that. In other words, I had the skills that I needed to get things off the ground. And that's not to say that I'm the world's best at any of those. But I was good enough at all of them to get to get going. And instead of needing 2 million of seed money or 5 million in Series A, I said, I think through the expertise that's been gained, I can forego that, spend a little bit more time, but I was fine with that because I think product market fit usually takes a minute, especially in these business to business sort of complicated industries. And so if you try to go too fast, you're going to break the feedback loop. You won't know if you have product market fit because you're sort of like powering through it with money. And so, I was like, I'm okay with it going a little slower. And I have the skills therefore that I need to get things off the ground. So that's a long way to say-
Eric Jorgenson: And you're mid 30s by this point, by the time you had acquired all those skills.
Sean Devine: Probably late 30s at that point. Yeah, or 37 maybe.
Eric Jorgenson: I think it's an underrated version of the story. I think the headlines are often like 24 year old raises $2 million round. But to your point, it's very interesting to have somebody who's like very broadly capable, sees the synthesis across all these different skills, mid career, like height of their powers starting something and with a perspective that I think is totally right. Like, there's an extent to which money can't buy product market fit. Money can buy time, and time buys feedback loops. But you can't necessarily speed up the feedback loops, especially with money, and you need a certain number of feedback loops. That's like the salient metric to actually like hit that breakout thing. And so more time, lower burn is better than a bigger pile of money that's disintegrating faster.
Sean Devine: Yeah, that's right. So I mean, if it was the case that you could raise money to go at a medium pace longer, I think it’d work pretty well. But the problem is you raise money and the pace goes up. And also, you're going to close a lot of deals that you didn't have any business closing. Because you didn't actually have product market fit, you just were yelling to enough people that eventually some people say yes, not to mention, you don't need the money because you have money. And so you're both going to pitch to people that may or may not be the perfect fit, you're going to price it in a way that is buying the business to some degree, and then you're going to mistake that dynamic for product market fit. And so, I've always been really suspicious of pouring sort of fuel on the fire early, kind of for the same reason I'm suspicious about taking too much medicine when I feel sick in that I like to feel sick. I like to know that I'm sick when I'm sick. And I feel like when you're an early- when you're early on in a business, you're sick. You don't have product market fit. And I like to feel the pain of that.
Eric Jorgenson: That pain is information.
Sean Devine: That's what I'm saying. It's really helpful to know where you don't have it. And just like if I took six Advil every day, every morning, I really wouldn't know how my body's doing. And I kind of feel that way about funding. Like I just, it causes a break between your experience and the sort of reality of the fit between you and the market. And so even before- maybe it was that reason that helped propel me to sort of learn more things because it at least meant that I could do what I wanted to do which is not raise money. In other words, I'd have to.
Eric Jorgenson: Yeah. Well, and you seem compelled to have a front row seat to the feedback loop. And having that broad set of skills like really helps you get there in every part of your business.
Sean Devine: Yeah, I think that's right. I think that's right. Well, the reason I brought up this story is that this tells us what's going to happen now, I think, in that now, many, many more people than was true seven years ago when I started XBE have what I had then. So like have the ability to program reasonably well, have the ability to research a market, understand the industrial dynamics of horizontal construction, can get into the mechanics of some financial management bit of the business, can think through their pricing strategy using the models that are out there, can do market research effectively. Like all these tools that used to take money directly or expertise which money could buy, well, the expertise is now abundant. And so what do you need the money for? It's just going to make you sick. Like, the money hurts you. It's just that it's better than the alternative. But if you don't need the thing that money gets you because it's free, the expertise is free, then why in the world wouldn’t one take it? So I am fascinated to see the next phase of how people start companies because I kind of think you'll see it, maybe this is just me projecting and imagining myself as the hero of the story and all the rest, but I think you'll see what I did, except now.
Eric Jorgenson: I think it's extremely true or could be extremely true in software in particular, like these businesses with like low capital requirements to get started, high pace of iteration. So let's get into the AI piece because I think a lot is about to change in software in particular. And I want to start with sort of your seat as like the current CEO and operator of a good size software business. Like there's bigger ones, there's smaller ones, but like you're a great sort of medium scale, like successful, profitable business with two dozen employees. And you're actively applying all these tools today. So, let's see, where did you start your sort of like discovery and usage journey with some of these AI tools?
Sean Devine: Yeah, so I think it's good to talk about this in two ways. First, me as an individual contributor, and then as it relates to our product, and then as it relates to our organization. So, where I started was me, personally, in that I communicate constantly. So I communicate constantly internally and externally and a lot that's published to our customers, to our users, etc. And I think maybe the first place where I started to see traction was just using GPT 3 as an aid to content production, which is a very sort of practical, boring thing. But I was spending a lot of time because, I mean, communication is leverage. That's a way to, instead of having a conversation one on one, I'd rather have a conversation through content with thousands of people every time I write something.
Eric Jorgenson: And when you say using, like just I'm going to dig sort of like uncomfortably detailed on some of these, like what does that mean? Is that having them draft? Is that having it expand from an outline? Is that having like Google Docs open in one window and GPT open in the next one and like going back and forth? Like, talk to me in detail.
Sean Devine: So it took me a little bit to maybe get to this place, but I'll describe how I use it now. And just know that it sort of evolved over the last six months. So I use GPT 3 now 4 Chat GPT as a collaborator, mostly. So I start at the very beginning with what my goal is, with the most upstream place possible. I’ll give an example from yesterday, I mean, even though- but all this was happening back in November or December when Chat GPT came out. So yesterday, I wanted to put out a newsletter. I had an idea, which just came from a conversation with a customer, about how if they provided comments on good catch safety incidents, like right in the moment, that they would help reinforce the safety culture of their organizations, which would lead to better safety outcomes and improve culture. So I had this- I was in a conversation with a woman named Amanda Moore. I mentioned that in conversation. I said, ah, that's a good topic for others. So, I made a note to myself, like just those words, safety incident comments plus culture. Then later that night, and this is an interesting part of the story, I think, it was like 11:15 at night. And I'm kind of tired. I had the Warriors game on. I was just in that kind of sleepy zone where I probably didn't have it in me to write a good newsletter. But I have this collaborator that's always ready and interested in helping out. And so I just started and said, I had this conversation – I should pull up the transcript, but it went something like this: I had this conversation with a customer named Amanda. In that conversation, and I describe the context I've just described to you, and I said, I think that would be a good newsletter to send out to everyone. So here's my idea. I'd like to build the newsletter and have one section, which is the sort of big idea and then a step by step how to about what software features to use. And then I’d like to cite research that shows that this is good out in the world. And I don't know what that research is, but I'd like to figure out that. So I said, okay, that's about the outline that I'd like. Can you flesh that into an outline? And am I missing anything? So it replies, and it sort of builds out an outline. Now I can kind of see what I'm talking about. I said, well, okay, I think we need a little bit more upfront about an example, I want to put an image down below so they can see. And then on the research, I think I want research now that you've sketched it, one piece about good catches and near misses, and then another about safety culture more generally. So let's take an aside and tell me all the research you're familiar with in those areas. Give me 10 examples. And so, it wrote 10 papers in those areas. And I said, oh, those two seem interesting. Give me the synopsis of those two. And it did. I said, okay, good. So we're going to use those two papers. And we're going to insert those as our example. Now, back to the outline. So the three features I want to use or highlight are inline commenting on the my incidence page, safety incidents subscriptions, and whatever else. And so, I said, so let's bullet those out. And then I said, okay, for each of those, I'm going to paste in the release note for each of those features. And then you take the release note, and then you write the how to, but it can't be more than two sentences. And this is all the basketball game’s going on, it's like I'm chatting with a friend. And if for those that know me, I'm chatty. So I will chat, chat, chat about the thing forever. And I'm coming up with ideas and I'm chatting. I'm saying no, that's not good. That's not good. That's not good. And it turns out all these attributes that annoy people aren't annoying anymore. Like it's annoying to work with someone that's that opinionated and chatty and has ideas and feedback.
Eric Jorgenson: Rejects 90% of the work that was just created.
