Latest / Construction Technology Podcast / Construction Technology Podcast - Episode 004
Transcript
- Luigi La Corte: Hello everyone. Welcome to another episode of construction technology podcast. I'm Luigi, a CEO provision. I'm here with Mike from build vision and Andrew from join. What's up gentlemen. How's the week going? How's this week been? Andrew Zukoski: Good customers in New York, customers in Las Vegas on the road. So I'm traveling now, so I got the ⁓ Decker behind me. Luigi La Corte: Are you traveling right now, Andrew? Yeah, the classic above doorway hotel look. That's good. And Mike, what's going on with you? Mike Powers: You Andrew Zukoski: Yeah. Mike Powers: Any drama in Vegas? Any good insights for the conference you're at? Andrew Zukoski: I'm at like pre-conference stuff and we're literally at the edge of town. So you look out the window, like you know we're in Vegas. So, no, not lot juicy goss yet. Luigi La Corte: Are you going to advancing pre-con in May in Phoenix? Okay, nice. be there as well. It's been a really cool week. I don't think there's been like a ton of headlines, but obviously I think on my docket today, there's that anthropic report from their economists, which is making the rounds, talking about every kind of element labor that's gonna be impacted by AI and where it currently is. That one has been going pretty crazy. Mike. Mike Powers: probably the best way to do it. Andrew Zukoski: Yeah, we'll have presence there. Luigi La Corte: It was just brought up a LinkedIn article about how good Claude is as an estimator. I have something to say about that. And then Turner posted some crazy, crazy data center numbers. So which one do you guys want to dig into first? Mike Powers: You want to start with the data center piece. Turner released its financial earnings this week on Monday, the third, I guess that's not Monday on Tuesday. So Turner is a wholly owned, % owned by a German company called Haktief. ⁓ Haktief is owned a company called ACS Grupo. So really this is the only lens we get into like the financial reporting structure of a general contractor. Andrew Zukoski: Data centered. Mike Powers: Um, what I found fascinating about this is, um, Turner was reporting 44 % year over year growth. Um, so looking at the numbers, uh, for their, uh, yeah, the Turner's revenue increased 40 % from 2024 reaching 29.2 billion. Uh, the company's backlog grew 34 % year over year to 44 billion. Um, Insane, uh, data centers represent approximately 37%. of Turner's backlog and continues to expand. And then I know Luigi's had up the article about ⁓ data centers slowing down because of permits and electricity capacity. So kind of seems like a tale of two different stories. ⁓ Curious to hear what you guys are seeing in the space. Like, are we gas to the floor running as quickly as possible on the data center side, like the kind of the Turner reports showing or Are you starting to see a little bit of friction, little bit of cracks and slow down on the data center build outside? Andrew Zukoski: What were the permits specifically that were slowing down? Was it power permits or was it just permits as a whole for data centers? I didn't read the article. Mike Powers: Yeah, like permits as a whole have been reduced and they were correlating it to power delays as one of the main driving factors. You can obviously understand there's specifically in the geopolitical space right now, there's a massive spike in natural gas prices, which over the last year, a lot of data center builders have switched from renewable energy to natural gas. saw that with the increase in profit share or market of GE, Vernova and Siemens who are both natural gas turbine producers. it's kind of a little bit of both. ⁓ I'd say it's like permits plus power delays. Luigi La Corte: pretty remarkable is just like the amount of growth that they've gotten, right? Typically they're growing like between three to 5 % per year. They grew 20 % between 2024 and 2025. And this looks like record backlog. I know this is a story that I've heard before from other people, like DPR seems very busy as well. They also specialize in data centers. Anybody who's touching data centers, like we spoke about before, know, seems to be quite busy. Like how long this is gonna happen for or like? how long they'll feel this growth, I really do wonder, but I mean, it's great for them. Mike Powers: I think the, I know we've talked to, don't, was it last pod or the previous one we talked about how Turner makes money. ⁓ So in the ACS Group O earnings report or earnings call by the CEO, they even referenced by, he even references by name Turner's position as an integrated service platform. ⁓ So it's kind of tied to this idea that the CM fee actually doesn't matter. There are other services, there's SourceBlue, XPL Offsite, Turner Engineering, Turner Technical Services, Self-Perform, First Equipment Company is where the revenue's driving. So if you think about it, they're increasing their revenue by 44 % and the dollar capture that they're holding onto is probably already increasing as well. ⁓ So if you think about the margin growth that they're able to capture, it's gonna be massive. I would love to see this compared to other... privately held general contractors in the top 10. know a lot of them are either family owned or private equity owned. So it's a little bit harder. But if I'm looking at this and I'm thinking about like, every dollar of construction that I put in right now, I captured 2 % because I have a 2 % CM fee. Turner's coming out here and say, Hey, we're not producing a dollar. We're producing $2 and I'm holding onto 4 cents of it. So it's the double whammy of their, but that, you know, 4 cents is either 1 cent, 32 cents. So doubling or so I guess it'd be 8 cents. So yeah, so know, quadrupling earning potential and also focusing on structures that have better EBITDA ratios because it's not as labor intensive with like SourceBlue, First Equipment, things like that. Luigi La Corte: They seem really well positioned. mean, not just in the U S also in Europe in 2025. I think this was kind of under reported. I only saw this recently, but they acquired an engineering company in Europe. Um, what's the name here? Is it Doran Dornan? Yeah. Dornan engineering construction firm. So like number one, they're basically pre-selling into 2026. They seem like they're going to be floored. Like they already, if, thing