Latest / Real Estate Meets AI / A Conversation with Dr. Sarah Buchner from Trunk Tools
Transcript
- Sandeep: One high rise in Manhattan, 120,000 files, 30 pages each, three and a half million pages, a stack three times taller than the building. Somewhere in there are spec conflicts, joint contradictions that blow up the schedule, and no human team that will ever read it all. My guest today built a brain that does. She was a carpenter at 12 on a farm in Austria and ran high rises for one of Europe's biggest contractors, got a PhD in construction data. Went to Stanford and now is running one of the fastest growing AI companies in the built world. Dr. Sarah Buckner of Trunk Tools. Let's go. Welcome. Welcome to the pod. Dr. Sarah Buchner: Thank you. I love that you start with the wildest stats. That's good. Sandeep: well, I am curious. Obviously, you know, there's there's a lot of things that you've done. Forbes 30 under 30, fellow. that's awesome. But I want to start at the farm. You've said you grew up poor, your dad was a carpenter, and working the farm job site was how you put food on the table. What did a 12-year-old learn on the job site that a Stanford NBA cannot teach? Dr. Sarah Buchner: A lot. That's I love that question. Now it's funny because I felt so misplaced at Stanford. I was like, I mean, was super excited, obviously, and honored to get in there. And I show up there and everybody's from like consulting or private equity and finance and like everybody's super fancy and flies private jets. And I yeah, I was pretty lost and I felt that I didn't belong. And now years later I'm like, actually that probably helped me more in my career than than most of the others. And so it's full circle this question. So I appreciate it. I think life skills is something that people are underestimating. And a lot of knowledge, at least over the last whatever few decades, was very much targeted around like white color knowledge. And I actually feel like a lot of the blue colour knowledge that I have helps me in my day-to-day at least equally, if not more, than a lot of my white colour knowledge. And the more AI we have, the more the blue colour knowledge is gonna become more and more important. So I always say, if AI is gonna eat everything, I'm gonna go back, build houses, be a carpenter. Okay. Sandeep: like that the way that you describe it is how truly every part of your knowledge, it's it's all coming together in what you're building. And and I love that. so spent eight years blue collar, became superintendent, then went to the GC side. And there was a moment on the job in Berlin that changed your trajectory. Is there something that you want to talk about that? What was that moment? How did you Decide to do what you do. Dr. Sarah Buchner: Yeah. it's not a great moment. I I had a fatality on on one of my job sites and that is something that I don't wish upon st my biggest enemies. But it is it is big unfortunately in construction still these days. Like there is f suicide, there's drug and alcohol problems, but there's also just like fatalities or like accidents based on safety and I had one of them and it shook me hard because there's police involved and all sorts of stuff and then I was just reflecting and I was like, what do I want to do with my career? Do I want to really stay in a job where I'm managing people who are put on risk of dying or do I want to change something? And so I decided to build a health and safety app. This was back in Europe. And this was also back in the days when apps were cool. how things have changed. and that kind of taught me that like, hey, I'll have more impact with technology in our industry, more positive impact. then just by managing job sites. that's why I ended up doing the PhD on what can we do with AI and data in construction. Sandeep: Yeah. Yeah. I love that. There's few things that you said in there that really resonate. One about, you know, not accepting the norm. That's definitely close to my journey as well. You know, watching the architecture world around me and being like, Yep, that that doesn't work. Let's redefine it and let's kind of see how we can improve that. But the other thing that gets me really excited to learn even more about your journey is really everything that you've built since is about making the job side safer and saner for people. And it's it's just incredible. And one thing that I think most people may not know who listened to the pod is that trunk tools didn't start as a AI company. Y'all first, as you said, were the trunk performed and then you killed that app. that Dr. Sarah Buchner: Yeah, w which was AI as well. So we we definitely had AI from day one, but with a different target, I would say, which and it was pre L L So it was very early L LMs. Yeah. Yeah. Sandeep: But you walked away from it is what I was getting at, which takes a lot of courage. I've done that before. So Dr. Sarah Buchner: Mm-hmm. Sandeep: what did it take? And what kind of foresight and risk taking, or how did you get to the point where you were like, I see a bigger wave? And, you know, we just gotta take the leap and go on and build something that we think can have even wider impact, clearly. Dr. Sarah Buchner: Yeah, it was a full circuit moment as well. I so did this PhD and it was paid by one of the largest channel contractors in Europe and their C Sweet gave me kinda like the topic, which were like, Hey, just figure out what