Sean Devine: Yes, that is- so, it's really interesting you say that. So one of the observations I've had about using these tools so far is that most people that I work with don't tell it to generate lots of ideas. And I find it fascinating. So the norm is to say, give me the answer. That's not the right way to do it. What you want to do is like the example I gave before, like tell me all the research you can fit into 2000 tokens, basically. I can read at, especially that kind of stuff, you can read at say 400 words a minute. I mean, you can read that fast. And so, generate as much as you can. And then to your point, I'm reading it basically as fast as it's writing it. I find the two things I like. I say, chuck the rest, let's go with this. That I think is so antisocial in terms of a workstyle that, unless you're disagreeable, like I am, and like it doesn't matter, you just end up working like that anyways, I think you have to learn how to be disagreeable. And it's been a pretty fascinating experience because it's sort of been very freeing to me because that's my norm is to is to kind of reject most of the ideas and want to generate more and then just go with the good one we come up with. And anyways, so it's a long answer, though, that explains on that one newsletter kind of what the process was like. Start with the big goal, give it feedback, sort of rabbit hole down every little part, zoom back out and take those parts and say, okay, now that, integrate other content that you have to provide context. I mean, you can provide thousands of words of context per question.
Eric Jorgenson: I think the new GPT 4, I think the input limit is like 25,000 words, which is like half of a book. That's a big, that's a huge input.
Sean Devine: Do you want me to be pedantic in this interview?
Eric Jorgenson: Please. Especially if I'm wrong.
Sean Devine: Yeah, so the current GPT 4 is 8000 tokens, or 8,288 or something like that, which is like 6100 words maybe depending on details. But it will be that high. So, what they announced is that it will go to 32,000 tokens. And so, four times the current limit. The current GPT 4 you can get if you pay for Pro is 8000, 8000ish, and that's two times what GPT 3.5 was. And I think it's actually very important to sort of get to know this because the context is unbelievably powerful. Like, you can just shove everything you know into that context. And then it will combine all the pre trained information it has from the knowledge of the world, plus all your specific content, and holy cow, it is unbelievably smart.
Eric Jorgenson: And that's about, I mean, 8000 words, just for reference, is about the transcript of this conversation, like we go an hour and a half, that's about what the dialogue transcript would be. That’s a lot of context, a lot of context. You could put probably your whole company's FAQ, internal or external was probably in the tens of thousands. A subcategory might be 10,000.
Sean Devine: So it's interesting you mention that. So we built out this feature called Hey, Kayla. And well, this will get to sort of the second category of thing I was talking about, which is in addition to this huge list of ways that I've personally leveraged GPT 3 and now GPT 4.
Eric Jorgenson: Wait, let's finish the newsletter story. So, I want to just like end cap that. How did you finish the newsletter that night during the Warriors game? How did that like come together? What was the total sort of like time investment? What did you end up with?
Sean Devine: Yeah, right. So yes, I did finish it and publish it, like completely to hit send or schedule at the very least. I think the total time investment was 45 minutes or so, end to end, from starting to hit publish. And the reason I remember that, I remember the game was still going. It was about 11:20 when I started and I remember it was just past midnight when I finished because I had like sworn to myself I wasn't going to work late that night.
Eric Jorgenson: Those are like half ass minutes. I think you didn't quite go that far. But like your 11:30 to midnight minutes are not your best minutes, not fully engaged minutes.
Sean Devine: No, no, no doubt, they were not. But the final product was, I should look it up, it was probably about 550 words, 600 words. So not that long. It was the format I said. It did cite two papers, it excerpted them, it gave the step by step one, two, three on how to do it, it had a nice preamble, it gave one example. The writing was consistent with what we had done before. In fact, that's a good aside. One thing I've learned is I give it examples, back to the spending of context where you've got this context of budget to spend. And so once I got near the end, I say, okay, now proofread that for any grammar issues or spelling. But I'm also going to give you three newsletters that are already published, that are in our voice, and are good. And just make sure it sounds like that. And very good at this. And so it reads those, it adjusts it, didn't make huge changes, but it makes changes, and that's not a huge deal to me because I wrote all those, so I'm good at writing my own voice. Now, it's not good in writing my voice. So you just give it examples. Of course, this means that other people can easily write in your voice too, which is amazing. It means our whole team can produce content that all kinds of sounds the same, which is fabulous.
Eric Jorgenson: Awesome. Okay, so that's a great example of sort of the individual contributor communication scale and sort of leverage that you can get. I think you were starting to head down the product example.
Sean Devine: Yeah, well, if you don't mind, since you put me back on track, I have one more individual contributor example that I think is more advanced that I think is in interesting, and it ties back to the like how will people found companies in the near future. So two weeks ago, I was scheduled to go to Con Expo the big, every three year construction convention in Las Vegas. It's exactly what you think it is. It's the $10 million setups from John Deere where you get to go and play with the excavators and all that. So exactly what you imagine, that's what it is, tens of thousands of people. So I was going there. But I was at my parents’ house in Albany, New York, visiting them and my sister who is pregnant with her first baby. And it snowed as its want to do in New York. And so it snowed a lot. And my scheduled flight was canceled along with all of the other flights that day. And I was rebooked like 40 hours later. Well, I couldn't go to Vegas then because that was sort of beyond the window that I was supposed to be there for some meetings and it meant that my trip would be too short. And it's difficult to get from Albany to there, so I just decided not to do it. And so I was faced with I had two days that I had scheduled that were now completely free. Now one of them was going to take a bunch of travel and other stuff. So I really had one day, which was Friday, that was totally free. And so, I said this is a very interesting opportunity. I am going to keep it blocked. And I'm going to test myself to see just what am I capable of doing in a day right now using GPT 4 as my collaborator. And so there was a problem that I had been wanting to work on, which was, I mean, the details are- we don't need to get into too much. But the basic idea is we have a trucking lineup in our software, which is imagine the 120 individual trucking shifts that need to happen in a given day. And they're all around the city. And they need different types of equipment. And they're doing different types of work. Like some are on an asphalt job, some are moving rock, etc. And let's say you work with 50 different trucking companies, some of them are big, some of them are tiny, they are in different places, they have different equipment. And ideally, you would solve that optimization problem to say, well, I can't violate any of my constraints, I can't assign anyone more than the drivers they have, I can't assign them fewer than the amount that they're kind of expecting or else I'll destabilize them, so on and so forth. I can't put them on equipment or jobs that require equipment that they don't have, etc. And what I want to do is I want to minimize the mobilization distance that everyone's got to drive.
Eric Jorgenson: So very complicated traveling salesman problem.
Sean Devine: That's right. I'm going to keep people- I don’t know if it is that complicated, but it's like not not complicated. So I want to keep people kind of close to home. And anyways, this is what these dispatchers do. And for a while, I had it on my list to say like, okay, we should write a solver that will solve this problem and propose solutions that respect all of the constraints that one has practically. And so, okay, Thursday, I decided while I'm traveling, I'm going to do research about my approach on this. So I'm not going to really do the work, but I'm going to read about some things, I'll jot down some ideas, I'll come up with some game plans, etc., like while I'm on airplanes. But my goal was on Friday, I was going to attempt to implement the entire solver. Now, I am not an operations research PhD. I've worked in a couple of optimization software companies, so I kind of know my way around a little bit. But I don't- I haven't implemented a lot of models and certainly not that- none that are like robust and commercial. But I thought, you know what, I know that topic well enough, and I have a very clear understanding of the goal. And I can program generally. So how about I just worked with GPT 4 like it's a PhD that I used to work with, like it's my PhD for the day. So I started at six in the morning, because I was jazzed about doing this, I started at six in the morning, I said, I'm going to go as hard as I can as long as I can and see if I can actually get this done today. And it took me until like 11 or so, but in one day, I was able to build, and I built it as a separate library, built a library that could solve the exact, not a toy problem, like the exact problem that we're solving with sample data that I generated from real scenarios that could handle all the different types of constraints, that could solve for the objective function that we wanted, that was fast enough to deploy in production in one day. And I went back and I looked at my transcript of the questions that I asked. And I mean, I asked, I don’t remember the exact number, I think I asked 180 questions during the day. I mean, an enormous number of questions because I didn't know how to technically implement any of it. Like, I mean, I started, and I from the very beginning said, this is what we're going to do today. Now, let's design the linear program. Let's get the variables right. Let's decide what solver to use. Let's draft the program. Let's consider alternatives. Let's- I mean, the whole- figure out how to host it. I mean, there was the issue of I wasn't going to run it inside of our app or use some web service, I had to decide that. It required some system libraries that didn't exist on our production stack. I had to figure out how to get those installed. Like all these things that any one of them would have knocked me off, or knocked me off in terms of it would have taken long enough that I just didn't have the time.