goes, if everything goes well and they can capture their backlog, they're going to have way more revenue than they have now. But. Mike Powers: Yeah. Start with a D. Yeah. Yeah, Dormin. Luigi La Corte: who is going to own the similar story in Europe? I Turner seems like a good shoe and like they obviously have a European owner and they seem well positioned ⁓ on the owner side or at least in like the pre-construction side in Europe as well. Mike Powers: Yeah. We're seeing a lot of mechanical electrical rep firms in the United States expanded to Europe because their customers are pulling them there. So the leading charge still seems to be the United States on data center build out. And then the European union would rely on data center builders from the U S to get expansion. Andrew Zukoski: what we see our partners have subsidiary or have offices in Europe largely. They're pulled there by owners and so it might be specialty pharma like science work but a lot of it's data centers. Mike Powers: kind of see. Luigi La Corte: Did you guys see that video of these protesters who were, I think it was in New Jersey. They were, I don't know what specifically they were, but they're clearly protesting the building of another data center somewhere in New Jersey. And they got the project canceled. I think they're going to turn it into a park. This made the rounds on Twitter, I think two or three weeks ago. And Gary Tant from YCU, I think responded to the guy who said triumphantly that we won is like, this probably would have generated about a hundred million in tax revenue. But you know, you have a park now. Mike Powers: Yeah, same thing happened. Luigi La Corte: I think there's going to be some real opposition, right? Like we saw even recently, and this is not necessarily construction-based, but you know, what I talk about with my customers is like what's going to happen in a couple of years. And what consensus seems to be is like there will be, action will be taken because these models are very capable. There was a democratic socialist in New York who just proposed, I think kneecapping LLMs to work on medicine or. finance or law. I think you've probably seen this, Mike. And I have to imagine that there's going to be more opposition to data center buildup. It seems to become like the poster child for, or maybe like this proxy or scapegoat for opposing AI. Mike Powers: law engineering. Andrew Zukoski: It's the most visible easily thing. It happens in a place and it has strict requirements terms of power going in and workers going in and out to build the thing. Does that make sense? Yeah, I don't They might be bedfellows, I saw that headline ⁓ that looked to me a of professional groups. ⁓ looking to use regulatory capture to solidify position in society as the blessed providers of medical or legal advice or engineering expertise. I don't why, don't that one's right. I question whether amount of we're putting into data centers is the highest and best use of society's resources, but the... Keeping LLMs from talking about engineering seems so incredibly wrong headed. Like, you already have to have an engineer stamp things. they've already got that on lock. We already require people to take on the personal liability and like assurances that like things are gonna stand up order for them to be built. I don't understand. That'd saying that, you know, computers can do math, but you can't do structural simulations because it threatens, you know, the livelihood of an engineer. So I was frustrated to see that. That felt like groups who were deriving good from limited access to expertise, like fighting to control access that expertise. Luigi La Corte: Yeah, I don't think there's really any stopping us. I think we're just gonna have to get on board. But look, we can move on to that topic. I think it's a really interesting one. Mike, you wanna say anything else about the Turner numbers? Mike Powers: I think overall it's just a great business. Wish I could be investor in the Turner Corp. Not sure how I feel about it having to be structured underneath ACS Grupo, but not financial advice. think the Turner entity is a very, very interesting second tier data center play. Luigi La Corte: Hmm. Yeah, like they, they, they're a fantastic business. think anybody who's touching data centers is like, is ripping it right now. I think even like in the Canadian context, ⁓ PCL seems to be doing ⁓ well. have some exposure in the U S they're not really in data centers, but like the growth in construction seems fantastic. I forgot who posted this LinkedIn post, but it was, I think it was a Josh from, from tri Roger. He was showing, of the returns of, I'll have to go find the post, but it was the returns of three publicly traded SaaS companies, 20%, minus 18%, et cetera. And then the year-to-date returns of construction companies, which are all up and to the right. So it's a good news story for construction companies right now. Okay, let's talk about Anthrobic's report. Let me just pull it up on screen. Basically, I think it was an economist from Anthrobic published this report on the theoretical capability and observed usage by occupational category. Basically, where's AI currently and where do we think it's gonna go? And they did this radar chart, which is really hard to read. basically, if you look at management, finance, computer, legal arts and media, AI is coming for your job, office and admin, basically AI is coming for your job. The red coverage part here is where it currently is, which kind of seems hard to believe, right? Like from my perspective, AI is such a critical part of my job now, I'm using it basically every second of the day. This seems like it's underrepresented. Like I was surprised at how small the penetration actually is. What were your guys' thoughts when you saw this? Andrew Zukoski: extreme skepticism at the blue line. Like, tell me about the theory and the derivation of what fraction of tasks can be performed by LLMs. Tell me about the hard laws of the universe that are being composed to ⁓ drive this theoretical blue line that drives ongoing investment into a money-losing business. ⁓ Luigi La Corte: Are we warm? Andrew Zukoski: Yeah, the blue line seemed like all hype to me. ⁓ Some of them, the computer and math one, like, yeah, okay, that's probably a pretty high number, but architecture and engineering, so much of architecture and engineering is site