we can do with our data. That was like, you know, we have collected data for years or decades, like just figure out now what we can do with it. And I went off and did my research and, you know, everything and came back years later. Was like, Honestly, you can't do much with your data. Because it's super unstructured, it's siloed, we have to create all these structure around certain processes around how to get structured data and so on. And fast forward a few years later, I I started trunk tools again as a tool to create better alignment from the field to the back office, which is something we still do today, but with very a different angle. And I had AI engineers from day one, and LLM started becoming more popular. This was before even ChatGPT 3 came out. and I was like looking at the technology and I was like, what else can we do with this? We were trying to solve a tiny problem in trunk perform with LLMs. And I was like, wait, this seems like it can also solve some of this unstructured data problem that I have experienced in my PhD. And I sent off my favorite AI engineer, was like, figure out what else we can do with this, with the amounts of data that we have and all of this. And came back, had like whatever, 18 different MVPs. And we put the AI engineers in a room and a bunch of superintendents, and a bunch of beer and wine. And we sat there for hours and went through all of these 18 different versions of what we could do with LLMs. And Two or three of them really hit home. And then we were like, okay, cool, let's let's explore that. And then one of them was a trunk text, which became our largest product back in the days. not anymore, interestingly enough, but that that was like a real hit where it was literally just you can text a phone number, a question around your construction side, and it's gonna respond back with an answer. Like so simple. And and that's construction, right? Like the more simple you can go, the better it will hit. And in the background, obviously it was super complicated, trying to get the right data, structuring the data, pre-processing all of the stuff. Sandeep: Yeah. Dr. Sarah Buchner: yeah, so that's that's how this happened. And like I remember just raised the series seed. I think I just raised $10 million and had my first board meeting. And they invested on trunk perform. And I was in a boardroom and telling about everything. And then in the last 10 minutes, I was like, hey, I want to show you something new that we built. And showed them this text functionality with the rag product in the background. We didn't even know that we built rag because it wasn't a term yet that we really understood. I think the term wasn't Sandeep: Yeah. Dr. Sarah Buchner: fully there at that time. And so we built RAG for construction and my main board investor was like, play it again, play it again. And he he was like, I'm so excited to have to leave the room. And I was like, Hey, I'm gonna take one of the $10 million and throw it against a new thing. The other nine we're gonna keep on on the thing you actually invested in. And thankfully the board was just very easy and they were like, Okay, cool, we trust you and then very quickly we spent nine million on the new one and about one on the old product and and here we are. Sandeep: That's that's what that's what comes out of great partnerships is definitely something that I hear over and over from fellow founders, but also my board as well, right? that's if you have a great partner, there's a lot of really cool things one can do. And that's exactly what it sounds like you've learned from your story. Okay, so here's like the the thing that we do for every podcast. It's the AI meter, it's a scale of one to five. one is the industry is AI averse. Five is that we have fully adopted it, we are fully a Gentec AI is everywhere. Dr. Sarah Buchner: It's funny. We have a scale from one to seven, so yes. Sandeep: So where where are we at? A one two five, where where do you think we are at? Dr. Sarah Buchner: The indust the A C industry Sandeep: Yes. Dr. Sarah Buchner: at one. Sandeep: And what and why? Dr. Sarah Buchner: maybe I don't know if you give me periods, maybe one point two or something like that. But No, it's Sandeep: All right, one point two. The point two because I asked why. Well either way, but why? Dr. Sarah Buchner: if it's black and white, it's definitely not a two. but it's Sandeep: Sure. Dr. Sarah Buchner: it's not fully a one anymore. why it is not further? Sandeep: Yeah, add some color commentary so folks can understand where you're coming from. Dr. Sarah Buchner: Yeah. so we are as a company at a five, I'm as a person at a five. so once you have touched the five, you realize what could be possible. And once you see this business models ha w would have to change, roles are changing, like whole companies would be turned upside down. And so I haven't seen a lot of that in our industry. it's interesting, channel contract is actually further in my experience right now, general contract is a further than other participants in in the industry, definitely further than owners, way further than owners' reps. Like that's that's shocking where they're at architects and engineers are also further behind So, but when I'm looking at the most forward part, which is channel contract is right now, and I'm looking at the best ones, then I would say those are at like a three or four. But this is such a small piece