رونوشت کامل Ep. 61 - What Palantir Actually Sells, and Why It's Not Another App

back to designers and ask them to at least highlight or give me a 00:03:53.439 --> 00:03:54.499 cover letter on the changes. 00:03:54.819 --> 00:03:58.219 And that was always like pulling like teeth. But Trunk Tools is now automating 00:03:58.219 --> 00:04:02.226 that. So you get a bulletin come in, it automatically highlights the changes. 00:04:02.621 --> 00:04:03.445 You don't have to do a thing. 00:04:03.945 --> 00:04:07.179 Well, it's also hitting RFIs, right? It's also hitting submittals. 00:04:07.679 --> 00:04:11.775 And when we think about standardizing ops, right, and providing that, 00:04:12.286 --> 00:04:18.379 you know, sort of resource to those field operators, what I see it as, you know, 00:04:19.048 --> 00:04:23.699 it's sort of like this like mentor within the company that structure, 00:04:23.699 --> 00:04:27.599 standardizes, keeps it moving in the right direction, looks out for risk. 00:04:28.439 --> 00:04:33.497 And just this last week, you know, I had a QC manager just pass along the submittal to an owner. 00:04:34.122 --> 00:04:38.179 Without doing a review because he was too busy and it was wrong and and you 00:04:38.179 --> 00:04:42.284 know the classic owner statement is did you even review this right, 00:04:42.859 --> 00:04:48.289 so this to me is carte blanche we absolutely did and think about getting that 00:04:48.289 --> 00:04:50.619 first pass before your your, 00:04:51.099 --> 00:04:55.319 your your your field manager ever even actually has to do the first pass someone's 00:04:55.319 --> 00:05:00.159 doing it for him so to me it's about speed it increases accuracy and help standardize 00:05:00.159 --> 00:05:01.411 ops out in the field, right? 00:05:01.997 --> 00:05:05.209 Yeah. And this is the kind of AI that we've talked about on the podcast, 00:05:05.209 --> 00:05:09.139 really appreciating, right? Which is looking at ways to elevate the craftsman 00:05:09.499 --> 00:05:10.679 at the heart of project delivery. 00:05:11.019 --> 00:05:16.341 Give me time back so I can be out in the field managing risk, right? Yeah, 100%. 00:05:16.754 --> 00:05:21.927 So, all right. If you're a little curious, check it out, trunktools.com forward slash podcast. 00:05:22.519 --> 00:05:25.230 And if you mentioned construction conversations, you get a discount. 00:05:25.706 --> 00:05:29.972 Discount. Demos are free, and you guys get a discount, so let's go. 00:05:30.618 --> 00:05:34.388 So the title is Forward Deployed Engineer at Palantir, right? 00:05:34.912 --> 00:05:37.463 Which also, like, we could have a whole podcast just about that title, 00:05:37.463 --> 00:05:39.503 because I think, you know, there's a lot of interest 00:05:39.823 --> 00:05:43.533 right now where you have all these different things that can be done with these 00:05:43.533 --> 00:05:48.323 models, but you don't always have maybe the organizational domain context needed 00:05:48.663 --> 00:05:50.043 to embed them effectively. 00:05:50.043 --> 00:05:55.623 And so the FDE in itself is a really interesting kind of growing job title. 00:05:55.623 --> 00:06:03.003 But so forward deployed engineer at Palantir and maybe what is the first thing 00:06:03.207 --> 00:06:05.656 as you start to dip your toes into construction? 00:06:06.266 --> 00:06:09.017 What kind of what kind of stupid questions are you asking to begin with? 00:06:09.557 --> 00:06:13.253 Yeah. So when I when I started with with Kavanaugh, I came in, 00:06:13.253 --> 00:06:15.446 you know, having never been on a construction job site. 00:06:15.897 --> 00:06:18.843 And so I kind of opened up to Joe and other people and said, 00:06:18.843 --> 00:06:22.413 you know, I'm going to ask a lot of questions that seem stupid, 00:06:22.588 --> 00:06:24.453 but they're also from an outside perspective. 00:06:24.453 --> 00:06:28.613 And so part of the value here is that I might challenge some assumptions that 00:06:28.613 --> 00:06:31.793 you have that have been had in your field for 40 years. I could be wrong 90% 00:06:31.793 --> 00:06:36.508 of the time, but if you walk me through it, we'll learn together and understand where those edges are. 00:06:36.932 --> 00:06:41.487 And so we kind of started in that vein of just sitting down and be saying like, 00:06:41.632 --> 00:06:43.095 okay, I think cost codes are dumb. 00:06:43.339 --> 00:06:45.813 And they were like, what do you mean cost codes are dumb? I'm like, 00:06:45.813 --> 00:06:49.873 I don't understand why somebody who's telling me that they're paving on a job 00:06:49.873 --> 00:06:51.133 and their job title is paver. 00:06:51.330 --> 00:06:55.364 Why do they also have to know this 12-digit number? That doesn't make sense to me. 00:06:55.713 --> 00:06:57.823 And so we kind of had this interaction where I'm like, okay, 00:06:57.823 --> 00:07:02.703 so the utility of a cost code is that you can map costs, but why does it have 00:07:02.703 --> 00:07:04.856 to be a computer-readable number? 00:07:05.054 --> 00:07:09.533 Okay, because all of this was designed in the 1980s to work with like Fortran, 00:07:09.533 --> 00:07:12.923 where it has a 12-character limit, and you've just kind of inherited that. 00:07:12.923 --> 00:07:16.832 It's kind of like that example of like, why are railroads the size they are? 00:07:16.989 --> 00:07:20.043 Well, it's because that's about the size of two Roman horses side by side. 00:07:20.403 --> 00:07:24.332 Follow the causality, you kind of end up with that locked in place. 00:07:25.290 --> 00:07:29.035 I mean, it's such an interesting example of the power of an outside perspective. 00:07:29.035 --> 00:07:33.035 I think, you know, whenever I'm hiring a new person, I really challenge them 00:07:33.035 --> 00:07:36.245 to ask a lot of questions and don't assume anything makes sense. 00:07:36.565 --> 00:07:40.648 And I think taking that approach with a bit of humility, like you did, 00:07:41.292 --> 00:07:45.385 yeah, can output some interesting conversations. So you start kind

of mapping 00:07:45.385 --> 00:07:48.024 and maybe before we kind of go into that whole endeavor. 00:07:48.576 --> 00:07:51.705 Sponsor break. If you spend any time running a project, you know, 00:07:51.705 --> 00:07:54.863 the answer is almost always sitting in the contract document somewhere. 00:07:55.414 --> 00:08:01.015 The catch is that it's almost never in one document. It's split across the contract, 00:08:01.015 --> 00:08:04.615 the specs, and some flow-down clause three layers deep that nobody connects 00:08:04.615 --> 00:08:06.403 until it's already cost money. 00:08:06.606 --> 00:08:10.475 That's the whole premise behind Document Crunch. That risk doesn't live inside 00:08:10.475 --> 00:08:13.177 any single document. It lives in the gaps between them. 00:08:13.404 --> 00:08:17.920 And most teams never look at all of them together until they're standing in a dispute. 00:08:18.616 --> 00:08:23.183 The average construction dispute in the U.S. now runs about $60 million. 00:08:23.846 --> 00:08:26.238 According to Arcadis. It's a lot of money. 00:08:26.638 --> 00:08:30.613 And the root cause they point to isn't fraud or bad intent. It's people not 00:08:30.613 --> 00:08:32.228 fully understanding what they agreed to. 00:08:32.443 --> 00:08:35.322 So Document Crunch builds a platform around closing that gap. 00:08:35.520 --> 00:08:41.138 Crunch AI reads the documents, surfaces the risks, and it cites the exact page and clause that's big. 00:08:41.486 --> 00:08:45.323 So you're not just taking its word for it. Project Assist lets your team act 00:08:45.323 --> 00:08:48.573 on it, ask questions across the entire project document set, 00:08:48.573 --> 00:08:52.945 hand draft your submitals, notices, and RFIs right from the chat window, 00:08:53.160 --> 00:08:56.193 and the whole thing runs on your own company standards. 00:08:56.193 --> 00:09:00.933 The office and the field are working off the same playbook for bid through closeout. 00:09:01.240 --> 00:09:06.223 No disconnects. It's live now across the U.S. and Canada. Go to documentcrunch.com 00:09:06.223 --> 00:09:09.594 forward slash construction conversations and book a demo. 00:09:09.890 --> 00:09:13.843 Like, let's maybe just talk about, like, what is Palantir and what is the foundry? 00:09:13.843 --> 00:09:16.647 Starting with Palantir, what is Palantir? 00:09:17.110 --> 00:09:21.964 Yeah, so Palantir is a company that's been around for about 25 years now. 00:09:22.497 --> 00:09:26.423 It was created in the wake of 9-11. There was a pretty bold statement by our 00:09:26.423 --> 00:09:29.563 founders that they thought the attacks would have been preventable because 00:09:29.883 --> 00:09:34.563 the data existed, but it existed in siloed systems that were siloed both by 00:09:34.563 --> 00:09:37.443 kind of technological limitations, but also because, you know, 00:09:37.443 --> 00:09:41.867 you have the CIA and the NSA, FBI, they have different allowances of who can see what. 00:09:42.058