Eric Jorgenson: Become a whole project unto itself.
Sean Devine: Except none of- there wasn't a single moment the whole time where I had any fear of not knowing. Because it doesn't matter if I don't know. It doesn't matter. Like it doesn't matter. The expertise was completely irrelevant. And so I think that's a pretty good example.
Eric Jorgenson: Did it ever- it's an incredible example. It shows another sort of interesting counterintuitive thing, which is like your first one was, most people try to get it to get the answer right, instead the correct play is like generate a ton of options, select the ones you like. In this one, I don't think people are used to asking 200 questions in a day. Like that is a counterintuitive way to approach things of just like firing off uncertainties constantly. But I'm curious, in all those answers, were you ever worried that it would lead you astray? Like you had enough domain knowledge to kind of know, to be able to vet what was coming back. But the odds that it is correct on 180 questions in a row, like it doesn't- Open AI doesn't at least claim that it's that correct. They kind of caveat a lot of stuff and say it's prone to be incorrect about stuff or mislead you or it's possible to generate false information. Do you run into that at all? Or do you think that's kind of just a hand waving caveat?
Sean Devine: No, no, no, I don't think it's hand waving at all. I think it's a real issue. I mean, I think GPT 4 is spectacularly more capable than GPT 3 is in my own sort of personal use. Its ability to reason about things is, so to speak, way better. But I'd say a couple things. So one-
Eric Jorgenson: It is capable of passing the MCAT, the GMAT, the LSAT, like all of those. You can't pass those tests without the ability to construct logical things that are-
Sean Devine: Yeah. And I'd say it's very good at that work. So a couple of things. So one is there's a lot of leverage in the questions themselves. So for example, I've learned that, as I said before, you need to explain your goal and everything you know to it. Every time I see someone post some sort of gotcha example, I mean, if I could mute gotcha examples in my life, I would. Because I don't understand. That's like gotcha people in real life. Like, no, tell the other person all the things you know, and then they'll be working with the same knowledge you have. And then there's no gotchas. I mean, you can get gotchaed together, but you're not going to gotcha them.
Eric Jorgenson: So when you say a question, like what is- you gave examples before but like when you say a question, are you typing hundreds of characters of context into these like one individual query?
Sean Devine: I think my queries tend to be longer I think then others would. I tend to copy and paste a lot of stuff, like other code, like I said before, release notes, like answers from before. I mean, I have a lot of content around so I feed back stuff. And sometimes, having one ever lasting conversation, I found that's not really the right way to go. Because it can kind of- it'll fixate on its previous answers from one thread. And so sometimes you kind of want to start a new thread so it forgets. Like its lack of a memory is a feature sometimes. Like people would be well served to forget sometimes in conversation, like stop thinking about the thing that we were just talking about; we're going to start anew. And so one point is, be clear about your goal. Point two is provide it the context that you have, everything you- Point three is have it explain its thinking and have it break down the problem before it does any more work just like a person. So, what's the secret to- secret number one problem solving is knowing what problem you're solving. Secret number two to problem solving is make the problem smaller. It's just the same with this. So first, tell it what problem you're solving. Second, make the problems littler. So for example, you don't say, hey, write me a solver that does the following. No. First, it's let's think through the objective function and constraints. And let's iterate on those and make sure we get them right. Now, once we've got those right, I'm going to scrap that conversation, say here's our objective and constraints. Now, let's formulate that into a linear program or mix integer program or whatever it is. Okay, so now it does that. You go, okay, now we've got that. Now, let's select our solver. Now, in fact, in this whole thing, right at the beginning, you can say, outline a project plan for us that are all the steps we need to go through. And what I've found, back to the principle number one, which is tell it what you know, I know a lot about those things. So I just quickly jot down, in my experience, the following steps are required to do this, op op, op, op, op, op, op, op, op, op op. I may have forgotten something, read that, insert anything you think is missing, we'll talk about that, and now you've got a project plan, and then you go to step one, and then you go down the rabbit hole, and then you go back out the rabbit hole, and you have the answer to step one, and the answer to step one feeds step two, rinse, repeat as you go down and up. So you can imagine a lot of questions along the way in that process.
Eric Jorgenson: I mean, this is like the collaboration that you imagine Iron Man has with Jarvis. Like, oh, just throw that model together. Oh, compute that. What am I missing? Like, it is incredibly fluid sort of high respect give and take process.
Sean Devine: And that's a- zero that I've said is speculative. This is just my lived experience of it. Like this exists right now. And GPT 4 is good enough that my take is if it never got better, it's good enough like to do- to fundamentally transform- Yeah, I mean, it's already great.
Eric Jorgenson: So that solver is probably a good bridge into some of the product changes, which is maybe chapter two. So I think that's a really good sort of progression. So, you're running a software business, you spent seven years building a product that's incredibly sort of perfect for your customers. It continues to get developed. You've got a team of developers still on your team working hard. How is your product evolving both as a product itself and for your customers?
Sean Devine: Yeah, so I made a list in preparation. I did a little research before we started talking about all of the, in the last four months, what are the AI infused features that we've shipped, just to be sort of literal about this answer. So here are some. I mean, are the specifics helpful?
Eric Jorgenson: Yeah, I love specifics. Let's go.
Sean Devine: Here's some more. Okay, incidents. So there are incidents that happen that people track in the software, like the paver broke down, there's a fire on the road, that kind of stuff. So people on teams want to see the headlines of the incidents sort of in notifications and other places, like it's good to know what's going on. The problem is that these incidents are created in the field by people with cell phones- there are many reasons why the content is going to be shaky. They're trying to put a fire out, things are hairy, their fingers are too big for their phones, it's cold out. You know what I mean? There’s a lot going on. So it's kind of a stressful- it's underrated how stressful this is. It upsets the people involved that they all sound lousy in this communication. Because you read it and like the spelling is bad and the punctuation is not there, and it's kind of stilted. It's just not great. I was like, well, guess what, we'll just rewrite what happened into a sort of declarative single sentence that says, this is what happened on the incident. And we'll call that the headline. And then instead of leading with the description that the person put in, which again doesn't make them look that good most of the time, we'll make them look good by rewriting the incident headline with effective English. So we shipped that. Is that going to change the world? No. But sure appreciated by everyone. Okay, we added safety risks to jobs. So well, we'd already done that. But we used past incidents and general knowledge of the types of risks associated with different types of jobs to suggest what safety risks may be inherent to a given type of what we call job production plan, which is like a crew day. So sometimes overhead line danger is a thing, sometimes like night is a whole new set of problems. Cold is a whole new set of problems. And so we built something that suggested the likely safety risks given the full context of the job; that was new. We generated, we added something that created what are called toolbox talks, which are the communication that the foreman has with the crew about the plan for the day and about the safety risks in the plan to mitigate them. So there are two problems that are interesting there. One, most these guys are not Shakespeare, nor actors. So they're writing and they're reading is, like they would admit is not their strong suit. So there's a reason why they are foreman on a paving crew; that's not what they want to do is write a document and speak to a group. And so it's sort of stressful to both write the content and deliver the content. And so we just generate now the communication, and that communication’s available to everyone else. Another thing we do, everywhere where we generate content, we now make it audio also. Like everywhere audio. Because two reasons. One, a lot of our users don't like to read too much. And they don't read too much, to be direct about it. And so, listening to things is easier. They're in trucks going to jobs, it's sort of easier. Also, the people that would be communicating to them don't really like to read either, like read aloud. Like, they don't like to perform a thing. Plus, it's a little bit difficult at job sites anyhow. And people show up at different times, and it's loud, all these things. And so the audio is nice because it both shifts the time and the place to be more convenient for everyone. So again, generating the sort of like messages plus converting them into audio.