inspection, is construction administration, it's physical on-site work, even for A &E, let alone like trade or labor installing things. I don't know how that one gets to 85 and so. ⁓ Yeah, I suspect the blue line, but adoption patterns in, like the adoption pattern in sales is higher than I thought it was. I don't know what that is. Maybe it's like information retrieval or something. Luigi La Corte: think it's prospecting, it's ⁓ customer research and it's probably, probably some like generative outbound. Our sales team uses AI quite a bit and they're using a mixture of Apollo and Apollo sequences, which could be AI generated in research. It's probably a few tasks where people are using it. But Andrew, I think you raised this point, maybe it was Mike. Architecture and engineering projecting high AI penetration or theoretically AI coverage construction, obviously like very different. I think they're looking at sub 20%. Mike, you have this in the bar chart that you can pull up instead of this radar chart? Mike Powers: see if I have where to go. I have too many of these things open right now. Luigi La Corte: I you sent it in the group chat. But I will say, so I did an internship at Arup. So Arup's an engineering firm. They're not multidisciplinary. I did a few things there, but this is when I was still an undergraduate in civil engineering. And I was trying to figure out what I wanted to do. So I was doing some structural, some geotechnical and some municipal. I think, okay, as an intern, it's probably not as representative of like what people are currently doing. ⁓ I was really surprised at how formulaic engineering was. Like you literally have a book, either you're doing steel design or doing concrete design. And this is just for those two disciplines. Okay. So in civil, can do a lot of, ⁓ a lot of specializations. If you're doing structural, you're designing steel structures, you're designing concrete structures. And to design these things, like you literally reference a book with rules, effectively a bunch of ifs and thens. So when I see this, then I see, okay, theoretical coverage across engineering is maybe 82%. I honestly believe it. I do believe it. I think by and large for simple cookie cutter buildings, again, I haven't done engineering since 2014 when I was an intern, but if things are still the same, if you're still designing columns and beams and spans the same way, maybe using some software, I think that largely becomes automated. It feels even at the time that if you had a spreadsheet with enough ifs and thens, could make it work. So to me, that number... I know who knows where this number would be plucked it from thin air, but the like where the data where the kind of workflow gravity is an engineering is designing structure at least for it for civil. And to me that feels automated. Mike Powers: So the way that I looked at and thought about this, you look at a, if you look at A &E, so an architect and engineer, do they spend 80 or 82 % of their time on a computer? And if that statement is correct, can LLMs support, accelerate and help that 82 %? I don't think this is saying LLMs are gonna replace it. I know we're talking a little bit about estimating and accuracy in the future. Andrew Zukoski: Hmm. Mike Powers: because I saw and looking at construction is 18%, 20 % of your time as a PM on a job site in a computer and 80 % of your jobs walking around the job site. ⁓ I think that correlation matches. And if that's the economic model they were looking at to build this, ⁓ they're saying, hey, we think anything you do on a computer can be solved with LLMs. ⁓ If so, here's what that TAM looks like. Andrew Zukoski: you Mike Powers: ⁓ You're seeing 80 % saturation for architecture and engineering task already. ⁓ I thought that number was high. ⁓ I don't know many architects and engineers that are, I guess, like it's against how do you define LLMs and usage in AI. People have been theoretically using AI to make machine learning estimates or decisions in Revit for 20 years. So ⁓ I think overall it's interesting from... You know, the people that are listening to this calls perspective or this conversation, it's like, you know, figuring out where in your workflow as a contractor, architect, or engineer, you can leverage LLMs to make your team better, more efficient X, Y, and Z. It's kind of how I thought about this. I think this is good for the industry. This is good, like good holistically, a perspective that there is opportunity for improvement. Luigi La Corte: you seem, I think a little bit more cynical about whether this is going to be like, do you, do you, do you not believe that let's say LLMs are here to stay and they will be more impactful than they are now? Cause I know when we talk about this, seem, uh, to take the opposite side. Like this is a bad business model in general. What, what, what do you, what do you think? Andrew Zukoski: Yeah, I don't know if I'd say it's bad business model. mean, LLMs are here to stay. It's like a radical new ⁓ fundamental capability that's been introduced to computers. ⁓ solves natural language processing. And like, also solves a fair amount of reasoning in a way that people have been trying to do with computers for 60 years. ⁓ So I think they're definitely here to stay. ⁓ And I think they can be viewed in a chain of innovations that have happened to computers. I'm a software engineer by background. The way that we wrote code in 2022, pick a moment in time before AI-generated code got worthwhile, was already radically different from how we wrote code 10 years before that and 20 years before that. ⁓ So I think it's going to be less disruptive to the economy than Anthropic might have us like to believe. ⁓ So I think we'll find it seeps in. tools that have integrated LLMs, most of them aren't going to, I don't think you're going to get rid of them. You can take Claude from my cold, dead hands when it comes to writing code. And it's extremely powerful research tool as well. It's better than Google was before Google got worse like 10 years ago. Now they're going to start more and more of these companies are going to start selling ads like, I don't know, like, you know, these results might get less trustworthy over time. I don't know. I'm just I'm skeptical of the pattern of people have something to sell telling us that it's, you know, the thing that's going to change absolutely everything in the