--> 00:09:46.583 So how do you kind of create a cohesive data model that brings that together 00:09:46.583 --> 00:09:50.899 while still respecting the privacy and, you know, the kind of access restrictions? 00:09:51.603 --> 00:09:57.268 From there, for the first kind of decade or so, our main customer was the US government. 00:09:57.453 --> 00:10:01.023 And then about 12, 15 years ago, I'm not exactly sure on the timeline, 00:10:01.308 --> 00:10:04.513 the realization was had that the same problems exist in the commercial space. 00:10:04.513 --> 00:10:08.363 So we got started with a few big companies, PG&E back in the day, BP. 00:10:09.291 --> 00:10:13.226 And really, it was a start of saying, how do you connect all of your 00:10:13.626 --> 00:10:17.126 different data sources, not in terms of making a dashboard, but in terms of 00:10:17.126 --> 00:10:21.116 making an operational layer that lets you make decisions based on that data 00:10:21.116 --> 00:10:25.718 and push those consequences out, either back to those systems or to the people who need to know. 00:10:25.904 --> 00:10:29.985 So it becomes a platform of action as opposed to just a data aggregation platform. 00:10:30.287 --> 00:10:34.086 So that's broadly speaking what Palantir is. I mean, we could talk about that 00:10:34.086 --> 00:10:36.021 and there have been people who talk about that for hours. 00:10:36.294 --> 00:10:40.772 But to get into what Foundry is, Foundry is our main commercial software offering. 00:10:40.952 --> 00:10:45.306 It's alternatively called AIP ever since LLMs came to the scene and we were 00:10:45.306 --> 00:10:47.724 kind of building agentic tools on top of that. 00:10:48.177 --> 00:10:53.222 I like to think of it as a collection of Lego blocks that are integrable by design. 00:10:53.553 --> 00:10:56.846 Anyone who's ever played with Legos knows that, you know, maybe this model is 00:10:56.846 --> 00:10:59.796 a spaceship, this one's a car, but you can mix and match them. 00:10:59.796 --> 00:11:02.761 Like by definition, any two Lego blocks can connect. 00:11:03.046 --> 00:11:06.736 And so by giving this powerful set of tools, as opposed to saying, 00:11:06.736 --> 00:11:10.424 like, here's how we're going to solve your problem, here's the tool that we're building for this 00:11:10.633 --> 00:11:14.206 abstract problem across your industry, it allows people the flexibility to build 00:11:14.206 --> 00:11:18.016 exactly what they need, as opposed to having to kind of conform themselves to 00:11:18.016 --> 00:11:19.282 what exists in the market, which has 00:11:19.532 --> 00:11:22.504 been very true in construction from the people I've talked to. 00:11:23.636 --> 00:11:30.671 So what I hear is kind of like modularized software with respect to data. 00:11:30.782 --> 00:11:33.206 And you haven't said ontology yet, which is impressive. 00:11:33.926 --> 00:11:37.746 I know, I know. I try to get to that point of the thing. That's one that people, 00:11:38.086 --> 00:11:42.366 I think, can bounce off of because it's a jargonistic term. But the ontology 00:11:42.366 --> 00:11:45.706 really is one of the secret sauces. I'd say the ontology and the forward deployed 00:11:45.706 --> 00:11:48.229 model are kind of two pieces that have become, 00:11:48.940 --> 00:11:52.585 adopted by other people as well other companies but the ontology, 00:11:53.108 --> 00:11:56.683 the simple way to put it is that it's the mapping of the 00:11:57.043 --> 00:12:03.273 business valuable concepts to the data with a system of action and accessibility 00:12:03.273 --> 00:12:06.531 access restrictions on top of that those are the three legs that i think of 00:12:06.862 --> 00:12:10.367 you can say the nouns verbs and and kind of access structure 00:12:10.721 --> 00:12:13.523 but what it is is a way of encoding not just data 00:12:13.923 --> 00:12:17.183 that can exist in a system but the tacit business knowledge that goes around 00:12:17.183 --> 00:12:18.744 that But in construction, you know, 00:12:19.163 --> 00:12:23.033 you can't just ask an estimator to say, well, tell me how you costed that. 00:12:23.033 --> 00:12:26.006 They're going to tell you, I've done it for 30 years. I know that's what it costs. 00:12:26.232 --> 00:12:30.393 I don't, I maybe don't even have like accessibility into my thought process necessarily. 00:12:30.423 --> 00:12:34.413 Maybe if I sat down for a while, I could tell you, but how do you capture that 00:12:34.413 --> 00:12:38.633 value and encode it into a system beyond where it would exist in a database? 00:12:38.633 --> 00:12:42.123 So it's a lot of different pieces there. And if you ask 10 different Palantirians, 00:12:42.123 --> 00:12:45.653 you'll get 10 slightly different answers on what the ontology is because it's 00:12:45.653 --> 00:12:50.131 a philosophy as opposed to like a tangible thing in some way i mean it. 00:12:50.803 --> 00:12:56.521 I feel like we've maybe we always needed the articulate clarity of an academic 00:12:56.695 --> 00:13:00.853 in our industry to like actually communicate some of these things in the way 00:13:00.853 --> 00:13:01.815 that we could consume them, 00:13:02.349 --> 00:13:07.103 that's helpful i think is it too large of a generalization to say that there's 00:13:07.103 --> 00:13:08.886 There's a couple of models in which, 00:13:09.462 --> 00:13:11.313 companies are approaching AI. 00:13:11.313 --> 00:13:17.630 It's there's a disaggregated set of data that we're kind of dropping a large language model into, 00:13:18.182 --> 00:13:22.303 and then the model you just described, which is like, hey, let's actually go 00:13:22.303 --> 00:13:28.293 in and do some mapping and understand the context and connections between these data sets. 00:13:29.314 --> 00:13:33.044 And then fit them into these modular pieces. It's an oversimplification and 00:13:33.044 --> 00:13:36.019 probably a terrible analogy, but like, are those opposing models? 00:13:36.582 --> 00:13:40.854

Yeah, I would say that there's the kind of naive approach, which is almost reminiscent 00:13:40.854 --> 00:13:43.724 of, you know, whenever a new technology comes out, people try to throw it against 00:13:43.724 --> 00:13:45.884 everything. You saw this with blockchain, where they're like, 00:13:45.884 --> 00:13:49.214 oh, you know, daycare on the blockchain, babysitting on the blockchain. 00:13:49.214 --> 00:13:50.814 You're like, why are you leveraging this tool? 00:13:51.774 --> 00:13:55.514 You actually saw the same thing with the steam engine back in the day with electricity. 00:13:55.514 --> 00:13:58.084 They're like, well, I don't know. Maybe if we just shock people, 00:13:58.084 --> 00:13:59.237 they'll lose fat or something. 00:13:59.854 --> 00:14:02.994 New technology applied broadly. So a lot of companies are like, 00:14:02.994 --> 00:14:06.011 okay, cool, LLMs are great. What if we just throw everything at clawing? 00:14:06.429 --> 00:14:11.704 It turns out that LLMs are not actually well-suited for that because they are 00:14:11.704 --> 00:14:13.470 generally trained. They're next token predictors. 00:14:13.761 --> 00:14:17.144 So the ontology, and this has been what we've been doing for the past couple 00:14:17.144 --> 00:14:21.524 of years, is you could think of it in a sense as retrieval augmented generation 00:14:21.524 --> 00:14:25.654 in that you're telling the LLM, hey, here's the version of reality you care 00:14:25.654 --> 00:14:28.412 about, and you have to return results that make sense with this. 00:14:28.772 --> 00:14:31.464 So people will say, oh, you're training the model. Technically speaking, 00:14:31.464 --> 00:14:32.987 you're not actually adjusting the weights. 00:14:33.277 --> 00:14:37.014 But what we've recently been doing, and Alex Karp has been talking about in 00:14:37.014 --> 00:14:41.844 CNBC and elsewhere, is leveraging open weight models to actually do the training, 00:14:41.844 --> 00:14:44.584 to say, hey, you're going to take an open weight model and you're going to make 00:14:44.584 --> 00:14:47.508 it your own by capturing all the pieces that are meaningful to you. 00:14:47.705 --> 00:14:51.464 And that's kind of that paired with the ontology becomes a very powerful way 00:14:51.464 --> 00:14:55.759 for you to maintain sovereignty over your underlying business process, 00:14:55.986 --> 00:15:00.104 your tradecraft that you've developed over 100 years in the construction industry 00:15:00.104 --> 00:15:01.953 and in any industry. Hmm. 00:15:02.861 --> 00:15:06.629 So what does a forward deployed engineer do within that? 00:15:07.162 --> 00:15:11.719 Yeah, so a forward deployed engineer, this concept kind of came out of our work with the U.S. 00:15:11.719 --> 00:15:15.379 Government, which was, as I understand it, because I wasn't at the company 25 00:15:15.379 --> 00:15:20.509 years ago, was a counter to the position of having software developers sit outside 00:15:20.509 --> 00:15:25.080 the problem,