of them. And it but when you see how they operate these days, you're like, wow, this is incredible. They have a 10x output per person. It's not about laying people off. That's what people don't understand. Like AI is not about firing people, especially not in our industries where there's such a labor shortage. It's about more revenue per head. And I'm seeing it with some of our customers and obviously in our own company and it's incredible, but what it takes to get there is brutal change management. And people are in general not people don't like change. That's human nature. And what is needed right now is a lot of personal changes and a lot of corporate changes. And that needs really good leadership and pretty aggressive leadership. And that's why a lot of it hasn't happened yet. That's my thesis. Sandeep: for the most part, I'd say I agree. That's definitely what we have continued to see across the board as well. But I will say that there is a little bit more light that we've seen, you know, and a little bit more hope in in other practices, where I will say that there's at least folks trying. I will say that I am finding less and less, because you know, again, we talk to hundreds and thousands of. owners and developers, other architects on a yearly basis. The number of folks that are averse is reducing. That's not to say that, my gosh, good job, you're less averse. The industry is moving with or without them. But I will say that it's trending in the right direction, but probably far too slow. Dr. Sarah Buchner: Yeah, we have the same thing. We have seven stages of AI adoption. If you are still piloting and you are five stages, you're still in a one. that that doesn't Sandeep: Yeah. Dr. Sarah Buchner: count. And if you've rolled out Microsoft Copilot, it doesn't count either. Because rolling something out without enforcing adoption or training people to actually know how to adopt AI, you haven't done anything. You have checked the box. And I see a lot of checking the box and it's so sad again that as an industry we're falling behind with yet another transition in comparison to law or accounting or whatever. there's finance. you see other industries you're like, they are at a three or at a four. And we yet again are stuck in a one without really needing to be there. Sandeep: That's true. I will say the excitement that I also feel from this is because so many of the folks that are, mid to large size in the industry that are stuck on that one, which they surely are. You are absolutely correct. There's room for other players to grow. I mean, now is the time where the mid-sized person Dr. Sarah Buchner: Mm. Okay. Sandeep: can become the large person. This is the Dr. Sarah Buchner: Yeah. Sandeep: time where, you know, we can truly reshuffle the the deck and to me that's exciting about the the world that we're living in. Dr. Sarah Buchner: incredibly. I gave a I gave a speech yesterday to a room full of fifty mid market channel contractors and subcontractors and they're like in the whatever hundred to five hundred million revenue bracket they are so much faster and more agile and like chumping in this and therefore they can get rid of the backlog faster. They can grow way faster. Like, yeah, you can double your revenue all of a sudden per year without increasing headcount. And like if you can do that, there is so much more demand right now for AEC than there is supply. So yeah, as you said, like pushing yourself again in construction, getting into the ENR four hundred, I don't think has ever been easier. Sandeep: Yep. okay, so kind of switching gears a little bit, let's get muddy boots, let's let's get into it. when a superintendent on one of your sites has a question, what happens? what are you actually enabling? what is one or two or three great concrete examples of where AI is actually helping? Dr. Sarah Buchner: So i let's take what you just said. Like a superintendent or somebody in the field or even in the back office has a question about a construction site. How they're interacting with our system doesn't matter to us. And I think that's a new wave of AI tools that really figure this out. So they can text us, they can Microsoft team chat us, they can email us, they can call us, literally phone call us, they can go on the web app, however they want to interact with the system doesn't matter to us. It's all going into the the same what we call cortex. What we do when we start with a new construction project, we are starting to integrate into all these different systems of record, right? Like Autodesk ProCore SharePoint box, whatever it is. And then we take the data and we train one brain for each and every project. by the way, this works in pre-construction, works in design, whatever it is. I'm just gonna use the construction workflow. So we train something that we call Cortex, and that's a lot of pre-processing for the AI people. That's a lot of you know specialized models, especially because the multimodality of construction is like drawings and and tables and schedules. And so this brain then knows everything about this project. And then this brain is being leveraged by different agents through an agent harness and orchestration, all of that. And meaning these agents are the smartest people on your job site. Because as you said in the beginning, nobody can remember three and a half million pages of documentation, but AI can. And AI is really good at bringing