world. Luigi La Corte: That's fair. I mean, the people who are going to go IPO and raise capital are telling you that it's a really good investment and that they're going to disrupt all white-collar jobs. Andrew Zukoski: And it's gotta happen right now. Luigi La Corte: I, uh, it kind of feels believable to me. mean, every week, every week I'm still blown away by how disruptive this technology is. Uh, we hired a marketer recently. Uh, we worked at Autodesk. He joined us, um, never shipped a line of code in his life. Yesterday we launched a webinar, basically trying to show people how to use Claude because it's, it's a great, it's a great tool for, for our customers. And, um, I said, Pratik, I'd like you to publish this pop-up on the website using our repo. He's like, what's a repo? I said, you just have to use Claude, figure out how to ask it to point it to this folder and then ask it to do this thing. And he did it. And he shipped a line of code and it's live on the website. If you go to provision.com, you'll see a modal that pops up asking to join a webinar. And that's done by him, someone who's never coded before in his life. And I think it's still the tip of the iceberg. at how disruptive this is going to be. So if someone who's never coded can pick up and start coding, you know, like I think the TAM of what we're going to be able to make is just going to increase, but also it's going to be really deflationary. I still really believe that SaaS is going to have a really hard time. And this is from someone who sells SaaS, right? Like if your software is not doing a very specific thing, I think it can be vibe coded to a Anyway, I'm covering a lot of topics here. Andrew Zukoski: Thank Luigi La Corte: ⁓ But I'm still kind of blown away by how powerful this technology is. Mike, how do feel? Mike Powers: Yeah, I think there's always like the history doesn't repeated rhymes perspective and like the people like kind of to touch on Andrew's point, people that are most certain that HCI is coming are the ones that profit from it. ⁓ but that doesn't like, doesn't mean they're wrong. ⁓ the internet like did change everything and the.com providers like pet.com promise that everything can be bought online. ⁓ It's it's complicated. Like the technology we're really over the confidence level is inflated based on these like incentives and commitment biases. Like you're going out to raise capital, ⁓ open air raise $110 billion this year, which is four times larger than any IPO initial offering, any IPO ⁓ initial public offering in history. ⁓ They're burning through capital crazy. ⁓ Claude is a little bit, or sorry, Anthropic is better. think Anthropic will burn $3 billion this year on Right now they're at 18 billion in revenue. Yeah. So they're scaling. Andrew Zukoski: Wow, that's less than I thought. Luigi La Corte: What's their burn this year? 30? Mike Powers: three, anthropics projecting three billion in burn. Luigi La Corte: That's it? Crazy. Sorry, is that, ⁓ that's like net burn. Andrew Zukoski: That's remarkable. Luigi La Corte: Yeah, they're almost there, right? They just went from 14 billion to 20 billion in revenue in three months. Mike Powers: ⁓ So it's interesting. It's interesting. And then you're going down the path of what's Apple been doing. The new M5 chip is going to be able to run models on the edge, which means you can run on a local device. You actually don't need the data center built out. So the cost of inference drops drastically. So a lot of things have to align for the valuation of these companies to hit. So when you look at this chart, they're looking at, my guess would be again, they're looking at labor. So say someone that does computer science spends 96 % of its time on a computer. that's X billion dollar industry, we can go capture 96 % of it. Andrew Zukoski: Yeah, okay, so what about on-site work? Mike Powers: That's the other, that's the other 20 % of this model. think there's like for construction specifically, the other 82%, that's not LLMs. That's like theoretically BLMs, robotics. And we're going to talk a little bit about that. Maybe we kind of shifted that conversation and get your thoughts on the robotics side, Andrew. Andrew Zukoski: Yeah. Yeah, mean, Luigi, think you had a stance. What was the stance? In five years, everything on site is going to be... Luigi La Corte: Thank So, okay, this is what I believe that in five years, robotics, mean humanoid robotics will have a meaningful presence on site. I think this for two reasons. One, the dexterity of these models are increasing exponentially. If you go look at what these, sorry, of these robots, so if you can look at what they did three years ago, they kind of rigid and not very capable. Now they're doing Kung Fu and backflips. Okay. And they're picking up eggs. So I think the dexterity and the capability of these models, specifically their hands and their ability to mimic human behavior is increasing exponentially. And I don't think it's very far let's say, okay, being able to pick up a wire, ⁓ the wire and it ⁓ into breaker. Right? Like if you do that, then it could do some subcontractor work. It can't be incredibly capable, but it could do some of it. Andrew Zukoski: you Luigi La Corte: So, I mean, five years is quite a long time. Look at where LLMs have gone within the last few years. And I think what LLMs done, so the second reason is what LLMs have done in robotics has helped with vision. So vision language models combine your vision the identification of ⁓ to ⁓ identify things within the robot's field of view. And also getting scarily good. yeah, my contention within five years, you will see robots on site. And I think there's going to be crazy, crazy opposition to that. Right. It'll be one of like the last bastions of kind of like protected labor will be, well, my body can go on this site and like, I don't want an AI or robotics to go on site. So I see both two things. think that the robots will there, but I don't think humans are going to let it, let it happen, especially unions. Andrew Zukoski: What's the, so has been the adoption curve for other labor saving, mechanical tools site? I looking around trying to find it and it was, DeWalt actually puts out study every about, job site readiness for electrification, right? And this year they're trumpeting that finally everyone agrees that job sites can be electrified and you can use battery powered electrical