Eric Jorgenson: And are these- sorry to interrupt. Are these all sort of with the same like Open AI API that you're doing? Are these all different tools? Are you-
Sean Devine: Yeah, so the audio we use Google for now. All the rest of them are Open AI that I mentioned. Some of them existed before the current, some of them were back with the original GPT 3 API. We've migrated most of them to the Chat GPT version, well, and now we're on the GPT 4 beta. So now, most of them are on GPT 4, not all of them, but most of them.
Eric Jorgenson: And are the engineers like super excited to work on all this stuff and are like, oh my God, we can solve problems we've never been able to solve before?
Sean Devine: It's a good question. I think it depends. I wouldn't say on average that's true. I think it's some people are really into it. Others are, I'd say, they don't really see it any differently than any other feature. And I'd say that everyone to some degree has some fear of it all. Because my own estimation of my personal productivity is that it's something like 50% higher than it was three months ago, ish, at least that, now maybe a little higher. I think that 50 to 100 is probably what we'll see inside of our engineering team within two months, one month. I mean, we're right there now.
Eric Jorgenson: Because of stuff like copilot?
Sean Devine: Yeah, because of, well, a few things. Because of copilot for sure. So copilot is good for 25%. I think that, as it stands now, chat GPT and other sort of like more zoomed out tools, they're good for another 25, they're good at refactoring code, they're good at answering questions about like how do I X? How do I Y? What's the right algorithm for Z?
Eric Jorgenson: Please tell the bug fix story that you told me the other night. While we're in product land, this is a crazy one.
Sean Devine: Yes, I mean, it's amazing. So two things. So first, we had this sort of production bug, like red siren bug, that took down a bunch of buttons, basically, in our app. And all of a sudden, we started to see in our Slack support channel like customers say, hey, this button’s not working, what's going on. And then someone else said, this other thing’s not working. And once you hear like three of them, and they're all involving what I knew to be a type of button, I was like seems like there was something that broke a type of button component. But I hadn't looked at any of this code at all. And we were kind of in the transition. So most of the India team was kind of settled down for the night. It was early in the morning. So it was like a little bit like a what are we going to do kind of moment. And so I said, okay, I'll look in and see if I can figure out what's going on. And I looked at the recent poll requests that were merged, found one that seemed like it probably was the culprit, except it was big. It was big because it was sort of modernizing, so to speak, like this pattern that involved various components. And anyways, so I'm like, shoot, this is a thing to read through. So I thought, okay, I know what I'll do. I'll take the diff of it, and I'll just paste it into Chat GPT, GPT 4 because it could afford the full context of the poll request. And then I took all of the reports from Slack, like users have reported the following, and just literally verbatim pasted them, this, this, this, this, this. I'm pretty sure the bug is somewhere in this poll request. Write a list of the most likely culprits. And it wrote five, one, two, three, four, five, like it could be this, it could be this, maybe this, think about that. And one of them was it. And it was kind of like two of them were it. It was like the two of them were hunting in the same area, and combined, they ended up being the issue. Later that same day, similar thing happened where a teammate named Pankaj, he asked a question, and he was sort of stumped on this drag and drop bug he was experiencing in a feature he was writing. And he was banging his head against the wall. And he just sort of yelled into the Slack abyss, like does anyone know why this may be? I'm stuck. And so just to have some fun, I took exactly what he wrote. I didn’t even change it. And I said, a coworker just said this, write a list of the things to- like a checklist of the debugging steps to find the issue. Sure enough, number two, it was it. It said I bet that there's a bubbling issue or some other components grabbing the event and preventing the bubbling from going further. Like so look for that. It turned out that's what it was.
Eric Jorgenson: That's crazy. Is this- that's your method of campaigning to get more people to just use GPT in your company directly? Are you actively trying to get more of the team to sort of get this reflex?
Sean Devine: Oh, yeah. I mean, to say that I'm obsessed with this point I think would be underserving the reality. And I actually find it very challenging. I think about this literally all day long, which is how to motivate everyone to fully embrace the tools. And so, my current sort of philosophy on this is that for me, my own work, what six months ago was good now looks kind of so-so to me, like actually. What six months ago was fast now feels very slow to me. And speed, not to be like an Einstein fanboy, but speed and quality are relative. They're not absolute measures. Like there's no such thing as goodness when it comes to the work. It's all relative to what's possible. And so when you have seen that what's possible has fundamentally changed, that in something like half the time, you can do some amount better, whatever the metric of better is, you can't unsee- I can't unsee it. And so my mission is to communicate that clearly. And what happens now is I think people hear the enthusiasm sometimes as like fanboyness or like hobby, hobbying of some sort. And while there's some truth to that, I think, it's more like straight up- it's more like being compelled to- in other words, I can't- it's clear that we don't have any choice. Like, if we all of a sudden got 50% slower overnight, the business would have trouble. Well, having what's possible become two times higher is the same thing. It's the same thing.
Eric Jorgenson: Especially if competitors figure it out and you don't.
Sean Devine: Absolutely, absolutely. Now, for us, and thank goodness on the following, we have grown very steadily, year after year after year, that's not stopping. So we'll grow, I don't know, 60, 70% this year versus last year, probably about the same the year after that, maybe even more than that. And so, when faced with a doubling of capacity, let's call it a doubling of capacity, for me- and that's like, well, great because now I've just solved part of our medium to short term scaling problem, basically. Now, if we weren't growing like that, boy, it'd be a trickier situation because then you say, well, we have to absorb that productivity either into growth, but if you don't have that, you'd have to absorb it into lower other costs, basically, efficiency in other ways that that better work could generate, or into better prices because of better negotiation or better features, or you're going to have to reduce the cost. But it's going to go somewhere. It's going somewhere. And so for us, it's a little less stressful, I think, than it could be because growth is a pretty good tonic. So growth allows me to say to the team, hey, good news, this is not going to result in the team needing to shrink. We're growing really well. This solves a scaling problem. But that doesn't mean that- the other side of that coin is that every team member is obligated to go as fast as possible. There's not a choice on that part. That's the deal, basically. And long answer to a short question, but I do literally everything I possibly can every day to communicate this all day long. Leading by example being the first one.
Eric Jorgenson: Yeah, it's a very interesting- I think I saw- I was reading a post where the sort of leap in tooling and capacity for an individual contributor, this is on the order of the personal computer revolution. It's more difficult to grok because it's software instead of like you used to have to do this on a piece of paper, and now it's on a screen, and look, you can Command F and sort and search and do all this amazing stuff. So it's going to take a little bit, it's maybe a little bit harder, takes a little bit more of an experience because you can't tell just by looking at someone using a computer that this is happening. But I think that's well put. Like the expectations of output for each individual person have gone up. And this is why. I need you to learn this skill of using these new tools and increase their efficiency and their speed and their quality. Like it's interesting that you call that both of those. I think that's a really important point.
Sean Devine: The quality jump is not talked about as much as it should be. Like the work is just better. Like my work is better than it used to be. I've been doing a thing recently where before every meeting I have, I take four minutes to, in four minutes, attempt to prep better for the meeting. Just do like, I just sort of open a session and say I'm entering the meeting, here's who's going to be here, here's the topic, here's what it’s going to be for, like what are the topics I should have on my mind? What are the- and then I do the same technique I did before. I've become a five times better meeting attendee. I am completely ready for every interaction with like a teeny bit of- like no time, a few minutes, the time I would take just making a coffee or whatever, nothing.
Eric Jorgenson: I think the other thing that people may be surprised by is your example around sort of the solver was very logical. Like there are correct answers, there's correct- the code either compiles or doesn't, performs the function or it doesn't, is something else that you- the examples that you've shared with me, it is great at creative work, especially if you follow this method that you're recommending of like give me 10 options, then flesh out the best one. Like you're using it to name features. You're using it to make a follow up on like icons, brainstorms. Yeah, we haven't even talked about the images yet, honestly. Like go into the icons, the images, feature names, stuff like that.