get a list of specifications, and then try and design something to solve that. 00:15:25.405 --> 00:15:29.989 The problem with this is that any specification list is at best an imperfect 00:15:29.989 --> 00:15:31.702 map of the underlying territory. 00:15:31.836 --> 00:15:34.645 You're going to leave something out. It's going to be sanitized in some way. 00:15:34.930 --> 00:15:38.959 You're going to ignore some edge cases. And so by putting our engineers as close 00:15:38.959 --> 00:15:43.958 to the actual problem as we could, and this involves sending people to war zones back in the day, 00:15:44.539 --> 00:15:48.289 sending people out to the North Sea, me going out to Kavanaugh Construction 00:15:48.289 --> 00:15:52.716 every other week for three months in the winter of 25, I think it was, 00:15:53.227 --> 00:15:58.359 so that I can leverage not just my intellectual understanding and how I encode 00:15:58.359 --> 00:16:01.119 a problem into a map, but also just sit there and be like, oh, 00:16:01.119 --> 00:16:04.859 like I would have never realized these things about how this actually works 00:16:04.859 --> 00:16:06.602 if I wasn't there seeing them and living them. 00:16:06.811 --> 00:16:09.981 So that's what we do. We put our engineers as close to the problem as we can 00:16:10.149 --> 00:16:12.960 so they can own the outcomes for the client. Yeah. 00:16:13.602 --> 00:16:17.571 I mean, is that Kaizen, like the Japanese concept of like get close to the problem? 00:16:18.090 --> 00:16:22.485 Yeah, I think that makes nothing but sense. So like tactically, 00:16:23.122 --> 00:16:27.529 building an ontology, like day one at Kavanaugh, and we'll maybe talk more about 00:16:27.529 --> 00:16:31.169 what that implementation look like, but maybe walk me through the tools. 00:16:31.169 --> 00:16:36.719 Like in my head, that's like, it's a really complicated whiteboard. Are you using like Miro? 00:16:37.059 --> 00:16:40.709 Like maybe I'm just like oversimplifying everything, but walk me through how 00:16:40.709 --> 00:16:45.470 you start to map a hundred year old plus companies, you know, context. 00:16:46.137 --> 00:16:49.643 Yeah. So I would say day one, or it's almost kind of day zero, 00:16:49.858 --> 00:16:52.909 you start off with conversations to understand what the problem space is, 00:16:52.909 --> 00:16:54.647 you know, what matters to the customer. 00:16:54.931 --> 00:16:58.469 That's part of what I do in my day-to-day job is kind of sit down and say, 00:16:58.469 --> 00:17:01.471 hey, you know, I understand the toolset. You understand your problem space. 00:17:01.709 --> 00:17:03.044 Where is there some synergy there? 00:17:03.760 --> 00:17:07.104 And I think breaking it down into what are you trying to do, 00:17:07.353 --> 00:17:11.565 right? So construction, if you have to horrifically oversimplify it, 00:17:11.565 --> 00:17:14.010 and this is a quote I'm stealing from Joe Petra, actually. 00:17:14.902 --> 00:17:19.251 Put someone who's not sick in a truck that's not broken against a job that needs 00:17:19.251 --> 00:17:21.694 work. If you can do that, you're doing construction. 00:17:22.233 --> 00:17:26.081 And so once we boil it down to those pieces, you know, that's really the core 00:17:26.081 --> 00:17:27.356 concepts of the ontology. 00:17:27.583 --> 00:17:30.931 So you're going to have, you know, your materials, your equipment, 00:17:30.931 --> 00:17:34.581 your people, your contracts. In a hospital setting, it could be your patients, 00:17:34.581 --> 00:17:37.469 your doctors, your operating bays, your records. 00:17:38.311 --> 00:17:42.031 Understanding what those core pieces that you work with and how you interact 00:17:42.031 --> 00:17:45.230 with them is how you start to build the nouns and the verbs of the ontology. 00:17:45.561 --> 00:17:49.461 And then from there, you know, there's the piece of connecting to the data sources, 00:17:49.461 --> 00:17:50.936 which is a bit more technical. 00:17:51.238 --> 00:17:53.841 How do you make sure you're importing the right access restrictions? 00:17:53.841 --> 00:17:56.768 Very important in hospitals, obviously, with PII. 00:17:57.047 --> 00:18:00.981 Important in an HR context in construction. You don't want someone to be able 00:18:00.981 --> 00:18:03.038 to just say, hey, what's everybody make at the company? 00:18:03.722 --> 00:18:07.841 And then from there, once you have a starting ontology, and it's iterative, 00:18:07.841 --> 00:18:10.411 it's not like you connect all the data, then build all the ontology, 00:18:10.411 --> 00:18:13.752 then build the front end. It's really building the plane as you fly it. 00:18:14.211 --> 00:18:17.751 So you start from a point of like, okay, what's something that you can't do 00:18:17.751 --> 00:18:20.225 right now or that you're doing in an imperfect way? 00:18:20.358 --> 00:18:23.731 Let's build that out. And then from there, what's the next marginal piece, 00:18:23.731 --> 00:18:25.931 right? Because these businesses are integrated. 00:18:26.210 --> 00:18:31.511 They're not just kind of siloed. So when I approve an invoice, something happens. 00:18:31.511 --> 00:18:35.491 It maybe gets written to an ERP or it goes to the CFO for a further approval 00:18:35.491 --> 00:18:38.771 or it writes back to the project engineer to say, hey, this happened. 00:18:38.771 --> 00:18:41.562 It's changing your profit margin on that job. 00:18:41.852 --> 00:18:45.621 How do we start to map for every decision? What are all the consequences that 00:18:45.621 --> 00:18:49.143 can be pushed? What are all the necessary kind of outcomes of that? 00:18:49.486 --> 00:18:52.508 And how do we make all of that happen more rapidly and more clearly to people? 00:18:52.764 --> 00:18:56.641 Because in construction, it's really, I like to think of it as undoing the death 00:18:56.641 --> 00:18:59.480 by a thousand cuts that happens across a job site. 00:18:59.869 --> 00:19:03.660 You know, every five minute

delay, cause you had to pick up the phone is costly in some way. 00:19:03.885 --> 00:19:07.301 And it can be hard to identify what that is, but you know, that there's, 00:19:07.301 --> 00:19:10.411 there's fat to trim somewhere in the business. And this is what we've heard 00:19:10.411 --> 00:19:13.347 from everyone we've spoken to is like, I think this could be better. 00:19:13.527 --> 00:19:16.807 I just don't know exactly where to start making it better. Hmm. 00:19:17.736 --> 00:19:21.105 So, and maybe, I don't know if you can speak to this or not, 00:19:21.465 --> 00:19:26.745 but McCarthy and Palantir had a big announcement, and the word that I picked up there was pulse. 00:19:27.305 --> 00:19:30.925 And I think one of the questions that I had was, is that proprietary to McCarthy? 00:19:31.245 --> 00:19:34.645 Is that an offshoot of the Foundry? Can you speak to the difference between 00:19:34.645 --> 00:19:39.623 what McCarthy-Palantir partnership is working slash producing versus the Foundry? 00:19:40.235 --> 00:19:45.725 Yeah. So, Foundry is best thought of as a, you know, that's why it's named that. 00:19:45.725 --> 00:19:47.603 It's a Foundry. It's a place where you build. 00:19:47.934 --> 00:19:51.962 So what our customers build inside of Foundry is their own IP. 00:19:52.189 --> 00:19:55.730 It's based on their data. So they own what they build. 00:19:56.015 --> 00:19:58.695 Companies will oftentimes choose to name it something, right? 00:19:58.695 --> 00:20:01.875 So I believe Kavanaugh called theirs Tom after their founder, 00:20:01.875 --> 00:20:03.304 Total Operations Management. 00:20:03.670 --> 00:20:07.357 I believe Pulse is the name that McCarthy has chosen for kind of their platform. 00:20:07.601 --> 00:20:10.608 So effectively, it's a branding of what you're building yourself. 00:20:10.805 --> 00:20:15.095 Because at the end of the day, it's a marriage of our tools and your subject 00:20:15.095 --> 00:20:19.468 matter expertise and your business logic. and it creates something that you, so, yeah. 00:20:20.190 --> 00:20:21.881 That's helpful. So- 00:20:23.093 --> 00:20:28.313 As maybe Palantir has kind of started to get into construction and had some 00:20:28.313 --> 00:20:32.213 of these conversations and mapped some of these processes, I guess, 00:20:32.893 --> 00:20:35.633 what's the organizational view of construction? 00:20:35.633 --> 00:20:38.433 You know, I think there's like the McKinsey model, which is like, 00:20:38.590 --> 00:20:42.909 if we just all adopted tech 15 years ago, we'd all be much more productive, 00:20:43.327 --> 00:20:44.843 which is a bit maybe reductionist. 00:20:44.843 --> 00:20:49.693 But what is, has Palantir started to mine some different takeaways from these 00:20:49.693 --> 00:20:55.265 implementations to say like, wow, here is the true constraint of, of this industry? 00:20:55.842 --> 00:20:59.823 Yeah. So, I mean, I think the first thing to say is that the structure of Palantir 00:20:59.823

--> 00:21:03.365 is very kind of non-hierarchical in the sense that we do have a C-suite, 00:21:03.655 --> 00:21:06.573 but kind of within the company, it's more of an artist colony. 00:21:06.573 --> 00:21:10.903 So I have positions on what I think is valuable in construction are people who 00:21:10.903 --> 00:21:14.415 work in insurance. Maybe they don't have any thoughts on that, or maybe they do. 00:21:14.733 --> 00:21:20.773 But what I can speak to is that the takeaways that I've seen is that breaking it down by GC. 00:21:21.377 --> 00:21:24.093 Self-performing company, and engineering architecture company, 00:21:24.093 --> 00:21:27.893 some of the pieces I've seen is that for GCs, a big problem is that there's 00:21:27.893 --> 00:21:31.636 a black box at the center of their business operation, which is the subcontractor. 00:21:32.231 --> 00:21:35.823 Effectively, what a GC does is underwrites the risk of saying, 00:21:35.823 --> 00:21:37.380 hey, we're going to make this happen. 00:21:37.537 --> 00:21:40.427 You don't have to worry about the exact details. You don't have to go and find 00:21:40.555 --> 00:21:45.252 your electrician, your drywall person. We'll handle that. We'll project manage it. 00:21:45.711 --> 00:21:48.543 But then they have to go find those people and understanding, 00:21:48.543 --> 00:21:52.861 you know, one problem you can imagine is if a sub goes default, what's left to be done? 00:21:53.318 --> 00:21:56.773 How do you think about concentration risk of a given subcontractor, 00:21:56.773 --> 00:21:59.773 especially if you're across North America? Maybe your relationship with them 00:21:59.773 --> 00:22:02.351 in the Northeast is different than it is in Texas. 00:22:02.859 --> 00:22:06.623 So that kind of difficult problem that doesn't necessarily have a definitive 00:22:06.623 --> 00:22:10.609 answer because you're dealing with risk, that's one place we've seen a lot of value. 00:22:10.934 --> 00:22:14.140 The project management of just the communication of making it clear. 00:22:14.702 --> 00:22:19.174 Project managers in general, their job is to make things run because they're 00:22:19.174 --> 00:22:22.109 pulling from so many different places. So the more efficient you can make them, 00:22:22.335 --> 00:22:23.676 the more projects they can run. 00:22:23.978 --> 00:22:27.416 And then beyond that, you can start to extract information about how things have gone. 00:22:27.828 --> 00:22:31.774 In the self-performed space, the one that I think is most important is like 00:22:31.774 --> 00:22:34.506 closing that field operations gap. 00:22:34.860 --> 00:22:39.670 And over the last 20 years, as construction tech has gotten more and more capable, 00:22:40.059 --> 00:22:44.320 a lot of the burden of entering and capturing that data has landed on the people in the field. 00:22:44.753 --> 00:22:47.724 And so you have these situations where, you know, now half of a foreman's day 00:22:47.724 --> 00:22:52.244 is spent in a site