tools instead of gas or compressed air tools, right? I was like, okay. then other one, self-driving are like... ⁓ Something that's very real now. I was very skeptical of it. Great reason to think that my skeptical stance on all this AI stuff is not to be believed because I was totally fucking wrong about that one. the, like a lot of the technologies that go into that have probably had a faster and larger impact being adopted piecemeal, right? Where like driver assist safety technology has just gotten way better in the last like 10 and 15 years. And it has, I found a me see if I can share my screen on this thing. found this. you know, national highway safety thing, looking at lane keeping assists where they say that like having these lane keeping assistant like tools, like, which is not a full self driving car. It's like one piece of it. Like, I'm sure in the Riverside, ⁓ terrible. Luigi La Corte: Andrew, you're the riverside. class. Andrew Zukoski: where I know how to use technology. Screen, entire screen. There we go. All right, it's a really thrilling visual I wanted to bring up, collection of text in boxes. Key thing here is lane keeping assist equipped vehicle models, 24 % less likely to be involved in fatal road departure crashes and non-equipped models. And that's a bigger impact than I would say is a bigger impact. Luigi La Corte: Hahaha Yeah. Andrew Zukoski: than all the self-driving cars have had economically, like to date, ⁓ even in the scale. it came, this partial deployment of the technology came like much, much faster than full self-driving cars, which are still available only in limited geographies and scaling quickly on one and like pretty pretty slowly on another. And so guess the thing that I'm saying is, I don't know if you're gonna see, I know if you need humanoid robotics, I don't know if that's gonna be the fast one, won't it be adoptive, like other time-saving technology we've where it's adopted piecemeal for specific tasks? Or you literally think it's gonna be that Turner Construction says we've the Turner Labor Division and they can self-reform every scope because they got robots to do it. Luigi La Corte: Yeah, I don't know. I don't know how it's actually going to play out. just know that I'm confident. I have a theory that the technology will be there. I think the human adoption will be a completely different story as we've seen. I think what you've shown there illustrates that perfectly. Irrespective of the technology, it's not like this is going to diffuse instantly. I was dealing with a general And I think the change management is a different beast in and of itself. You can't just plug in the latest and greatest. Obviously you have to work with work with humans to make that happen. But Mike brought up a point a few, a few episodes ago, was executives Turner who left and spun up a new GC and then had all this bonding capacity. What's to people from leaving their current general contractor, subcontractor, and just taking stance of, ⁓ you know, of AI enabled legal firms. I'm going to be robotics enabled. You we're going to compete on labor costs, which is lion's share of costs. I don't think there'll be anything stopping them. And then if you look at, let's say like the adoption curves of technology, let's say since like the refrigerator and the, like the automobile, these things are, these lines are getting steeper. Andrew Zukoski: Are they? Like you've got that graph, pull that up. Yeah, pull that graph up. Are they getting steeper? Like we adopted cars really fucking quickly. Cars are much better than horses. Like we went, we were done with horses like in the lifespan of a horse. Luigi La Corte: Yes, I have a graph. I definitely have a graph here. Mike Powers: You Luigi La Corte: I mean. All right, here's a chart for you. Here's a chart. So over 25, what's that? ⁓ Andrew Zukoski: It doesn't have cars on it. It doesn't have cars on it. So we adopted smartphones at the same pace we adopted radio. That's not going faster. Radio was a century ago. And the internet was slower than radio? Yeah. Luigi La Corte: No. Well, I guess radio had crazy product market fit. guess, I guess radio had crazy product market fit. What was happening before radio? don't know. was a little bored on maybe, but yeah, this does seem a little bit odd. mean, internet was like apparently slower than, than the radio, but smartphones are pretty quick. I mean, I I can believe this, right? Like I know this isn't audited, but ⁓ Andrew Zukoski: This isn't audited. Luigi La Corte: This is from a reputable domain Andrew, okay? WindBuzzer.com Andrew Zukoski: Yeah, yeah, windbuzzard.com, ooh. Luigi La Corte: Yeah, yeah, I like the New York Times font here. Andrew Zukoski: Yeah, this is the paper of record. I'm surprised there isn't a crypto section. Sorry. Luigi La Corte: The ads that I've been hit with are really interesting on this. But I mean, you don't feel this Andrew, like you don't think that we've adopted AI, let's say, I mean, faster than the internet. Andrew Zukoski: Ah, it, oh, I mean we definitely. It's like, it's been adopted faster than the internet, but I believe that it's less different than the things that come before it. Like, if you look at, like we adopted Pokemon Go faster than we adopted smartphones, right? Like the crazy ramp on Pokemon Go is a craze, but it's not as meaningful of an adoption. There's a, like one, it's a lot easier to adopt quickly. because you just have to install an app. Like there's a platform that it lives on. And to, well, I don't know, I'm playing Pokemon Go. There's probably a bunch of people who like met their spouses playing Pokemon Go. So it's like a very, like very impactful thing. But I don't know that the fundamental capability of AI is as meaningful as radio or internet. Luigi La Corte: Really? Andrew Zukoski: Yeah. I don't know. I could be wrong. I could be wrong on this, but like, the, story of your market or learning to code. That's so awesome. It's amazing. It's so cool when people like get their hands in there and start to code. And that story has happened with Stack Overflow and with WordPress and with no code tutorials and with Excel and with tapping someone on your shoulder and asking for some hands. And like, I think AI is actually further lowering like the barrier to entry there. It's like getting an easier and easier on ramp. Luigi La Corte: I was. Andrew Zukoski: But that's a story