Sean Devine: I mean, on something to sort of bring it all together on the product side that I think touches on a lot of this is all those other features I mentioned, those were kind of what we did when we were getting going. They were relatively small. They were kind of enhancing a little thing along the way. But once we had maybe eight of them under our belt, they kind of knew how to do it right, our prompt engineering skills were pretty good, we understood how to chain things well, blah, blah. We were prepared. I looked at our strategy. And one of the three items on our strategy is to what we call productize operations, which means to take all the work that we have like operations staff members do to help train new customers, provide customer support, explain things, problem solve, make quizzes, all that sort of stuff, that part of our work is a real problem for scaling for us because it's very difficult to train it. Because it requires a ton of expertise. In fact, you mentioned earlier how many words are in frequently asked questions in a business. So because of this feature, I know how many were in ours. So the good news is that we had years’ worth of release notes and newsletters of content around for every feature, every topic, or not every, lots of them. And so we said, geez, I wonder if we could just use this big corpus of knowledge that we have in order to build a bot version for most of what our operations staff does. And so the answer on how much content, 127,000 words was at the time the sort of size of the release note and newsletter library. And so the first thing we did, which was a fun use of AI, is I said, okay, like our test versions of the answers, the system was good, but it wasn't great. And one of the problems was it didn't have enough language, like it was missing some words. Like it didn't know some words that you need to know, like terms of ours. And so what we did is we wrote a little script or some features that went through every release note and every newsletter and extracted every glossary term that we believed existed in anything we ever said. So that generated, let's say, it's 1000 terms. And then we said now draft a definition for each of them based on reading any of the things that seem related. And that generated, I think, in the end, it was like 450 glossary terms and then we went through and did some editing on. So then we had release notes, newsletters, and glossary terms. And we built on top of that a feature called Hey, Kayla. And Hey, Kayla is you just ask it any question whatsoever, but I mean, any question – how do I do this? I'm going to meet with drivers and I want to explain to them why this thing is good for them. Write it for me. My trucking surplus is too high and I want a step by step program to reduce it, especially focused in this area, write it for me. Anything, anything. And so, we built this feature called Hey, Kayla. And it first sort of assembles, back to this context point, it assembles the related content that may be related to the question. It creates, and this is interesting, back to what I learned personally, it creates at first a clear understanding for the bot of who's asking the question, what their context is, like this is a person that's a driver, they work in this geographic area, they work for this company, in case you see it in documents, etc. And now like, here's the question they have. I listed for you various reference materials. Now, in this amount of budget, you give me back an answer. So, we built up this this feature Hey, Kayla, and it's amazing. It's like it took this area that was a huge stress of mine, which is how the heck are we going to add marginal operations capacity that is expert when it takes like 18 months to become expert, and half of them will quit. Like the math is really bad on that problem. And no longer a concern of mine. Like, the problem- so we release it. It is sort of mind blowingly good. Internally, it's mind- It's incredible how good it is. Especially, it took a few rounds of iteration for it to get great. But now I think it's pretty great.
Eric Jorgenson: Do you run into challenges around just customers not expecting an interface like that to be good yet? Like, they just got good. Normally, those chats are just so frustrating.
Sean Devine: Yes, yes. In a couple of ways. So, I mean, it's impossibly good. So, I get it. I have a lot of- I empathize on this point. So I mean, for example, we've put out multiple newsletters where we give examples, like these are- when we upgraded to the GPT 4 engine as an example, the first three questions that I asked, I'm actually going to look them up while we're talking because I think that they're unbelievable. So I looked up three questions. And I said, I mean, I hope this gimmick would work, said, okay, I'm going to just use the exact verbatim answers to the first three questions I asked it and say I didn't like ask twenty to find three. I just asked three, pasted them, and here they are. Let's see. So here are the three questions. Number one, I said, I'm a logistics manager, give me a step by step plan for remediating late driver arrivals using tools in XBE. It wrote a 10 step, bulleted plan that was on the nose, like just as good as an expert would have done. Number two, I'm a trucking manager. I'm going to have a meeting with my drivers. I'd like to convince them that XBE includes many features that make their lives better. Write my speech with the top five reasons they should be excited. Did exactly as asked, perfect. Like multiple people have used this speech now. It's excellent. What is the timecard pre approval? How does it differ from an admin approval? When should we use which strategy? This isn't- I mean, of course, you don't know the context here. That is an extremely nuanced question that it got right.
Eric Jorgenson: That sounds like a 25 minute email to write.
Sean Devine: Because it's- you know what, you're right. And it's the shortest answer, which is why it would take the longest. And it's very nuanced. It's a very nuanced point, or somewhat nuanced point. And it got all three of them. It just got it right. So to your question, I think a couple of observations. One, it's sort of weird that people don't say what their question is really. So there's very much like a Google search impulse that people have, which is like, for example, they'll type in the words material transaction. Or they'll type in the word like timecard or route. They type in nouns basically.
Eric Jorgenson: Yeah, which we've been slowly trained to do by search engines for like the last 20 years.
Sean Devine: And the results are always interesting because it's like they're right, because what we do is we both answer the question, and then on the left side, we provide all of the reference materials that we have gathered for it. So let's say there are 75 documents that we've assembled that we think are related to this question enough to care. So those are on the left side of the screen, the right side of the screen, kind of like Bing, is the answer. Now it's funny, when someone just like puts a noun in, like material transaction. I saw that one this morning, the answers are fascinating. Like that one, it was sort of a remarkably good answer. It said, I see you asked about material transactions. Here's what they are. And here's like the different situations where you may care about them. But it was kind of like if you just sort of like blurted out a word, but you were obligated to politely sort of engage in an effective conversation. It did pretty good, pretty well. But on the left side, it had just various reference materials. But the interesting thing is, if you just took 10 seconds to say exactly what you want, like someone told me to enter a material transaction. I'm a driver. I don't know how to do it. What do I do? Like it’d get it right basically every time. So you're right. It's like, I don't know if it's the Google training. I don't know if it's the suspension of-
Eric Jorgenson: I wonder if it'll change- like how much the inputs will change. Like, is it actually that you would need them to submit a short form? It's like, what's your role? What company are you at? What problem are you trying to solve? Or what information do you need? And then you can turn that on the back end into a- like turn nouns into a story through a different user interface.
Sean Devine: I think my take on that- I think maybe. I'm not sure yet. Like, it's clear to me that, I mean, just yesterday, we added that, we just merged something that added a bunch of context that we knew about the person that- sort of of the sort you just said, which says like here's a bunch of stuff I know about them. And it really improved the answers. We had before and afters where it's sort of remarkable how context will affect things, just like people, but you give it- you frame the situation, and it's much more likely to sort of shoot correct in uncertainty. So I think maybe.
Eric Jorgenson: Which is interesting. This is the first interface that like when you respect it by giving it the most context, you can actually expect it to reciprocate with like meaningfully better results. Google doesn't do that. The more keywords you add, it just gets more confused.
Sean Devine: It's interesting you just said that. Just as an aside, today, I broke a rule that I've never broken before, which is my wife and I have bought like various stocks over the years, and the rule is we don't sell. We only buy.
Eric Jorgenson: I like that rule.
Sean Devine: Yeah. And today, I broke it. For the first time.
Eric Jorgenson: What’d you sell?
Sean Devine: Google. Yeah, I just, I'm a big believer in kind of doing the things I believe, so not just talking, but like doing them. And I just was- I was talking to Teresa, my wife, and I was reflecting on how little I use Google now and how business model wise- I mean, technically, I don't doubt that they've got what it takes. But business model wise, they really are up against it on this one. And in my experience, business models are extremely powerful. And so I broke the rule today, first time, sold Google because sort of related to the point you made plus my own experience we talked about.
Eric Jorgenson: It's going to be a fascinating case to play out. I mean, we know that they have capacity, we know that they own Deep Mind, we know that they have been, in theory, working on this for a long time, but they also have the most to lose and the conflict- like it's going to be- it’s the innovators’ dilemma of the decade. Like it's going to be so interesting. I can't imagine that they do nothing. But every day, like yeah, it's fascinating. Okay, let's try to like realign with the sort of part three, which is organization. We've touched on it a little bit. But I want to kind of give you room to run on this because I think this is- all of these are applicable, but individual contributor and organization managers, I think a lot of the audience here is founders or investors. And I'm fascinated to sort of hear how your vision of an organization and how it runs, the inputs and outputs of it, the expectations of it sort of have changed in the last couple of months as this has really come- the power of this has sort of become obvious to you.