dots together. So when we do pre-processing, we go through an ontology into a knowledge graph, which is the same neural network approach that the human brain has. So you have this super powerful brain. How the people interact with it doesn't matter. But if a superintendent, for example, calls trunk text and be like, Hey, what is the color in the lobby? This brain within a few seconds is going to tell you it's Benjamin Moore, Matt White. Or if the executive is like, Hey, what are the biggest risks right now in this construction side to fall behind schedule? It's going to tell you here's the top three risks and here's how you should avoid them. So that's pretty basic QA. The way more interesting part for me are automatic agents that are doing the work without any human involvement, meaning that we have trained agents on this construction workflows that know exactly when to jump in. One that's more relevant for this audience is where we have a almost complete understanding of drawings, 2D drawings. So for example, before you send a new set of drawings to the channel contractor, we have the AI agent run on the architecture side and finds problems. But that's two-day flash detection, trade to trade comparison. Or was it just version to version comparison and be like, here's three things you forgot to actually like mention? Like if you sent this to the channel contractor, here's the financial impact of this. Right. So we can calculate before they make a decision how much the decision is gonna cost, how it's gonna change schedule and so on. So you see what's happening there, and usually we do this in a way that before something goes back to the channel contractor, the agent grabs it. runs and goes back to the architect and says, this is gonna be a $200,000 decision. Are you sure? And the architect says, yes, cool. It goes back to the channel contractor. So we have dozens of agents that are really taking on human level work. And that's also how I'm seeing the future of all of this is agents that automatically work based on new files, based on new data points, doing human level work and pinging the human again to then take a decision or move it further. Sandeep: I like how you describe that. Dr. Sarah Buchner: My favorite agents are the ones that directly work with other agents, right? we have a drawing agent that finds problems, hands it to the RFI agents, writes an RFI, waits for a response, calculates the impact, rides a change event, writes a change order, right? So that's when you see agents really we always say they dance with each other. Once the agents start dancing, then you have a real like transformation. Sandeep: Yeah, I I love how you describe that. Cause right now, just to put it in context, for example, we're doing all of that just from an architecture's standpoint, right? Because we're a services firm. So we when we get RFIs from human, contractors, our agents will respond to it. They'll pre-write it and then the architect who obviously is the license stamp behind it, they'll be like, Yep, that looks correct. ship ship it off. And it sounds like you're just doing that. across the entire, for example, construction in this case, example. That's I I love that. Dr. Sarah Buchner: Yeah, it's funny sometimes we have agents work work against agents. we had we had a job recently where the channel contract that was using our RFI agents to write the RFI and then the architect was using our RFI agent to respond to the RFI and so it was just like funny back and forth based on the same brain. Sandeep: But it's a good log too, right? it's logged. So, if there's ever any conflict in the future, that's the thing that we enjoy the most about it is the having logs and making sure that where the decision was made is super clear. It's not in some email sitting somewhere. Or if it is sitting in the email, it's still connected to to the brain, which is amazing. That's awesome. I love those examples. They're very, very clear. They're very, very actionable. And it sounds like agents are talking to each other. I imagine what the future will be as the agents continue to get smarter. How much do you think will folks continue to delegate? there there may be a world where perhaps humans are less and less in the loop? Where do you think that that makes sense? if one was to try that in in the world, you know, that that you're in, where would be the safest place to have fewer humans in the loop? Dr. Sarah Buchner: I think the the tick is actually way further than the industry's leisure trust is right now. So I give you an example. We have a submittal agent that reviews submittals automatically and finds open RFIs and the spec sections and so on. And one piece of this is like that you need to find the right spec section for that part of the submittal, right? And in some systems of records, humans make that connection, right? It's like this product comes to the spec section, whatever it is. We have run as a middle agent hundreds of thousands of times. this is a real number, if not millions at this point, I haven't checked recently. And after about, I don't know, a hundred thousand runs, our confidence level to find the right spec sections was so high that we decided to overwrite the human spec section connection that was just wrong ten percent of the time. that's just a tiny, very detailed example, but We are not releasing anything under 85% accuracy, which, people can say that's not