that could have happened five years ago. Luigi La Corte: I really disagree. think I've just seen enough evidence. I've seen enough evidence to say, well, okay, I will say this. I saw the internet diffuse ⁓ as a child. I was born in 92. So I can't really say that this time is different. ⁓ But I can say the magnitude of this change feels significant. And I'll give you one anecdote. Andrew Zukoski: Good. Luigi La Corte: I was working at a general contractor's office last Friday. It might actually been, it might've been Thursday, but we were catching up on Friday. And I made a post about this. He had two positions open for a project coordinator. But in combination of using provision and Claude, he just said, I don't need to hire that coordinator anymore. But this was, this was a need that he had had one week prior, but within that week using our tools and Claude. Yeah. It seems like this is a lot more deflationary than the internet, a lot more meaningful. If you can go from like, have a job position to it's filled in a week because I've discovered a tool, to me that seems significant. I, Mike, how do you feel about this? I mean, like, I don't know. I think you're a similar age to me, but. Mike Powers: So ⁓ I think going back to the anthropic report where 20 % of contractor construction tasks are done on a computer, I think it's drastically impacting for those 20%. The other 80 % is no impact right now. Things that would need to happen to change my mind would be countries like China or countries in the Middle East that are using robotics to build buildings first. China has led the charge in modular construction, has never really taken off the United States. ⁓ Contractors that have tried to absorb all building risk like companies like Katara have blown up So I think if you're like looking at historical norms, so let me give you an example of hey this this contractor no longer needed a project coordinator That falls into that 20 % bucket. I would say not the other 80 % bucket where the actual building is being built But that story changes on the AEC side. So I'm sorry the AE side the AE side. It could be a lot more disruption from a workflow perspective. And I think there's just like nuances from that. LLMs are changing, but LLMs are just changing to 20%. Is the upside of VLMs slash robotics gonna be able to impact the next 80 %? I think things would need to change. And the first kind of pieces of grass that would pop up would be in other countries, not the United States for a lot of different reasons, from unions to permits to all these different kind of struggles the United States have on. construction where a top-down approach in China can drastically just push out like, hey, we're using robots now to build buildings. Andrew Zukoski: Yeah, we'll able to move much faster. Luigi La Corte: It's there, isn't it already, Mike? Like there was a famous article that came out of the New York Times covering a manufacturing plant in China. I know manufacturing is different than construction. Mike Powers: different. I think that's different. Tesla's 90 % automated in Austin. Factories in China are 100 % automated with robotics. It's doing one bolt in one specific controlled environment. think it's a very big jump that I don't think we have the data to prove that. it's kind of even like, mean, even the robotics, if there was product market pull for robotics, Andrew Zukoski: very different. Mike Powers: the amount of VC dollars that have gone into robotics companies and construction to underwrite the cost, what we think would relate to better adoption in market. You still really don't see the Boston's job. I know there's videos of Suffolk or Yugalpane having a Boston dynamics robot on a site, but what is it actually doing? I think it's an interesting, it's just an interesting lens. Like, yeah, like I think elements are solving problems, but going back to that chart of like, solving the 20 % problems, not the 80 % problems in construction. It's could solve the 86 % or 85 % problems in AE. but it's a different task. and somewhat like the foundational model early creators, like, what's the guy from meta, the AI, the AI from better that just left. ⁓ He's like, ⁓ solve L word problems. Not, yeah, not visual problems. Luigi La Corte: ⁓ Yom Likum. Mike Powers: So it's interesting. Yeah, I don't think it was a good answer and I'd have to see certain things have to play out to make that position change. Luigi La Corte: Okay, well here, take this perspective. I was working with a subcontractor, subcontractor, and we were talking about the biggest problems that they were facing, for them it was labor productivity. Okay, so time spent on site. When I asked him about the breakdown of work on site, he broke down like this. 20 % of the time is coordination, access to lay down, getting on site at the appropriate time. 20 % is set up and take down. 50 % of the time is the actual work. So that 50 % of the labor hours seems like coordination problems. And these coordination problems aren't really a function of human labor. It's a function of good planning. So why not then? Why not? If we use LLMs, why could we not get more productive and impact labor this 80 % that we're saying is untouchable? Because it feels like at least we'll get more productive on the labor side. Mike Powers: So I've been a mechanical subcontractor. data point is, ⁓ we're running duck above head on the floor of this building right now. That 50 % can't be accelerated by all of us in that example. It's you taking a 2i6 piece of duck bolting it into the ceiling, hanging it, running it, moving your scissor lift six feet forward, doing the same thing over and over again. That's the majority of the work from a labor perspective or pipe fitters, metal. Nothing in the LLM world is touching that right now. It could have robotics like a Go Canva who just hangs drywall, wrote Go and actually do that on the mechanical side. That's a robotics There's a layout problem. There's companies like Dusty Robotics that are trying to solve for the layout problem there. It just, there's a lot of pieces that need to be overcome before that happens. that, ⁓ Andrew Zukoski: Doesn't, isn't, but Luigi, you're saying that there's like, of labor time on site, 50 % is putting the duct in overhead on the 13th floor, 20 % is lay down coordination stuff, 20 % is like set up and tear down. And it seems like you're saying better planning