Sean Devine: Well, I love this topic. And I think a lot of it's very scary, so just as a disclaimer, trigger warning on all of the following. So okay, just I'll go rapid fire on this one. So one, value of expertise has cratered. And I think it's critical to go through the organization both in terms of who you currently have on staff and who you are thinking about having on staff and ask the question for everyone, to what degree is that person's role about what they know? And if the answer is a lot, they either are going to have to learn some other things, or you're not going to hire that job anymore. And I don't think that that's- this is, I think, a very sort of counterintuitive moment in this area in that previously, these sorts of technical innovations impacted the bottom up. This is, to my mind, the first time ever that it's going the other direction. So these innovations impact the top first and then go down. So they impact the highest paid, most credentialed, most educated people more, or as much at the very least, as they do the low level, people with tons of expertise, where when all of a sudden, it's rendered to not matter much. And I don't mean, I'm speaking in absolutes, I don't mean it in absolutes, but it matters a lot less. And so I think that's step one, is say, hey, have we invested a lot in expertise, do we have a lot of capacity that's about expertise. And if we do, we can repurpose that capacity into other things. The larger- like a smaller organization like mine, it's not a huge deal because you don't have specialists, like just not big enough to have specialists. If I was in a bigger organization, like I've been in the past, this would impact us a lot because you have a lot of specialists in bigger organizations, and those highly specialized sort of technical roles just aren't that valuable in the very, very near future. And so I'd say that's sort of category one. Category two, I think recruiting generally is going to change, I think is changing a lot right now. So a couple of examples. We are focused primarily right now when we recruit in two areas. One is problem solving by itself. So whereas we've reduced the amount we care about their technical expertise, and I'm saying whether they're a technical person or an operations person or a salesperson, but what they know matters a lot less than it ever did before. How they approach problems and break them down to their components, basically do the work I talked about before, like decomposing problems into smaller problems, evaluating the option space, going down the promising paths, so on and so forth, that skill is the one that's valuable. And figuring out how to- it's much more valuable relatively than it used to be. Because it used to be that you'd have to actually do the thing, but now the doing the thing is getting a lot easier. And so knowing what to do is the thing. There's a lot of quotes-
Eric Jorgenson: It happens more often, like knowing what to do. I mean, 180 questions in a day, like that's the iteration. That's the feedback loop now. That's the reorientation.
Sean Devine: There’s this great quote that I've heard Horace Dediu say, the guy that writes Asymco. And he says something like the answers, my answers are free, but the questions you got to pay for, some sort of pithy thing like that. That's becoming all work basically. The answers are free, the questions you got to pay for. And so who's good at asking questions is the skill to interview for. And I don't know that we figured out how to interview for that yet. Like, we're actively working on solving that problem right now to say like, how do we identify? Now, shocker, this is something I'm deeply engaged with GPT 4 on, trying to solve this particular problem. We've architected entirely new sort of interview rubrics that are built around this one. So that's point one about recruiting. Point two is stamina. And I think this may be slightly unexpected if you haven't sort of been deep into this. But this style of work is physically difficult. It is unbelievably intense. Because there isn't time off. It used to be you'd be like waiting for someone to speak, or you have to have a meeting to collaborate on a thing, and so you're one of six people, and so, on average, you're only engaged one sixth- you only matter of half the time, you're only physically speaking one sixth of the time, etc. It's not that way when you're working in this style. You're basically constantly working. And it's like a very counterintuitive thing when you hear it because you're like, well, I thought that the machine was doing it for you. You're like, no, no, no, no, no, no, the machine is doing things, but like you don't have any time off because you're just constantly engaged. And I've found, I mean, as you know, I'm sort of a bit of a health focused person and good shape, and I found I am beat at the end of the day. I am really physically giving it my all every single day I work. And that kind of work ethic, I think it's a combination maybe of work ethic and stamina, sort of like just the ability to stay focused for that long and just to endure that kind of pace is, I'd say, probably not for everyone right now. And I think many people could develop that stamina. Work ethic is a bit of a question. I'm not sure on that one. But so to answer your question, we're trying to recruit for those things, problem solving first, stamina, work ethic two, and then the third one is small identities, that finding people that have low ego that won't be upset by all of this. Because, I mean, if you were to construct a machine to destroy identity, I think you'd come up with this. Like it is punishing in that it takes things that people's entire identity are built on, the ability to write well, the ability to be clever, the ability to know how to chant into the API the things one needs, and it's like great at all of them. And all of a sudden, as an individual, you're like, what am I? Like, what am I now? I have, and this is like a tremendously dark aside, and I'm going to leave out a couple of details that'll be clear in a second. But I actually have one step removed personal experience now where someone's business is sort of in the bullseye of what's going to be affected this year, or is affected maybe now, because of AI related advancements in that industry, that attempted suicide on Friday. And in their note about it, they wrote, I can't be me anymore. That was the punchline of the note. Now they survived. They are going to be okay. But that's not like a conjured example. That's like manifesting this point around identity deconstruction. And my own experience is that nothing is more valuable to people than their identity like in the world. And I've never seen anything more effective at tearing down identity than this. And so the third thing we're recruiting for is small identities, just people with very low ego, very little attachment to who they are, so to speak. Because I think it's critical. I think just it's hard to endure emotionally the kind of transformation of what's valuable otherwise.
Eric Jorgenson: It's very interesting. And I appreciate how sort of deeply you've thought about this at all the different levels of engagement. And I identify, even I’m early in the rabbit hole, you're many months and many, many hours ahead of me here. But it's overwhelming. Like, it's overwhelming to figure out how to interact with this new thing. It's overwhelming to confront the reality that it's, I’m not going to say omniscient, but it knows orders of magnitude more than we ever will. And it knows it instantly. It's like that demoralizing feeling when you're like playing chess against the computer, and you think about a move for five minutes, and it moves instantly, and it whoops your ass. It's that for work. And there's a very interesting- like it's on your team, that it the good news, like the thing’s on your team. But what you said about the pace or the stamina is really interesting because it can't work without you. It's not working in its sleep. It's making you more effective for every question that you can ask it, every hour that you can spend in the saddle. But you've got to be there for it to multiply your efforts and your output. I hadn't considered that before. But I think that's a really sort of good thing to understand about it.
Sean Devine: I mean, it's one of those where I think one could either understand it, theoretically, or just like give it a couple of months. It'll be pretty clear because it's very much a- it's a very real experience.
Eric Jorgenson: I know so much of- it sounds at least like a lot of what you have learned has been sort of self taught, and you just kind of jumped in the sandbox and started playing around with it. Are there resources or any places that you've gone to learn this stuff that you would recommend or that you found helpful?
Sean Devine: So I would say, I a few months ago subscribed to all of the best like AI centric newsletters I could find. And I have very much in hindsight benefited from that. Now, there's a lot of news that's come out. So it's a little overwhelming. It takes a bit of a commitment to stay up on it. But if I could subscribe to one, having I think I subscribe to five now that I read, but the one that I think could suffice is called Ben's Bytes. And I'd say it's similar to some of the other ones, but it's a bit more right down the middle, which is kind of what I was looking for. And now the newsletter itself is not going to teach you enough, but it will point to all the things that would be enough. And so, yeah, I just started to read the details on a lot of things. And then, I'm kind of a doer, and so I just started to implement some of them. And then, that kind of creates its own energy and feedback loop. But Ben's Bytes I'd say is really good, and especially if you commit to reading anything that seems relevant from it.
Eric Jorgenson: Okay, interesting. Now, I know we wanted to kind of stay focused as much as possible on sort of the today, like what are the tactics and stuff that you're changing right now. But I want to get in your head a little bit as sort of the leader and owner and this is your baby, this is your livelihood, this business is it. How do you think about the landscape now? Are you thinking differently about sort of the size and structure of your company and maturity? Are you more worried about competition? Like, in theory, the barriers to entry for building the software are much lower than they have been. Is that like- how have those thoughts evolved I guess?