trailer, keying in invoices, keying in time cards, 00:22:52.244 --> 00:22:55.304 doing production tracking, and they're not out on the field making sure the 00:22:55.304 --> 00:22:58.704 job is actually running day to day, which is what they're trained and what they want to do. 00:22:58.976 --> 00:23:04.823 So how can we still get that data, which is valuable, without burdening the field to do so? 00:23:04.951 --> 00:23:09.444 A kind of trivial example of this that I like to think of is that any piece 00:23:09.444 --> 00:23:13.294 of data should be entered only once, and the only data you should be responsible 00:23:13.294 --> 00:23:15.537 for entering is the data that only you know. 00:23:15.728 --> 00:23:20.044 So when I pull up my time card, I shouldn't have to tell you what job I'm on. 00:23:20.044 --> 00:23:23.222 I showed up that day. Someone told me to be there. I didn't just wander in. 00:23:23.391 --> 00:23:28.124 So I should already know what job I'm on, who I am, what equipment's in my cache, 00:23:28.124 --> 00:23:32.124 all of that. The only piece I tell you is when I clocked out and when I clocked 00:23:32.124 --> 00:23:34.780 in, right? Because that's the net new piece for me. 00:23:34.885 --> 00:23:39.824 And it makes it seamless because I'm not going to make data entry errors because 00:23:39.824 --> 00:23:42.610 it's very easy and I'm owning the data that I'm creating. So that's, 00:23:42.865 --> 00:23:45.048 I think, one key piece in SPC that we've seen. 00:23:45.627 --> 00:23:49.774 I mean, I think you got at least a handful of cheers when you said that, 00:23:49.774 --> 00:23:50.564 because I think you're right. 00:23:50.564 --> 00:23:56.064 I mean, I think we have understood the power of data and I think at times not 00:23:56.064 --> 00:24:01.334 gone through the work to map the inputs of data well enough before we've set 00:24:01.334 --> 00:24:04.115 expectations around what should be recorded. Yeah. 00:24:05.222 --> 00:24:08.822 And I think, you know, because of what's happening with AI and some of these 00:24:08.822 --> 00:24:13.092 models being trained, you know, everyone's suddenly kind of rereading the data 00:24:13.092 --> 00:24:14.867 sharing clauses in their contracts. 00:24:15.273 --> 00:24:19.432 And I think that makes sense for all kinds of reasons. But I think at the same 00:24:19.432 --> 00:24:23.722 time, we overestimate the value of a lot of data. 00:24:24.242 --> 00:24:29.522 And I think a lot of data in these organizations hasn't yet been mapped. 00:24:29.942 --> 00:24:34.472 And I think when you look at means and methods as a key kind of component of 00:24:34.472 --> 00:24:38.222 competitive advantage, I've made the argument before, we've made the argument 00:24:38.222 --> 00:24:42.052 before on the show, that a lot of that still remains in people's heads, right? 00:24:42.052 --> 00:24:48.717 And so, like, true IP won't show up on a spreadsheet, is not really showing up in a data set. 00:24:48.996 --> 00:24:56.622 How do you handle and manage, as you go through and create this ontology, the missing IP? Right. 00:24:57.127 --> 00:25:01.254 Yeah. I mean, I think to your point, one comment I wanted to make is that you're 00:25:01.254 --> 00:25:04.807 exactly right that a lot of what's being tracked is never used. 00:25:05.844 --> 00:25:08.034 I remember making this argument to Kavanaugh. It's like, well, 00:25:08.034 --> 00:25:11.469 you're not tracking what color socks your foremen are wearing because you know it doesn't matter. 00:25:11.626 --> 00:25:14.406 You're probably tracking a bunch of stuff that you're also not using. 00:25:14.621 --> 00:25:19.064 Like there's this dream that if we have data tracking so good that the estimators 00:25:19.064 --> 00:25:23.461 can now kind of run like, you know, a postmortem and get better the next time. 00:25:23.728 --> 00:25:26.944 But statistically, the problem here is that unless you're like in the data center, 00:25:26.944 --> 00:25:30.164 space where you might do the same project or very similar projects over and 00:25:30.164 --> 00:25:34.993 over again, unless I'm paving the same road in the same conditions 10 times, 00:25:35.237 --> 00:25:37.509 I don't really have enough kind of 00:25:37.869 --> 00:25:40.744 statistical power to really fine tune that. 00:25:41.304 --> 00:25:45.324 I'm really reliant on that IP in my estimators heads, in my schedulers heads 00:25:45.324 --> 00:25:50.731 to know, yeah, I have a feel for this. I remember 10 years ago when we paved that road, it has a clay. 00:25:52.844 --> 00:25:55.634 Substrate that makes it harder to dig or something like that these little pieces 00:25:55.634 --> 00:25:59.570 of information and so to answer your question of how we encode that, 00:26:00.223 --> 00:26:04.824 part of it is that as you make decisions in foundry that becomes a source of 00:26:04.824 --> 00:26:06.899 data you can do this in a few ways you can 00:26:07.154 --> 00:26:09.924 do it very easily it's just like an audit log of like this person did this at 00:26:09.924 --> 00:26:12.193 this time you could ask for an explanation 00:26:12.448 --> 00:26:16.094 you can start to have ai models suggest answers and then if you disagree with 00:26:16.094 --> 00:26:20.674 it say hey i this is why i disagree that becomes part of how the model is suggesting 00:26:20.674 --> 00:26:23.716 it for you so that's one way that you can start to encode those pieces. 00:26:24.199 --> 00:26:27.554 So you can do it bit by bit. You can try to start it from, you know, 00:26:27.554 --> 00:26:30.634 from the very beginning of saying, hey, just speak to me for an hour. 00:26:30.861 --> 00:26:35.704 I'll record you. I'll pump that into Foundry, use AI FTE to try and encode it 00:26:35.704 --> 00:26:37.577 as a set of rules. And then you work with me. 00:26:38.244 --> 00:26:42.834 But really what it is, is like, how can how can we give you the best tools that 00:26:42.834 --> 00:26:46.444 let you that can

work on the parts of the problem that a human being is like 00:26:46.444 --> 00:26:48.063 slower, inefficient as at 00:26:48.376 --> 00:26:51.874 and leave you with the ability to focus on those pieces that you are better 00:26:51.874 --> 00:26:56.387 than any AI in the world that dealing with because of your experience in that field. Hmm. 00:26:57.487 --> 00:26:58.567 Yeah, that makes sense. 00:27:00.844 --> 00:27:05.354 And McCarthy maybe destroys the thesis of the question I'm about to ask a little 00:27:05.354 --> 00:27:06.918 bit, or the McCarthy partnership, but, 00:27:07.493 --> 00:27:12.894 I can imagine it's a little bit more data rich working with a fully vertically 00:27:12.894 --> 00:27:16.324 integrated contractor, right? there's just a lot more parameters. 00:27:16.824 --> 00:27:19.921 There's a lot more variables you can seemingly control. 00:27:22.137 --> 00:27:27.554 Is that a factor as Palantir evaluates opportunities and kind of thinks through 00:27:27.554 --> 00:27:30.814 the model of how they integrate with or interact with the market? 00:27:31.486 --> 00:27:35.264 Yeah, I mean, at the end of the day, as powerful as we are and as good as our 00:27:35.264 --> 00:27:38.017 software and forward to put engineers are, we're not actually magicians. 00:27:38.284 --> 00:27:41.744 So if you don't have the data, there's not much we can do with that. 00:27:41.744 --> 00:27:44.216 But we can be very creative in how you get that data. 00:27:44.506 --> 00:27:50.164 An illustrative example would be cost allocation, right? So you can be creative in this. 00:27:50.624 --> 00:27:55.094 If I am doing two tasks today, right? Like I am, you know, breaking the side 00:27:55.094 --> 00:27:59.125 of a road and I'm also picking up trash on the side of the road. 00:27:59.696 --> 00:28:05.664 And we kind of know, roughly speaking, how much time it should take to do a 00:28:05.664 --> 00:28:09.557 unit of work there. We kind of have a sense of that from historical estimates. 00:28:10.027 --> 00:28:14.204 I can say, okay, cool. I look at the production I logged. I did four units of 00:28:14.204 --> 00:28:16.668 trash and four units of raking. 00:28:16.796 --> 00:28:21.474 Well, I know that the ratio is like two to one, so I could infer that my labor 00:28:21.474 --> 00:28:25.528 costs for that day should be split in that way across those two line items. 00:28:25.690 --> 00:28:27.565 Again, kind of a specific example. 00:28:28.378 --> 00:28:32.189 There's so many pieces like this where you can back into the data you wish you 00:28:32.189 --> 00:28:36.975 had without making someone put it in by hand because you already have enough of those pieces. 00:28:37.341 --> 00:28:41.090 But yes, in general, if the data doesn't exist, it could be hard to work with. 00:28:41.398 --> 00:28:45.479 But, you know, we've done very interesting data collection and data aggregation 00:28:45.479 --> 00:28:49.127 exercises in the past, all the way down to like sticky notes on the side of someone's desk. 00:28:51.019 --> 00:28:58.049 So what was, did you have kind of a presupposed model of construction that has now been challenged? 00:28:59.689 --> 00:29:04.307 For context, you know, I was a business econ major coming out of school and, 00:29:04.876 --> 00:29:06.529 kind of ended up in construction. 00:29:06.989 --> 00:29:12.358 And my first project as a project engineer, entry-level kind of management position on site, 00:29:12.863 --> 00:29:18.880 it really destroyed the little I knew about the industry in that the folks building, 00:29:19.333 --> 00:29:22.209 were much smarter than I ever thought, more innovative. 00:29:23.409 --> 00:29:26.299 And just really good at problem solving. 00:29:26.299 --> 00:29:29.892 And I think at times the narrative outside looking into construction is, 00:29:30.391 --> 00:29:33.474 you know, it's a bunch of knuckle draggers who don't really know how to do much. 00:29:34.294 --> 00:29:39.393 Obviously, that's not your opinion, but as you came into interacting with the 00:29:39.393 --> 00:29:44.041 industry, what were some of your pre-suppositions that maybe got challenged? 00:29:44.558 --> 00:29:48.913 Yeah, I think, you know, from the outside, I kind of had this vision that it 00:29:48.913 --> 00:29:53.783 was much more orderly, like a colony of ants being kind of centrally commanded 00:29:53.783 --> 00:29:56.343 and they do the task and it was clear to them how to do the task. 00:29:56.683 --> 00:30:00.799 They do it on time and, you know, they report back that it had been done. 00:30:01.084 --> 00:30:05.303 What I didn't, I guess, know kind of from the outside is the amount of like 00:30:05.303 --> 00:30:08.593 rich contacts that people in the job site have. Like they're really handed kind 00:30:08.593 --> 00:30:12.593 of an undifferentiated problem and told, hey, you're close to it. 00:30:12.593 --> 00:30:15.507 Figure it out. It's kind of similar, actually, to the way that Palantir works. 00:30:15.734 --> 00:30:18.313 Nobody tells me exactly how to solve it. They're just kind of like, 00:30:18.313 --> 00:30:20.993 hey, go figure out how to make this project work. 00:30:21.336 --> 00:30:26.053 And so, you know, the subsidiarism of it in the sense that every problem is 00:30:26.053 --> 00:30:29.313 like solved at the lowest level that can be solved. There's a lot of that. 00:30:29.313 --> 00:30:33.093 Like there's a lot of trust and, you know, they make it work, 00:30:33.093 --> 00:30:35.323 right? You break the system a lot of the times. 00:30:35.963 --> 00:30:39.393 I know this kind of example of, I'm going to get the term wrong, 00:30:39.393 --> 00:30:43.103 but it's when you do a strike where you actually just adhere to the rules exactly. 00:30:43.451 --> 00:30:46.893 And that like destroys your ability to be efficient on job sites. 00:30:47.131 --> 00:30:50.283 I can't remember the exact term for that, but like that was one of the things 00:30:50.283 --> 00:30:53.616 that