could impact that other 40%. And it might also impact, I don't know what the loss rates are for rework, but like some fraction of that duct gets pulled, gets torn out, right? I don't know. Luigi La Corte: Yes. Yeah, it's like 10 to 20 % of the time. Andrew Zukoski: Yeah, so there's like another, there might be another 10 % there that like better planning could accomplish, which is at least more plausibly something you do on a computer, right? Like, yeah. Luigi La Corte: Yes. Like what I'm saying is if we put the robotics aside, LLM will actually help indirectly in that 82 % that is in this anthropocore report saying that ⁓ can't actually touch. That's what I'm saying. But I also do believe that robotics will impact the site within five years. Andrew Zukoski: Yeah. We talked to... Yeah, we talk to our customers, typically general contractors, many of them self-perform some trades. And they say that being it's very difficult to execute, but they're being very disciplined about the flow of information from owner through designer through GCE through trade through installation leads to measurable 5%, 10%, 15 % improvements in their self-perform. capabilities and efficiency on site, which I think is a point for like computer work can meaningfully impact productivity on site. ⁓ Mike Powers: So my pushback on that is the self-perform with general contractors do lower things. Carpentry, ⁓ like laborers, like up trash, laying down mechanical pipe and coordinating with the electrical sprinkler plumbing contractor is complicated. BIM has been around for years and BIM might even... ⁓ Andrew Zukoski: Yeah yeah carpentry concrete. Mike Powers: be the example of making this worse. I don't think even in that, so we break down like, that mechanical subs like, hey, 50 % of the time is actually doing the, actually installing 20 % is coordinating 20 % is take down setup. Like the take down setup is getting the things to the 13th floor. Even if a robot solves that problem, it's still gonna take. Andrew Zukoski: That's an interesting take. That's not going away. Mike Powers: 20 % of the time to have the lift outside or the elevator bring the duct work up. first construction was in Philadelphia while they building the Comcast 2 tower, which is the tallest building in Philadelphia now. employees and ⁓ workers would to wait two to three hours on the ground to get up to the 70th floor of construction. That doesn't get solved by a robot. That doesn't get solved by an LLM. I think a lot of those pieces are what makes construction complex. Yeah, if we're in a future world where there's robots flying around on drones that can go from zero to 74 in two seconds, maybe that increases efficiency. So there's a lot of pieces there that I don't think just like, we have this workflow tool now that helps us do this in the back office that helps us lay out better. I gotta see a lot more data points before I swallow that pill. Luigi La Corte: So basically go to construction, you're safe. Your job's not going to get disrupted. I think that's the takeaway. Yeah, I agree. Mike Powers: Best industry in the world. Why wouldn't you want to go to construction? construction person I've met is always super passionate about what they do. My dad skylines of every city he's worked in, in his office. And when we go to a city, he goes, that's a Kone elevator escalator. And there's not other industries. Like if you go to a JP Morgan banker, they're not like, ⁓ yeah, I helped move a penny. I made a penny off of this transaction by helping my customer buy or sell this stock. 40 years ago. So. Luigi La Corte: You know what? You'd be surprised. Andrew Zukoski: And families stay in it. People encourage their kids to go into it. That's another big marker. My wife's doctor, and she's not going be recommending that career to our kids. Luigi La Corte: Okay, we have like about seven minutes left before we go a little bit over. We have a few topics we can talk about. So Bechtel creates a C-suite AI role. Talk about AI and data strategy. I know we've been talking about this, you know, kind of in and around a bit tangentially. Or we can talk about VLMs. Which one do want to go into guys? Mike Powers: Let's see the AI role, we can talk about the VLMs next week. Luigi La Corte: Good. Okay. Let me pull it up on screen. so this is, think about two weeks old, but Bechtel appointed, a, basically an AI czar, to, and pull out the article here on construction dive, ⁓ Virginia based builder tab, John Platt, 25 veteran to lead the tech transformation of its engineering procurement and construction delivery method. So Bechtel obviously global, very large, ⁓ construction company, ⁓ that's done a lot of work throughout the world. They've appointed a specific name of the role, but it's basically leading AI efforts across the entire organization. So I think my point on this ⁓ in the notes was not, there aren't many companies that are back town. There aren't many companies that have the overhead and the profit to like that, like the aggregate profit to appoint a, a role like this. ⁓ But I personally believe that this is like the right move, depending on what their day to day is. I think this is such disruptive technology that you do need a point person. Like in previous episodes, I've called this a tinkerer, but you need someone to lead the charge, like I think in any technological change. So depending on what they're doing, I think this is a really interesting role. ⁓ think if GCs and subs are looking at this, I personally think you need to designate a tinkerer of your own. Somebody who is tech literate, who has some sway in the company, who has some budget, ⁓ because it's already proliferating. and see how excited people are when they use Claude and they figure out they can automate a work breakdown structure, or they can get somewhat close to a defensible estimate. Now, I actually think the estimates are really bad, but that's a different story. Andrew Zukoski: Yeah, I feel like it scales down with company, right? Like we talk to, when we go and talk to, when we talk to the top 10 firms, right, there's IT organizations, then there's innovation organizations, and then there's like tech groups inside of a bunch of functional groups. So it's like this really scaled infrastructure inside the org. And at smaller firms, there's often like the tech guy, just like the guy is into tech. inside of given team. And everybody, it