Sean Devine: A couple of things. So as it relates to XBE specifically, I think we're in a really good spot thankfully. It's because the product is so complicated that it's so expansive, and it's not- there's some innovations where it's like there's one thing that is the innovation, and then there are some that are it's like there are 10,000 little things, and it's the collection of them that's the innovation. We are more like the latter, like 10,000 things. And I think that if we stood still, I would be concerned a bit more just for the simple fact that what took us seven years could take, well, if my numbers were right before, could take three or three and a half maybe, something like that, all other things being equal at least. Now, thankfully, we don't really have much debt, technical debt, so our infrastructure is really, really good, and our velocity is really good. And so we have picked up the pace and are now sort of deploying our capacity and expertise, both as it relates to leveraging AI in features, like the ones like Hey, Kayla, like safety risk, all the rest of the ones I said before. Like we've got one coming soon that's interesting, which is a bot timecard approval auditor, which is like a job that someone does now that's actually quite a like mid level job, that like we've experimented a little bit, and it's very clear you can teach GPT 4 to do this job well. And so that's an example of AI being in the feature. But then there are lots of features where it's AI helps us make features. And I actually think that in the community at large, there's insufficient talk about the former, or about the latter rather. There's a lot of talk about AI features, but there should be at least as much talk about AI helping make non AI features because most features or many features, at least, are non AI. And then lastly, there's the business side of it, which is we've deployed this in our sales and marketing game about determining researching prospects. I've always believed in our business that having individual sales and marketing strategies for every single prospect was the right move because we have limited Tam, so to speak. But now it's definitely the case. I mean, we can afford to completely craft totally bespoke plans for 100% of our target markets, a totally new game in that way. And so, I guess what I'd say is I think the threat is pretty real in that everything's faster. And I think if you stood still, you could get caught much quicker than would feel possible before. And so, in our case, we can actually press that advantage, though, which is like we're good at making things and we are in a good position. And so we can, as long as there is additional value to create, which there is for us given the size and scale of what we do, then if we just go faster, at a minimum, we maintain the current lead. And I think more likely, it just diminishes someone's willingness to try.
Eric Jorgenson: Interesting. Yeah, there's no risk of you slowing down it sounds like. Do you think that this will cause margins in your business to increase or decrease?
Sean Devine: Our specific business, increase.
Eric Jorgenson: Software at large, software industry in general?
Sean Devine: Decrease. Yeah, I would- I mean, put me down for an overall bet on deflation generally in this area. I think the price of things should collapse in a lot of areas. I think even if margin percentage has stayed the same, which I think will be difficult, except for organizations that are super agile and mostly variable costs. I would think you'll both see the revenue decline and the margin percent decline given, well, given leverage. I mean, it's a fascinating moment that way in that I've for my entire career obsessed about leverage. And it has taught me, I think, what to expect now, which is that everything that was good is going to turn bad. Like leverage works both ways. I mentioned that dinner the other day. I worked for Conway in 2007, ’08 when the Great Recession happened. And I remember going into work the day after, what was the name of the bank? The day after that bank went out. Lehman. The day after Lehman went out. And our volume went down 25% the next day. This is a $4 billion revenue company, 25%. And at the time, I was head of pricing and strategy and engineering there for the whole company. And we had 440 locations and 20 some thousand employees, but it's the locations’ massive operating leverage. So tons of leverage. In other words, the business was pretty great in good times. And I remember just walking into the boardroom with the executive team, and it was clear we were screwed, like super screwed because of leverage. So in other words, I had this one experience in my career where leverage worked the other way, and then most of it is that leverage worked for you. And now I think I can see pretty clearly how it's going to be the other way again.
Eric Jorgenson: Yeah, this is going to be a very interesting sorting function. And I think the distinction that you made earlier is good, like those who are many years into a very complicated product within the software industry I think are, especially if they have existing customer relationships and integrations and practices around it, are relatively safe. Maybe it's easier to approach building something to replace it, maybe that could happen cheaper, but also it's scary. I think the simpler the product, the weaker the moat for sure. But there's a lot of software with what is truly like a technical edge and a foundation that is like an engineering problem that hasn't been solved before, some anyway, maybe not the majority but enough that that's also a different case. It'll be fascinating.
Sean Devine: I think point solutions, so I would be very concerned if I was a point solution, like incredibly concerned. The more unit task the thing is, the more specialized, the more, even if it's technically advanced, I actually don't think- I don't think that there's any point solution that’s just very good at a thing that is going to come out clean, unless it's like, I mean, there are exceptions to prove the rule clearly. But for the most part, I think anything that's like that is it's either a large language model will be better at it, which is going to be true for a lot, or they're just going to be effective at creating a competitor cheap. And whether it's A or B, it's trouble.
Eric Jorgenson: Okay, as we kind of wind down, wrap up here, I want to- just you've touched on some of the softer side of it. But you're six months down this journey, you seem enthralled by the power of it and aware of some of the shadows that it casts, and I'm curious sort of how you feel about it.
Sean Devine: So, I think my career I'd split into three sections. And I've thought about this long before this sort of AI moment. And the beginning part of my career when I was in my 20s, my identity was that I was smarter, which is terribly gross and not that effective, but nonetheless, I think was the case. And I had a moment in my late 20s where it just became clear to me that I was a little impressed with myself and that I wasn't impressed by anyone else that was impressed with themselves. And I was like, well, that seems to tell me something. And so I decided like on a day that I was changing. And the thing I decided that would be true is that I would work harder than anyone else. And so, sort of phase one was smarter. Then I got very turned off by that as I should have been. And phase two was work harder. And that lasted a bit. And then I looked up and had worked- I was working for big public companies at the time and had a strategy, had a price in kind of like the applied science smart guy jobs. And it occurred to me I hadn't thought much about what I was working on at all, ever. Like I was just working on the hard problem. That's what I was working on. But like the goodness of it, just it wasn't part of my thinking. And including who I was working with and for on things. And I kind of had a similar moment, which is like, boy, I'm not really impressed by people that don't consider who they work with and what they work on. Like, I find that kind of gross. And so maybe I should listen to myself, like my own judgment of others, maybe I should apply it at home. And so I hit a new moment and said, okay, I think that being smart is kind of gross, and working hard can be gross if it's going to the wrong thing. And so the rest of my career, I'm just going to try to be courageous. Like, I'm going to try to do the hard- like do the thing that I know is right. And I think, and that sounds quite full of it. But I found it to be quite helpful, and it led to the best phase of my career where I was still working hard on things, but I was like judging much more the what I was working on and who I was working with. And anyways, long way to say that I think that we're in the moment of our lives workwise, so I was working at the beginning of the internet, so I graduated college in ’99. And that's right at the beginning, and I jumped right into it. So I worked through that. I worked through the Great Recession in a very sort of interesting job, as I mentioned. I was here for the iPhone and rode that careerwise and have done well. This is so much more than all those, and those were all huge. But it's obvious it's more. Like I can remember what those were. And I was enthralled by them too for what it's worth. And so I guess my point is that I just don't- I think it's so scary to see what's the degree of change and pressure that it takes, it's going to take a lot of courage to not go into a hole, to go into denial, to run from it, to lash out at it. And I think all those are futile and will lose and are cowardly. And so I'm just going to face it. That's how I feel. And that's not to say that- I'm sort of into it too. Like, don't get me wrong. I think it's exciting. I think it's exciting to be able to work at this pace. I like to work fast. Like I mean, don't hear me denying that I think that the work itself was fun, because I do. But it'd be callous not to every day take a step back and go this is a moment, man, this is the real situation. And a lot of people I know are going to be dramatically impacted by it, dramatically. Every college kid I know, literally 100% of them, many of my professional friends. And so, I just don't want that. I don't want to be a Pollyanna. I don't want to pretend that's not true. I don't want to- I just want to be at the front and do my best.
Eric Jorgenson: Yeah, that's incredibly well put. I think you've done a little better than facing it, which I'd characterize what you've described to me in this podcast as making the big scary monster your biggest ally and putting them on your team, like by your side. And that sounds like a winning strategy to me. But I think you're definitely correct to say that it's hard and it's scary and it takes a lot of character and courage to do it.
Sean Devine: Yeah, well, I'm trying. I'm trying. And I feel like, on that point, if it's scary for me, I mean, I'm in a great spot, I've had a great career, I'm financially stable, I have a family that's wonderful. I mean, I'm in like the 100th percentile of good situations. I mean, it'd be freakin’ really rough even if you were 25% of the way down that ladder. And yeah, I think that sounds like a situation that needs a lot of leadership to me. So, I'm just going to try to do it.