surprised me was just how much tacit knowledge everyone had in the field. 00:30:54.398 --> 00:30:59.163 And yeah, they're very capable people. like it's not the same to be non-technical 00:30:59.163 --> 00:31:02.343 as non-intelligent and i knew that going in but i think it's worth stating that 00:31:02.343 --> 00:31:07.857 people do sometimes have that assumption yeah no i think i think it's interesting because it's, 00:31:08.455 --> 00:31:13.153 it's it's like it's muscle memory like not not codified memory i don't know 00:31:13.153 --> 00:31:16.183 if that's that's the the analogy but yeah i understand what you're saying 00:31:16.503 --> 00:31:20.433 and it is interesting to see it in action which is kind of it's like organized 00:31:20.433 --> 00:31:24.313 chaos like how is this actually working And, you know, sometimes it doesn't at all. 00:31:24.403 --> 00:31:26.713 And obviously when you look for, from the outside of the end, 00:31:26.713 --> 00:31:30.373 like from a metric standpoint, from a mental health standpoint, 00:31:30.373 --> 00:31:31.684 a lot of ways it's not working. 00:31:32.805 --> 00:31:39.761 And I think we're at this edge of this new frontier, whether it's AI and its applications. 00:31:39.761 --> 00:31:43.283 But what I appreciate about the Palantir model 00:31:43.678 --> 00:31:48.751 is that you're getting really close to the problem with the minds who can potentially 00:31:48.751 --> 00:31:54.835 pair those problems with technology that can kind of modularize around 00:31:55.270 --> 00:31:57.381 the operation and drive some efficiency. 00:31:57.381 --> 00:32:02.901 So maybe digging into the Kavanaugh story, I think it would be interesting to 00:32:02.901 --> 00:32:07.305 understand this is an established business in Ontario. In Ontario? 00:32:07.613 --> 00:32:08.617 Yeah. Did I get that right? Okay. 00:32:09.175 --> 00:32:12.774 And heavy civil contractor, vertically integrated. 00:32:13.699 --> 00:32:15.852 I mean, it's probably a whole separate conversation to understand like, 00:32:16.247 --> 00:32:21.521 hey, that's a really forward thinking kind of C-suite to understand the value 00:32:21.521 --> 00:32:23.051 of a partnership with Valenteer. 00:32:23.544 --> 00:32:27.851 But that aside, what was the implementation path? 00:32:28.271 --> 00:32:32.061 Obviously, you're building out this ontology. Is that a 30-day effort? 00:32:32.061 --> 00:32:34.793 Is that a six-month effort? And then talking through. 00:32:36.851 --> 00:32:42.891 Change management 101, how do you go from changing behavior over time as you 00:32:42.891 --> 00:32:44.432 kind of roll out the foundry? 00:32:45.182 --> 00:32:50.671 Yeah. So I think that the path to engagement was we did like an initial engagement, came on site. 00:32:50.671 --> 00:32:54.941 I was there for a couple of days, identified a bit of data, started to solve 00:32:54.941 --> 00:32:57.451 a problem. And we started fairly simple. It was kind of a question of like, 00:32:57.451 --> 00:33:01.801 hey, our dispatch and our maintenance teams, there's some friction there. 00:33:01.801 --> 00:33:03.395 How can we help alleviate that? 00:33:04.105 --> 00:33:08.951 That then got us onto the path of doing a bit more of an engagement focused 00:33:08.951 --> 00:33:13.138 on things like how do you do closer to real-time job costing? 00:33:13.631 --> 00:33:17.471 And then from there, we moved into kind of a larger process of like, 00:33:17.471 --> 00:33:20.137 okay, cool. We all agree that there's some power here. 00:33:20.688 --> 00:33:24.621 Let's think about what it would look like to do everything inside of Foundry 00:33:24.621 --> 00:33:28.316 and chart a path towards that. It's like the North Star and we're charting a course. 00:33:30.096 --> 00:33:34.621 So you're building ontology in stages. You start from the data that you have. 00:33:34.737 --> 00:33:38.380 And then in the case of a company like Kavanaugh, they moved to starting to 00:33:38.380 --> 00:33:42.350 create that data inside a foundry. So there were multiple iterations. 00:33:42.730 --> 00:33:45.550 We threw a lot of spaghetti at the wall early on to try and figure out what 00:33:45.550 --> 00:33:49.440 the right way. There's an art to creating an ontology because it has to operate 00:33:49.440 --> 00:33:51.255 the way people think about the things. 00:33:51.453 --> 00:33:55.850 But there's also edge cases where you want to start to change how people work with the tools. 00:33:55.850 --> 00:33:59.900 Like, for example, if people are used to thinking about, you know trucks and 00:33:59.900 --> 00:34:01.440 equipment in the same way because 00:34:01.440 --> 00:34:04.207 they they all move you're like well actually trucks have a different 00:34:04.625 --> 00:34:07.920 property than equipment they they are carrying something or something like that 00:34:07.920 --> 00:34:10.076 so you have to kind of get into the the pieces there 00:34:10.419 --> 00:34:13.650 and so it was there were waves of iteration and we in building the ontology 00:34:13.650 --> 00:34:18.640 we laid the groundwork over the first 30 30 days extended that over 90 these 00:34:18.640 --> 00:34:20.070 are all kind of rough timelines, 00:34:21.052 --> 00:34:24.280 and then as we built out the use cases there was feedback from them saying hey 00:34:24.280 --> 00:34:27.750 like this is pretty cool it addresses this problem, but it needs to work this 00:34:27.750 --> 00:34:28.807 way. So there was rework. 00:34:29.069 --> 00:34:33.620 And that's, I think, part of the reason why it can be hard to understand how 00:34:33.620 --> 00:34:36.800 Palantir works from the outside in, especially if you're used to dealing with 00:34:36.800 --> 00:34:39.500 software providers that say, here's exactly what I'm going to do. 00:34:39.500 --> 00:34:43.131 Here's the milestone date. It'll work like this. Here's the UAT testing. 00:34:43.619 --> 00:34:47.220 We're trying to revolutionize construction, and that's going to

break some things. 00:34:47.220 --> 00:34:49.858 And it's going to be a little bit chaotic as we do it. 00:34:50.067 --> 00:34:54.083 And to their credit, they were very amenable to giving us that space to do that. 00:34:55.072 --> 00:34:58.598 And then I think what was the second part of your question? I mean, 00:34:58.598 --> 00:35:03.698 I asked like 15 questions in that one. So you did very well to even kind of piece out what you did. 00:35:04.000 --> 00:35:10.771 I think maybe just stopping there, like 90 days in, is part of the kind of map. 00:35:11.543 --> 00:35:16.178 Like understanding the existing tech stack and not to scare all the construction 00:35:16.178 --> 00:35:19.636 tech providers out there, but like, okay, this is what now goes away? 00:35:20.342 --> 00:35:24.528 Yeah, so I would say that in general, we don't necessarily approach from a, 00:35:24.528 --> 00:35:25.947 hey, we're going to rip and replace. 00:35:26.417 --> 00:35:30.208 Kavanaugh had that mentality as we got in of saying, hey, we'd like to just 00:35:30.208 --> 00:35:31.270 get this all in one system. 00:35:31.549 --> 00:35:37.035 But our design for Foundry is to be very interconnected. It can integrate with whatever you have. 00:35:37.366 --> 00:35:40.488 If you have a piece of software you like, we're not going to cause unnecessary 00:35:40.488 --> 00:35:43.858 change management by saying, hey, get rid of that thing that's working well, right? 00:35:43.858 --> 00:35:47.518 We'll integrate with it both bi-directional so we can write back to it. 00:35:47.518 --> 00:35:50.778 So if you're doing something with that outside of Foundry, you can still get 00:35:50.778 --> 00:35:52.521 benefits from being connected to Foundry. 00:35:52.887 --> 00:35:55.608 And then if you choose at some time to start to deprecate it because it doesn't 00:35:55.608 --> 00:35:58.373 have the features you want or you want to go a different direction, 00:35:58.704 --> 00:36:01.478 that's very easy to pull something out or to bring something in, right? 00:36:01.478 --> 00:36:04.817 You know, as these new tools come in for in construction tech, 00:36:05.037 --> 00:36:08.838 you know, we're not positioned that we want to get rid of everything. 00:36:08.838 --> 00:36:10.291 If it works, we're happy to work with it. 00:36:10.673 --> 00:36:13.258 And I think the change management piece was the one that you mentioned, 00:36:13.258 --> 00:36:14.574 which I think is very interesting. 00:36:15.026 --> 00:36:19.888 So when I first went to Kavanaugh and we started the year-long engagement, 00:36:19.888 --> 00:36:24.008 I made Joe two promises. The first one is that we are going to be subtractive before we're additive. 00:36:24.292 --> 00:36:27.488 We're going to make people's lives easier before we ever ask them to do a new 00:36:27.488 --> 00:36:30.288 thing because we have to earn the right, to earn the trust to say, 00:36:30.288 --> 00:36:33.608 hey, we're on your side. We're not just trying to make