great. I'm like, yes, but then, it's in the market and then very quickly it's gonna get to 95, 97% accuracy. And then people are still like, it's not perfect. I was like, how perfect are your humans? And that's the question when it depends who you ask, but we did an anonymous questionnaire into the construction industry. It was like, how often do you think your humans make the right decision? The average response was 65%. So this whole criticism around it needs to be perfect and is it accurate and it's it's it's just it's the wrong version of thinking because do you have rather a hundred micro decisions done in microseconds where five of them are wrong, or do you have the a hundred micro decisions done in at least an hour or ten hours where Probably at least a few that are still wrong, but in their mind none of them are wrong. And so there's there's a always a a correlation between efficiency and perfection that humans need to get over. And and it's it's time for this to happen. Yeah. Sandeep: it's inevitable. And I'm so so excited that you're on the leading end for it. So now that I know a little bit about what you're working on, I know a little bit about, how you got here, I'm curious to learn a little bit about you personally. okay, coffee or tea. Neither is fine too. If I don't know if it's just life juice. Dr. Sarah Buchner: There's this this is soup. Love soup. This Sandeep: Is that truly s Dr. Sarah Buchner: is soup, spring bone. I love that fries. This is like soup out of a cup. if so decaf, I don't do caffeine. Sandeep: Okay. Are you a morning bird or night owl? Okay. Dr. Sarah Buchner: Night. Sandeep: hammer or keyboard, which one is more satisfying? Dr. Sarah Buchner: Hammer, hundred percent. Especially the bigger the better. Sandeep: Austria or New York, better bread. Dr. Sarah Buchner: Better bread? Austria. No, there's there is no this is not even a subjective question. around the corner they open an Austrian bakery and they're charging sixty dollar per bread and Austria is furious that like they can charge. Sandeep: Is that bakery making a lot of money from you? Dr. Sarah Buchner: For me, no, I don't like carbs. Sandeep: that's funny. So what was the first thing that you stopped doing when you really got into AI and started adopting as as part of your your life, your business, what have you? Dr. Sarah Buchner: I don't know about the first one, but I think the most impactful one probably for this audience was I have my AI read all emails, decide which ones are important, and automatically draft responses to them. And it obviously has my co-recordings, my calendar, my notion, my Slack, my everything else. that was the most impactful one for me personally, because my EA did this before, but now it's just freed so much of the time. And I think the coolest one of people are not doing is we actually treat AI as a full executive. So we have an AI run in all executive meetings, in board meetings and stuff like that. And then we humans take a break so that the AI can do processing. And when we come back from a five minute break, the AI tells us like where was our logic flawed? Where did we lose a point? And like where what should we get back to? And honestly it's like having the most sophisticated McKinsey partner in every meeting, for free. It's great. Sandeep: Love that. Okay. Other than your own, what is the AI technology that you use every day? Dr. Sarah Buchner: All of them. No, not all of them. I'm obviously big on Claude Grogbot right now is really taking over. Sandeep: Love it. All right, Sarah, last question for you. you graded the industry at one point two as we kicked it off. But clearly, you know, folks are still navigating this change. What is one piece of advice? What should they be doing to be able to be as comfortable as yourself as they navigate this change? Dr. Sarah Buchner: Yeah. It's it's interesting. we offer full AI transformation, right? Where we come in and transform the whole companies. And when we do that, the first thing we do is we do a one-on-one training session with the CEO. And the reason for that is because I'm telling you, 90% of the CEOs we are working with say they want to do AI, but they're doing nothing themselves. And AI transformation is not one of these things that you can delegate, because it's a shift in mindset way more than it's a shift in just like picking a new tech. And that shift in mindset needs to happen and it needs to start with the C suite. And the most successful companies that are like now in a three or in a four, those are the ones where the CEO and the CEO themselves got their hands dirty. And nobody needs to be a coder, but like just i if you still use AI like a Google search, you're not you're not doing AI. We teach every CEO how to write their first own skill in like cloud, copilot, whatever they use. And once they've understood that, then it's really a like from there it's like it's smooth sailing. Sandeep: I love Doctor Sarah. Dr. Sarah Buchner: And my favorite thing around AI that nobody like that that people should be doing, ask your AI what your spirit animal is. It's like it's really, really fun. I'm a honey badger. It's really ugly, Sandeep: What is yours? Okay. Okay. Love that. Dr. Sarah Buchner: but that I did but it's aggressive. Sandeep: Sarah, I love this conversation. Thank you for keeping it real. Thank you for taking the time to join us today. Dr. Sarah Buchner: Thank you for having me.