Eric Jorgenson: How much do you feel like your technical expertise and background was necessary or helpful? And I ask that both as a literal question because I'm curious about it, but also because I don't want to scare people away from feeling like they are capable of doing technical things just because they don't necessarily have the technical background. But maybe we should.
Sean Devine: No, I don't think so. I don't think so at all. I think back to before, I think to the interviewer characteristics I said we were focused on now, which were problem solving, stamina, and identity. Those are the ones that matter. So, I think the reason I've had some success is that I'm a good problem solver. Like I'm systematic and break things apart and make big problems small problems and consider lots of options, all the by the book things. And I think that's critical. And that doesn't require- I mean, I think programmers may be on average better at that. But lots of non programmers are good at that and certainly can be. Stamina, that's its own thing. I mean, so again, I think a lot of elite programmers have great stamina, but a lot of non programmers have good stamina too. And then on the third one about identity, man, that one's a big one. And I think programmers are no better than others on average on that, or maybe worse on average.
Eric Jorgenson: Cool. Amazing. Well, thank you, thank you for taking leadership in this space, like intellectually and in sharing that with us in your company and now to anybody here who's listening. I think this has been a hugely helpful conversation for me, honestly. I'm still early enough to feel just like kind of overwhelmed and confused by the whole newness and power of it, to be honest, and I'm excited by it. But this has been really helpful to me in feeling like I can get a rope around it and turn it into a friend. So you've been an amazing example, and I appreciate you sharing all your experiences to help others do that.
Sean Devine: Well, I appreciate that. And I'm happy about that. I mean, so for anyone that listens to this and is interested in any part of what we talked about, whether it's the how to, whether it's an example. I mean, I'm interested in the leadership culture management topic, the strategy side. I mean, I'd say my interest is somewhat limitless on the topic. And I am not, by dint of the strategy before, I'm not sort of deeply enmeshed in the tech world because we didn't raise- we haven't raised money, we don't interact with venture capital, we aren't in that, we don't hire mostly in the US. So, a lot of the reasons people would be- oh, I'm in Kansas, like you, you're down the road. So that sort of takes me out of the scene.
Eric Jorgenson: Which I think is also- that's a huge testament, that's part of the reason I love holding you up as an example of this actually because like it doesn't have to be a tech scene to get into AI. Like, people in tech are excited about AI, but everyone should be excited about AI. And there are applications sort of all across the gamut no matter who, when or where you are, what you're working on, like this is leverage that can help you.
Sean Devine: Yeah, absolutely. So anyways, I'm interested in connections. Because I don't spend most of my time connecting with other founders in the moment, and yet I'm social, and I like to. So, reach out, I'd be happy to chat.
Eric Jorgenson: Where should people reach out? Where are you easiest to find or prefer to meet people?
Sean Devine: Yeah. So I think good on Twitter, that's always fine. I'm barelyknown on Twitter. Like, that's the handle. Yeah, barelyknown. All these years-
Eric Jorgenson: You can't get too popular with that handle.
Sean Devine: I mean, that's the- This has been a long play. Like, you know what would be great, is if I was well known, and I have the handle barelyknown. If you know me, that's exactly what I want in life.
Eric Jorgenson: It is, yeah. That'd be really funny when you have a million followers.
Sean Devine: So I'm barelyknown on Twitter, and that's probably a good place to start. But you could reach me anywhere else too, email or LinkedIn or whatever it is. So yeah, I welcome that. And then the second one, can I plug my-
Eric Jorgenson: I was just going to say any closing thoughts, anything else you want to share?
Sean Devine: Okay, so we talked earlier about that feature, Hey, Kayla, which is our big AI feature. It's named after my daughter, I think I mentioned that. She works for XBE as, well, she's both a manager in our customer operations group, and now she's sort of taking on some of the marketing duties as well. So her name's Kayla. And she's amazing. She's smart and fun and funny and adventurous and beautiful, all the things. Eric has met her and I will not put him on the spot, but he can vouch silently for these things.
Eric Jorgenson: You don't need to put me on the spot. I testify. Everything he says is true.
Sean Devine: So here's the thing. So she is all these amazing things. It's the one of the joys of my life to be able to work with her. It's just a fabulous thing. But you can't be good at everything. And she's not good at dating. She's not. You got to, back to the having a small identity, you got to know yourself. And she's blessed with all of these wonderful gifts, and it has not left enough for dating skill and capacity. And so she went with me on two trips to India, which was also amazing. And about half the people we know in India have arranged marriages, and she was commenting about how it seemed like such a reasonable way to find a good match of a partner, to have some collaboration from your parents. And so, I was preparing for this podcast and said, you know what, you know whose listeners are kind of exactly your demographic? I mean, I don't know you, listener, but I know you're possibly in the demographic. Eric's listeners, that's who.
Eric Jorgenson: This is a big, big pile of nerdy dudes for the most part. I've done demographic surveys and that's just basically what it sums to. You know who you are.
Sean Devine: That's right. And so of the listeners that fit that description, the percentage of you that have said, that are single that have said like, man, all the good girls are taken is not small. I also have friends. Except Kayla Devine in Kansas who is single and amazing and bad at dating. And so anyways, so I said to her, I’m going on Eric's podcast, and you have asked me if, like an Indian father, I would engage in finding a suitable match, and I have accepted the challenge. And so anyways, this morning, I made the website, Bae Kayla, just as a riff on Hey, Kayla. And it has a little gallery of photos, a little bit about her, and a little text box where you can ask her out if you meet the criteria she's listed. And so that's my plug, go to baekayla.com, and if she seems like a good match, and you seem like a good match given what she's listed out, ask her out.
Eric Jorgenson: I love this tactic. Like screw Tinder. We're going to make like a one woman landing page and just lean into the demographic, targeting the demographics that you really want to reach through a trusted filtered channel. Like obviously, what better demographic is there to go fishing for than listeners of this podcast? There's no better group of people in the world in my humble opinion.
Sean Devine: I mean, that may be more on the nose than you realize for this single woman. But anyways, there we go. There's my plug. None of you own asphalt companies, and so I'm not going to waste my time on that plug.
Eric Jorgenson: If you happen to be in the horizontal construction industry and understood what was going on- No, I have a feeling if you're a customer of Sean's, you've probably heard from him already. Which is, yeah. So it's a privilege to have you on to share all of your hard earned wisdom from a niche industry that has so much to teach everybody. And really all the conversations that we've had I've learned from and I've always admired how you thought about building your business, building your team, doing your work. And I'm glad we finally got to record one of those conversations and share it with people. It's a privilege. Thank you, sir.
Sean Devine: Man, me too. And I got to the end without saying how much I love your first book and can't wait for the second. So there we go. I didn't forget.
Eric Jorgenson: We'll cut that.
Sean Devine: Leave it in.
Eric Jorgenson: I appreciate you hanging out with us today. Thank you so much for listening. If you liked this episode, previous episodes you will also love is number 38, Chris Ho and Robert Hayes of Athena. We talk a lot about the leverage inherent in using EAs and personal helpers, investing in the corporation of you, a lot of the same themes even though we're talking about people leverage there and not AI like we covered in this episode. And for operators, CEOs, company builders, also episode number 28 with Jeannine Seidl, talking about how people teams really help build companies, thinking through your HR function, when to start which pieces of the puzzle, and how to handle those challenges that come up on the people side of things as you scale your companies. I'll also remind you, accredited investors can invest in early stage companies alongside me and my partners in Rolling Fun. There's a link to more in the show notes. Our guest today Sean is actually an investor in the fund. We're very proud to serve him and entrepreneurs like him to get capital into world class startups, so they can focus on their very important day jobs. You can hire a talented team to build some very excellent software for you by using this episode's sponsor Bread. Go to madebybread.com or click the link in the show notes to learn more about that. For a free way to support the show, please leave a quick review in your podcast player or text this episode to a friend or coworker you think would enjoy it. Or you can tweet it; that works too. I'm so glad you enjoyed this. I hope to put my words in your ears again soon.