your lives more complicated. 00:36:33.608 --> 00:36:37.730 It's not a top-down approach. We're here in the field with you making your lives better. 00:36:37.980 --> 00:36:41.658 And the second promise I made is that nothing we build for the field is going 00:36:41.658 --> 00:36:44.288 to make a guy in that field throw his phone across that field. 00:36:44.288 --> 00:36:47.240 It has to work. And so that was the way that we approached it. 00:36:47.482 --> 00:36:51.005 The change management piece was led largely by Kavanaugh's team. 00:36:51.678 --> 00:36:55.000 I think they onboarded like 800 people over a two-week period. 00:36:55.219 --> 00:36:59.893 It was very coordinated. They actually ran the trainings inside of Foundry and 00:36:59.893 --> 00:37:05.103 built kind of like almost like a ticketing system and like an Asana-type workflow 00:37:05.103 --> 00:37:06.783 to track things inside of Foundry. 00:37:06.783 --> 00:37:09.823 So, you know, we say, hey, this person's going to be trained this day. 00:37:09.823 --> 00:37:13.138 They get a notification on their phone that they had to show up at that time. 00:37:13.746 --> 00:37:15.487 And they approached it really well. I've seen like, hey, like, 00:37:15.696 --> 00:37:20.143 you know, another piece that we did, I think that helped a lot is that we identified 00:37:20.427 --> 00:37:23.556 people across the org who would be amenable to piloting with us. 00:37:24.051 --> 00:37:27.743 Kavanaugh was very non-precious about it. They said, we don't care if you don't 00:37:27.743 --> 00:37:30.973 have a CS degree. We don't care if you're a truck driver, a foreman. 00:37:31.124 --> 00:37:34.503 If you think you have value that you can provide in the back office by helping 00:37:34.503 --> 00:37:37.208 us build these things, come on and try it out. 00:37:37.579 --> 00:37:40.953 Open Casting Call. And so we're able to find people, find a foreman, 00:37:40.953 --> 00:37:45.543 find a truck driver, find an estimator who were able to say, hey, I have an idea. 00:37:45.543 --> 00:37:49.143 I want to make work. I see some brokenness in the system and then they were 00:37:49.143 --> 00:37:52.733 very good conduits because they can go to their guys and say hey look i'm i'm 00:37:52.733 --> 00:37:56.303 not an external i've been in your shoes i am in your shoes this actually is 00:37:56.303 --> 00:37:57.554 i'm here to help you you know. 00:37:58.517 --> 00:38:03.153 It's a really interesting approach to change management. I think it's so often 00:38:03.593 --> 00:38:05.053 underestimated as a variable. 00:38:05.813 --> 00:38:10.693 And, you know, I had enough experience in my career around software and process 00:38:10.693 --> 00:38:14.636 change where I tried to get really good at it because it was the thing that, 00:38:15.059 --> 00:38:18.972 you know, was kind of a deterministic in terms of whether or not something was going to be successful. 00:38:19.326 --> 00:38:23.588 Cool. So, I mean, outside of that, I

think just that open casting call to say, 00:38:24.486 --> 00:38:29.972 hey, if you want to be a problem solver, come kind of join this kind of endeavor. I like that approach. 00:38:30.682 --> 00:38:37.273 So if you were to look at what Kavanaugh is doing now versus what it was doing, 00:38:37.613 --> 00:38:41.503 is the main change, you know, you kind of started with like, 00:38:41.503 --> 00:38:44.310 hey, we need to build this unified language. 00:38:44.740 --> 00:38:52.923 And we want to ideally have only data inputs come from someone and the data 00:38:52.923 --> 00:38:55.976 be associated to that person's kind of specific expertise. 00:38:56.784 --> 00:39:01.653 What did success look like after? And obviously there's been some case studies, 00:39:01.653 --> 00:39:02.832 you've written some articles. 00:39:03.633 --> 00:39:07.783 What would Kavanaugh point to? Like, hey, this was the thing that did it for 00:39:07.783 --> 00:39:12.053 us. And did that align up to how success was defined beforehand? Yeah. 00:39:12.803 --> 00:39:16.091 Yeah, I mean, I would say that part of our model is that, you know, 00:39:16.091 --> 00:39:20.251 Palantir, we do sell services, the FDE model to get things running, 00:39:20.251 --> 00:39:23.750 but at our core, we're not a services company, we're a software company. 00:39:24.081 --> 00:39:27.521 And so I think Kavanaugh is a great example of, you know, we worked at full 00:39:27.521 --> 00:39:31.301 tilt for a year, and then we're able to hand it off, like we taught a man to 00:39:31.301 --> 00:39:35.045 fish, and now they're out there fishing, and they're helping other people fish with Kavtera. 00:39:35.446 --> 00:39:38.651 We don't want people to be dependent on us. It's not a black box. 00:39:38.651 --> 00:39:41.368 We really do want to share the ability to build in Foundry. 00:39:41.763 --> 00:39:46.311 And so I think that the journey, there were lots of twists and turns along the 00:39:46.311 --> 00:39:50.170 way of where we thought things would be valuable, pivoting our effort. 00:39:50.709 --> 00:39:56.051 And I think that the early wins in helping them get off a piece of dispatch 00:39:56.051 --> 00:40:00.291 software, making that more agentic, that was one thing that landed well. 00:40:00.738 --> 00:40:05.271 The time card app, that was a very fun process of building. I've never coded 00:40:05.271 --> 00:40:09.331 in React before, but we had one of their developers and on our side coded a 00:40:09.331 --> 00:40:13.791 React app, got it into the Apple iStore. So that was a new thing for me. 00:40:14.145 --> 00:40:17.665 And then getting that out into people's hands, I think that was a really big win for them. 00:40:17.810 --> 00:40:21.631 I think they would probably say, and Joe can comment on this at some point or 00:40:21.631 --> 00:40:25.811 Kavanaugh can, was that getting it out to like 97% of their workforce was a 00:40:25.811 --> 00:40:29.211 really big win there to kind of show that construction

can do it too. 00:40:30.031 --> 00:40:34.571 They can adopt tech if it's done the right way and it's value aligned in making 00:40:34.871 --> 00:40:37.872 their lives better and safer. I think that's the other thing to say is that 00:40:38.133 --> 00:40:42.105 it's not a zero-sum game of just making more profit and hitting the bottom line. 00:40:42.447 --> 00:40:44.902 Especially, I would say, in construction and physical fields, 00:40:45.139 --> 00:40:48.321 you want everyone to come home with all of their fingers, all of their toes 00:40:48.321 --> 00:40:50.196 not injured. You want them to be happier. 00:40:50.382 --> 00:40:53.701 And that's like an intangible ROI that goes into that. I'm sure it has knock-on 00:40:53.701 --> 00:40:56.106 effects. People stay at the job longer. They do better work. 00:40:56.349 --> 00:40:59.740 But, you know, there's a humanistic element to this that's not just, 00:40:59.879 --> 00:41:03.066 hey, how do we efficientize every process in the world? 00:41:03.799 --> 00:41:06.314 Yeah, we care about that here at Palance here, definitely. 00:41:07.180 --> 00:41:12.255 Who is Palantir's competitor? Everyone? Yeah. So my example for this is that 00:41:12.255 --> 00:41:16.425 like, who's the competitor of like a superstore? Like that's a Canadian store, 00:41:16.425 --> 00:41:18.944 like a Safeway or Sobeys or again, Canadian. 00:41:19.805 --> 00:41:23.925 We got Safeway. There we go. Is a butcher a competitor? Yeah. 00:41:23.925 --> 00:41:28.069 In the sense that they both sell meat. Is a cheesemonger? Yeah. A baker is a competitor. 00:41:28.785 --> 00:41:31.425 There's people who compete with different parts of our tech stack. 00:41:31.425 --> 00:41:35.696 Like Databricks is a partner of ours. We have an agreement with them on how we operate. 00:41:35.923 --> 00:41:38.570 They do certain things really well. We do certain things really well. 00:41:39.375 --> 00:41:43.822 So you could say that like inside of the Lego block ecosystem of Palantir, 00:41:43.967 --> 00:41:45.773 there's people competing on different pieces. 00:41:45.940 --> 00:41:49.965 I don't think that there's a company out there that is really a full competitor 00:41:49.965 --> 00:41:53.455 of everything that we do. You could maybe say something like Microsoft Fabric, 00:41:53.455 --> 00:41:55.647 though I don't know if I'd agree with that. 00:41:55.960 --> 00:42:00.065 But yeah, and the way we respond to this is by being integrable. 00:42:00.065 --> 00:42:02.645 If you have Databricks, you don't have to get rid of that to work with us. 00:42:02.645 --> 00:42:06.037 If you have something else, some other tool, some point solution, 00:42:06.322 --> 00:42:10.435 Viewpoint Vista, to name a few, bid to win, heavy job, whatever. 00:42:10.435 --> 00:42:14.732 We'll work with anyone because we're oriented at making the end customer's life better. 00:42:15.951 --> 00:42:20.204 So what maybe, you know, a big kind of segment of the audience that listens 00:42:20.204 --> 00:42:24.154 to the show is, you know, front lines, field ops folks, right? 00:42:24.154 --> 00:42:26.084 PMs, supers, project executives. 00:42:26.864 --> 00:42:30.284 And I think they're kind of stuck between this kind of rock and hard place, 00:42:30.284 --> 00:42:34.694 which is they're using ChatGPT, Claude in their day-to-day. 00:42:35.147 --> 00:42:38.914 There's some maybe light experimenting happening at project level. 00:42:39.390 --> 00:42:44.034 And a lot of organizations i mean some are making meaningful progress but a 00:42:44.034 --> 00:42:48.719 lot are kind of stuck in this like strategic analysis of like what is our, 00:42:49.230 --> 00:42:52.574 what is our cyber security policy around the use of chat gpt even though like 00:42:52.574 --> 00:42:56.674 everyone's using it on their you know personal phones so you know what would 00:42:56.674 --> 00:42:59.133 your message be to them of, 00:42:59.893 --> 00:43:05.269 understanding what's out there in terms of of just personally available frontier models. 00:43:05.675 --> 00:43:12.371 And, you know, obviously Palantir is kind of acting and operating at an enterprise layer. 00:43:12.813 --> 00:43:16.542 A lot of construction tech is kind of selling to the project level. 00:43:17.466 --> 00:43:21.354 Are there plans to kind of bridge that divide and go straight to projects? 00:43:21.354 --> 00:43:22.174 What are your thoughts there? 00:43:22.863 --> 00:43:27.474 So, I mean, the way we kind of enter into every customer is a bit different. 00:43:27.474 --> 00:43:30.164 Like there are certain places where we start at the project management level. 00:43:30.758 --> 00:43:34.125 For large organizations, we might be at a business unit level to start. 00:43:34.560 --> 00:43:38.934 Of course, the general positioning is that the deeper and more mature an ontology 00:43:38.934 --> 00:43:42.551 is across an org, the more value you can get from it because it's more connected. 00:43:43.284 --> 00:43:47.304 But you can build whatever you want on Foundry. 00:43:47.304 --> 00:43:51.494 The way I like to say it is if you have the data and you can describe a process, 00:43:51.494 --> 00:43:54.144 we can find a way to build it in Foundry for you. And of course, 00:43:54.524 --> 00:43:58.304 there's some interesting edge compute issues that people run up against as you 00:43:58.304 --> 00:44:00.490 start to get into the frontier. 00:44:00.705 --> 00:44:03.033 But for a lot of stuff in construction, it's pretty simple. 00:44:03.497 --> 00:44:06.226 It's simple problems that are hard to solve, if that makes sense. 00:44:07.091 --> 00:44:11.166 So I'd say for the project managers, the people who are like in the field and 00:44:11.166 --> 00:44:14.636 kind of managing the field, it's, you know, what if you just didn't really have 00:44:14.636 --> 00:44:17.876 to pick up the phone to email someone to ask them if they've seen the thing 00:44:17.876 --> 00:44:18.676 that they should have seen?