seems like they get time and budget. It might be fractional. It might be a fraction of their time and budget, but it's fractional and allows them to go out and understand things and carry it back to the rest of the org. One pattern we're seeing, and I'm curious if you guys are seeing this as well, is that... Large firms leaning like like we the data, we control the data, we control how it's used strategy. They have these large well-staffed setting up data lakes and data and data lake houses and data ware lakes and pulling data out of and then building ⁓ their own consumption of it on of it. At least a of them. did not appreciate having one of their AI vendors kneecapped by a public company's API last year, which is sort of interesting to hear. So the big ones are doing it themselves. It feels like there's, and then there's two other groups. One that's doing it is looking to existing vendors to help them understand and adopt this stuff. And then a third that's not really doing anything at a corporate level, although I imagine everyone at the company is running around with Copilot and Cloud and GBT because. everyone else's at every other company. Luigi La Corte: don't think Claude is as, diffused in construction as much as ChatGBT. Like I think people are still catching up to Claude. Some people, some constructors I spoke to, like I said, what's a Claude? What's an anthropic? I don't, I don't even know that is. To me, it's like, it's a staple of my workflow. but I think people are going to learn how amazing it is. And anyway, for the people who haven't heard about it, it's, it's effectively a ChatGBT competitor, effectively. I find... ⁓ the to be better. Now we can argue whether that's true or not, but the chat takes a more agentic approach to answering some questions and it's more free form, meaning I can compare ⁓ vendor quotes and do some bid leveling. can develop a work breakdown structure. I can develop a scope of work. And these are things you can do with an open AI as well, but just like having human labor, you evaluate the... the suitability and the power of the tool based on its output. And I find the outputs just to be more thoughtful. So in any case, I this is where the industry is going to be going. People are going to look to third parties or consultants, hey, I need an AI strategy. I need to work with some AI tools. What are we going to do about that? But Mike, you're with a lot more subcontractors. Where are you seeing them relative to general contractors to the degree that you're aware of what GCs are doing? Are they pretty far along? Are they adopting AI as much as, let's say like on this call are, like where are they? Mike Powers: Yeah, I think leaning into AI in some capacity. A lot of the executives at subs and contractors that I'm talking through are having a thought process of like, are we a tech company or are we a contractor? And would I rather allocate resources into becoming a better contractor than becoming a tech company, having to compete at scale with these tech companies that are getting invested by VC capital. It's interesting and like, I know we're going to chat a little bit about this next week. but like as these become commoditized, say AI takeoffs, AI scope gap analysis, does it actually hurt the contractor more than it helps the contractor? cause if your differentiator was before, Hey, I would our team. ⁓ had the industry knowledge to make decisions effectively, give that like good feel to the owner. And now everyone can do that with AI. I actually pulled up a funny, not a funny, but an interesting quote from Warren Buffett about this from Charlie, poor Charlie Salmanak, he says, and this was around a, they were at a textile mill factory and there was a new loom that would increase efficiency by 20, by 2X. Buffett's response was, hope it doesn't work because if it does, I'm closing down the mill. And why he said that and what you saw in the eighties was all of that manufacturing power went to China, to because it became commoditized. was, hey, I have a machine that does it 200 times better or 2X better. So now just a commodity price and it's a race to zero. As we talked about kind of the beginning of this call, like, You see Turner expanding into other avenues because they don't want that CM fee to be erased to zero. And I think this AI tooling only helps accelerates that. So it's going to be beneficial for contractors to lean in. We've the very ⁓ strong that we're going to continue to rely on all frontier models and never train our own models for that exact reason. We're not model builders. We're not researchers understanding We're going to bet on that. I think My bet for the contractors that win will be betting on the companies that help them better do that. I think, was it BuildOps or Service Titan? The head of marketing was on Bricks and Bytes this week and he said 71 % of his customers, which were mostly subcontractors, are using AI in some capacity. But it's through their tool, not necessarily, hey, I'm going to spin up a cloud code that I have to now maintain or have a team to maintain. And it comes down to hiring as well. Would you rather go work at a mid-tier general contractor as an AI engineer, or would you go rather work at Google as an AI engineer? fascinating. Because if you break that piece down, there's probably only 200 people that are actually impactful in the LLM space. And they're all getting eight-figure contracts from Anthropic OpenAI. And is a GC going to able to compete with that? Or would a GC be able to tap into that model? I always have a funny rebuttal to another. ⁓ construction tech founder that was bragging about the size of their AI team. And my note was, my AI team is the summation of anthropic open AI in Google, and you'll never be able to beat that no matter how much money you spend. So that's how I'm thinking about it. That's how I contractors are thinking about it. it goes back to that model that like, we're living in a bubble where AI is changing the world, but contractors are still building buildings and they see and hear all this noise, but they were also promised and heard all this noise from ⁓ Andrew Zukoski: you Mike Powers: know, procor or plan grid, making them 100 % more efficient. Andrew Zukoski: note you gentlemen ⁓ Luigi La Corte: Yeah, good chat. Thanks everyone for joining in. We'll talk again next week. Thanks everyone. Mike Powers: Hahaha Talk soon.