00:44:18.676 --> 00:44:24.066 Like the information flow, if you could undo that death by a thousand cuts, 00:44:24.066 --> 00:44:25.515 how much more efficient could you be? 00:44:25.771 --> 00:44:28.985 If you could leverage your insights from one project more easily to the next, 00:44:29.345 --> 00:44:30.477 would that make your life better? 00:44:30.773 --> 00:44:35.417 What's the shittiest part of your day and how can I make it go away is what I would say to them. 00:44:36.386 --> 00:44:42.486 Yeah, I like that. I mean, you know, Jack Dorsey, Square and Block and former 00:44:42.486 --> 00:44:48.006 Twitter founder, he wrote an article that was like the, it was the hierarchy 00:44:48.006 --> 00:44:49.368 of intelligence. Here it is. 00:44:49.673 --> 00:44:54.218 And, you know, he kind of made this point that in this age of AI that we are, 00:44:55.646 --> 00:45:01.306 our modern kind of management hierarchies and structures were built so that 00:45:01.306 --> 00:45:05.106 information could flow down and it goes all the way back to Roman times and 00:45:05.106 --> 00:45:07.658 the different... kind of organization of soldiers, right? 00:45:08.245 --> 00:45:15.106 And we are in this age where that kind of management structure is no longer 00:45:15.106 --> 00:45:17.491 required to pass information between people. 00:45:18.014 --> 00:45:22.440 And so how do we rethink what an structure looks like? 00:45:22.568 --> 00:45:27.026 And so he just kind of makes the point of how he's building block and how he's 00:45:27.026 --> 00:45:31.678 using AI to kind of be that thing that transfers information between people. 00:45:32.506 --> 00:45:36.426 And I think it's an interesting thought experiment to do at a project level 00:45:36.426 --> 00:45:41.496 of you have a supri of a pm px you have foreman you have all these different 00:45:41.496 --> 00:45:46.640 roles and if you could eliminate the friction between information flow what, 00:45:47.526 --> 00:45:52.526 what does that do for how how we organize our teams and and our processes and 00:45:52.526 --> 00:45:57.216 our workflows yeah i think there's there's a lot there what is what's the coolest 00:45:57.216 --> 00:46:01.123 thing you're doing with ai personally let's see me personally i'm. 00:46:02.332 --> 00:46:06.772 I think, you know, over time, I've built up like conversations in chat GPT that 00:46:06.772 --> 00:46:09.663 are kind of like interface with each other to say, you know, 00:46:10.070 --> 00:46:12.942 I mean, one thing I'm doing is I'm working on like an exercise program and I'm 00:46:12.942 --> 00:46:15.422 trying to make it optimal and kind of asking things like, hey, 00:46:15.422 --> 00:46:17.392 should I tweak, you know, the volume on this day? 00:46:17.392 --> 00:46:21.632 And so I think it's one of the coolest things is that you can build in like 00:46:21.632 --> 00:46:24.861 an understanding of who you are and you can kind of use it as like. 00:46:25.535 --> 00:46:29.282 You know,

you can kind of almost use it as like a store of brainpower to handle 00:46:29.282 --> 00:46:29.980 things that you don't have. 00:46:30.380 --> 00:46:33.092 There's lots of people who are going deeper than I have. Like I have not had 00:46:33.092 --> 00:46:36.382 time to figure out how open claw works and stuff like that. But I think there's 00:46:36.382 --> 00:46:37.676 some really cool stuff there. 00:46:38.592 --> 00:46:40.433 I think it's, it's kind of like, I'm trying to use it to get, 00:46:40.798 --> 00:46:44.742 to do the stupid parts of tasks that I don't like to free me up to do more of 00:46:44.742 --> 00:46:47.375 the stuff I actually like. It's just a general heuristic. 00:46:47.971 --> 00:46:49.550 That is a, that's a good rule of thumb. 00:46:50.148 --> 00:46:54.182 What, I know we kind of talked enterprise versus projects, but I think there's 00:46:54.182 --> 00:46:58.712 also, So, you know, there's probably midsize contractors, executives within 00:46:58.712 --> 00:47:00.450 that structure listening, thinking, 00:47:00.903 --> 00:47:05.895 am I big enough to warrant this kind of relationship with Palantir Foundry? 00:47:06.203 --> 00:47:11.512 Is there a model you guys use to say, you know, here's our kind of price structure, 00:47:11.512 --> 00:47:15.653 this is how it makes sense, and over time, where that investment pays off? 00:47:16.219 --> 00:47:19.722 Yeah, I mean, I would say that in general, we have a value-based pricing model. 00:47:19.722 --> 00:47:22.912 It's the only way that it would actually work for us to be able to do work with 00:47:22.912 --> 00:47:26.862 the U.S. government, which is kind of a big dog. They, you know, they have some money. 00:47:27.222 --> 00:47:30.752 And all the way down to, you know, smaller companies and everything in between. 00:47:30.752 --> 00:47:33.912 So we try to meet customers where they are and figure out, you know, 00:47:33.912 --> 00:47:34.947 what the path forward is. 00:47:35.156 --> 00:47:37.992 We're very flexible with how that works. There's lots of options. 00:47:37.992 --> 00:47:40.302 I mean, as you guys know, Capterra exists. 00:47:40.302 --> 00:47:44.812 So there are people who do services in Foundry that are not Palantir 4 deployed engineers. 00:47:44.812 --> 00:47:48.482 So that's one thing. But, you know, I would say that if you think that you have 00:47:48.482 --> 00:47:51.892 a problem that we could solve, then we'd be happy to have a conversation. 00:47:51.892 --> 00:47:54.412 And, you know, we'll be the first to say, hey, that's, you know. 00:47:55.392 --> 00:47:58.812 The juice isn't worth the squeeze there, right? We're not trying to license 00:47:58.812 --> 00:48:02.652 Max and get everyone working with us, even if it's not valuable to them. 00:48:03.012 --> 00:48:07.174 We want our customers to say that it was a beneficial engagement, 00:48:07.394 --> 00:48:10.342 they're happy with the value they're getting, and to find more value in the 00:48:10.342 --> 00:48:13.139 future. So if there's no there there, we're not going to push it. 00:48:13.798 --> 00:48:18.621 Yeah, it makes sense. How much interlap or interface do you have with the FDs 00:48:18.621 --> 00:48:20.941 on, you know, different kind of engagements? 00:48:20.941 --> 00:48:24.961 Like, do you have a construction vertical that's like, hey, these are best practices 00:48:24.961 --> 00:48:29.001 that we're seeing? And like, is there any information exchange within Palantir 00:48:29.001 --> 00:48:30.347 within construction verticals? 00:48:30.719 --> 00:48:35.011 Yeah, so there are kind of like loose structures that kind of emerge when there's 00:48:35.011 --> 00:48:38.672 a necessity for it. So like very much to your point of like the management structure, 00:48:39.311 --> 00:48:42.985 we try not to have internal structure for the sake of internal structure. 00:48:43.310 --> 00:48:46.831 So people kind of coalesce into teams of shared interest, right? 00:48:46.831 --> 00:48:48.255 They'll spin up Slack channels. 00:48:48.976 --> 00:48:52.541 So we, you know, you can sometimes like just say, send an email like, 00:48:52.541 --> 00:48:55.081 hey, I'm dealing with this problem. Does anyone in the organization have any 00:48:55.081 --> 00:48:58.321 experience with it? Like, hey, like I'm looking at BIM models for the first 00:48:58.321 --> 00:48:59.456 time, who's touched this? 00:49:00.054 --> 00:49:03.991 So there is that kind of like information sharing. It also gets encoded into 00:49:03.991 --> 00:49:08.771 the product by interface between our kind of product developers and our business 00:49:08.771 --> 00:49:10.847 development teams who are out on the job site. 00:49:11.500 --> 00:49:13.841 That kind of, you know, the product is developed at the edge. 00:49:13.841 --> 00:49:16.893 So there's a few different mechanisms by which information is ingested, 00:49:17.055 --> 00:49:18.843 shared, and pushed to the edges. 00:49:19.175 --> 00:49:21.992 But, you know, it's a radical responsibility. Like you're, you're, 00:49:22.178 --> 00:49:25.271 you could be here like three months and now you're at, you're the person responsible 00:49:25.271 --> 00:49:26.648 for making sure this is successful. 00:49:27.270 --> 00:49:31.449 Figure out how to make it successful right yeah we like self-starters and entrepreneurs. 00:49:32.218 --> 00:49:38.118 What what about you know you got these big pn software systems pro core autodesk trimble, 00:49:38.583 --> 00:49:44.831 what does how do those interact with boundary i mean obviously i'm sure there's 00:49:44.831 --> 00:49:49.111 integrations but do you guys take a point of view on like ultimately when you 00:49:49.111 --> 00:49:51.591 have a really mature ontology and, 00:49:52.331 --> 00:49:59.271 you're embedding the right kind of extractions from workflows so that the right 00:49:59.271 --> 00:50:01.215 kind of data is flowing into your system. 00:50:02.000 --> 00:50:08.407 Like what is the future of a PM

system? Yeah, I mean, it's unclear to me exactly 00:50:08.407 --> 00:50:12.327 how AI and PMs will interact in the future. 00:50:12.327 --> 00:50:16.240 I think there's a bullish case to say that, you know, at some point, 00:50:16.420 --> 00:50:21.597 most of what a PM does might be handled by AI or by like a better interface system. 00:50:21.597 --> 00:50:25.047 Now, this was also the promise of like the ERPs in the 1980s, 00:50:25.047 --> 00:50:27.903 and that didn't bear out. So I'm kind of agnostic there. 00:50:28.182 --> 00:50:31.397 Project managers have a lot of skill sets that are not just about collating 00:50:31.397 --> 00:50:34.748 information. It's about how to understand, you know, where can I push? 00:50:34.864 --> 00:50:36.803 Like, where is there slack in the system? 00:50:36.977 --> 00:50:40.627 How do I make this team work? How do I kind of, you know, there's a lot there 00:50:40.627 --> 00:50:43.157 that's not just like encoded in data movement. 00:50:43.570 --> 00:50:48.217 So I think, you know, there's a lot of software systems that I think will probably 00:50:48.217 --> 00:50:52.777 go the way of the Dodo because they were solving a specific problem in a specific 00:50:52.777 --> 00:50:56.307 field when the world was a little bit less interconnected. 00:50:56.775 --> 00:50:59.227 And I think the ones that survive are going to be the ones that figure out how 00:50:59.227 --> 00:51:04.321 to actually operationalize AI in a way that's not kitschy and actually makes people's lives easier. 00:51:05.055 --> 00:51:08.697 So, you know, our stance is that, you know, we're pro-competition, 00:51:08.697 --> 00:51:12.594 like let people try. We'll integrate with whatever works and we'll replace what doesn't. 00:51:13.651 --> 00:51:19.447 I like that. With what you know about the industry now and maybe what you know 00:51:19.447 --> 00:51:25.845 about kind of where at least the Palantir's kind of collection of technologies are. 00:51:26.877 --> 00:51:32.887 What do you see are the attributes in the field that will enable people to be 00:51:32.887 --> 00:51:37.042 more resilient and handle the change that is coming or is already here? 00:51:37.857 --> 00:51:41.227 Yeah. So, I mean, I think in the field, it's going to be like, 00:51:41.227 --> 00:51:45.437 first and foremost, you know, this, this isn't a put your head in the sand type situation. 00:51:45.437 --> 00:51:50.797 Like it's, it's coming, even if it's being forced down by kind of shareholders 00:51:51.117 --> 00:51:54.193 or, you know, the board, like people are going to want to use AI. 00:51:54.454 --> 00:51:57.697 And so I think, you know, one way to kind of protect yourself is to become opinionated 00:51:57.697 --> 00:52:02.517 on where the good uses of AI or the beneficial uses of AI are so that you can 00:52:02.517 --> 00:52:06.917 say, hey, like, I actually don't think that doing it this way is going to be is going to work for me. 00:52:06.917 --> 00:52:10.226 Like, you know, you have tradecraft, you know what you do. 00:52:10.534 --> 00:52:14.086 Understanding how that interfaces with it, I think, is going to be helpful because. 00:52:15.019 --> 00:52:18.154 We're in the stage where we're throwing this technology against everything. 00:52:18.450 --> 00:52:21.003 And I think there was a report, maybe it was Gartner or something, 00:52:21.003 --> 00:52:25.873 which is that like 90% or something of like AI pilots fail because they're just 00:52:25.873 --> 00:52:31.377 kind of throwing information and saying like, hey, magic oracle, solve my problem. 00:52:31.981 --> 00:52:36.893 AI guys are very fast, but they can be dumb in some ways. 00:52:36.893 --> 00:52:40.043 So I think, you know, becoming opinionated about it, learning the tools, 00:52:40.043 --> 00:52:43.623 right? Like, you know, you can use ChatGPT for free, get a feel for where the 00:52:43.623 --> 00:52:47.171 limitations are. play around with it, I would say is one way to do that. 00:52:49.399 --> 00:52:52.483 Have an idea about what you would do with it is one way to do it. 00:52:52.483 --> 00:52:55.743 Or you can also just choose to say, hey, I'm five years from retirement. 00:52:55.743 --> 00:52:58.559 I can probably ride this one out. That's completely up to you. 00:52:59.328 --> 00:53:03.933 But you might think about in that long career, what learnings can I encode that 00:53:03.933 --> 00:53:06.293 would be useful to pass on to the next generation? 00:53:06.293 --> 00:53:12.453 If there's some way to save other people from the pain I felt, how can I do that? 00:53:12.453 --> 00:53:15.942 And I think that's true, like definitely in construction in a lot of industries, 00:53:16.220 --> 00:53:19.403 you know, there's this element of, and when I've spoken to construction companies, 00:53:19.612 --> 00:53:21.553 a lot of the people in the field, like that's what they care about. 00:53:21.553 --> 00:53:24.600 They're like, we're a family in a lot of ways. So. 00:53:25.657 --> 00:53:28.255 Again, I'm kind of pontificating in a few different directions there, 00:53:28.255 --> 00:53:32.037 but I think you're becoming opinionated about it. Learn what it is, what it isn't. 00:53:32.642 --> 00:53:37.841 Yeah, I like that. What is, I mean, you went academic turned FDE. 00:53:38.161 --> 00:53:42.511 What are the other profiles of FDEs that kind of operate in this way? 00:53:42.801 --> 00:53:47.227 Yeah, I would say that there's a lot of people come from STEM undergrad. 00:53:47.482 --> 00:53:50.192 A lot of people come from a CS undergrad. There's a lot of that. 00:53:51.415 --> 00:53:54.685 There's a right tail of interesting people. I think for a while, 00:53:54.685 --> 00:53:58.561 our former CIO was an opera conductor in a former life. 00:53:59.328 --> 00:54:01.945 There's a lot of different paths to Palantir because, again, 00:54:01.945 --> 00:54:05.523 in a similar way that Kavanaugh wasn't precious about having a CS degree, 00:54:05.887 -->

00:54:10.148 yes, there's heuristics. Yeah, if you went to Yale, that's probably a good sign, I think. 00:54:10.835 --> 00:54:14.978 But if you can find a place here and you can provide value, we don't really care. 00:54:15.226 --> 00:54:19.458 We have a meritocracy fellowship. We bring people in before they've even gone to university. 00:54:19.661 --> 00:54:24.015 So we try to push back against that semiotic signaling but you know it's it's 00:54:24.015 --> 00:54:27.645 a history of having like been dissatisfied with the state of the world and wanting 00:54:27.645 --> 00:54:30.494 to fix it i think is kind of what the the through line is, 00:54:31.910 --> 00:54:35.712 i mean you're selling me on palantir a little more than i was ready for jesse, 00:54:36.294 --> 00:54:38.621 i wasn't ready to like you as much as i do uh. 00:54:40.119 --> 00:54:46.065 I think you know i i everyone probably has some narrative on you know all varieties 00:54:46.065 --> 00:54:50.302 of companies and and i think what palantir is solving for 00:54:50.713 --> 00:54:56.878 is you know best in class tech in theory right like like industry grade tech 00:54:57.292 --> 00:55:03.255 with the humility to go to the problem and and ask questions that may sound 00:55:03.255 --> 00:55:07.857 stupid at first and i think that's really powerful from a model standpoint because, 00:55:08.722 --> 00:55:14.715 it's normally we have the tool buy it if it doesn't work i guess you're just 00:55:14.715 --> 00:55:18.305 not very good at implementing tech and and i think that's a model that's been 00:55:18.305 --> 00:55:21.625 tried in our industry and and often fails and so it's it's, 00:55:22.293 --> 00:55:25.265 unique and i think it's healthy to see a new model coming up. 00:55:27.157 --> 00:55:30.479 So I normally end with like, what would you tell your younger self? 00:55:30.693 --> 00:55:33.126 But I feel like, you know, you look fairly young. 00:55:33.770 --> 00:55:39.138 What would, what would you tell your pre-Kavanaugh self, you know, 00:55:39.138 --> 00:55:44.789 coming from the academia world? What would you tell that person coming into construction? 00:55:45.534 --> 00:55:49.748 Yeah, I mean, I think there's, I think if I had to tell my younger self something, 00:55:50.268 --> 00:55:53.148 it would be that like, no one's going to give you permission to succeed. 00:55:53.148 --> 00:55:55.828 Like at the end of the day, like you have to go out and figure it out, 00:55:55.828 --> 00:56:00.358 right? Like academia, there's a very clear onerous causa, right? 00:56:00.358 --> 00:56:03.312 Like, you know, you do this, then you do this, then you publish these papers, 00:56:03.544 --> 00:56:05.650 then you make tenure, and then you do these things. 00:56:06.051 --> 00:56:09.988 I would tell myself that's like, that's comforting, but it's a constraint. 00:56:09.988 --> 00:56:13.802 So don't limit yourself to, you know, the path that you're on. 00:56:14.284 --> 00:56:19.218 But I think what I tell myself pre-Kavanaugh would be, things are about to get 00:56:19.218 --> 00:56:22.728 really fun. They're going to be hard, and you're going to be cold for a while in the winter. 00:56:23.268 --> 00:56:26.758 But you're gonna you're gonna make some friends and you're gonna really create 00:56:26.758 --> 00:56:31.462 something of value with the people around you yeah here's an amazon link for some good socks, 00:56:33.588 --> 00:56:37.727 good stuff jesse well i appreciate you coming on man yeah i was happy to be here.