Subscribe
Share
Share
Embed
The Punch List is the definitive weekly briefing for the leaders, investors, and developers shaping the future of the AEC (Architecture, Engineering, and Construction) and Real Estate industries.
In an industry that is "the second largest in the world but very small at the top," staying ahead of the curve isn't just an advantage—it’s a necessity. Join our three expert hosts as they bring their unique "different stakeholder" perspectives to the table to discuss the three biggest news topics of the week.
KP Reddy (00:00)
If you really wanna execute true project success, you have to have two things.
You have to have empathy and you have to have understanding. Understanding requires that I don't just have an understanding of the project for myself, I have an understanding how my decisions affect other parties.
If all we did was yell at our people all day, and had no empathy and no understanding of their point of view and how my decisions affect their lives, we would never be successful.
Jon Wright (00:31)
All right. Welcome to the Punch List podcast. We have an exciting episode today. We have a guest, KP Reddy, uh, who we will introduce to you and talk more about. We're excited to talk about, um, KP's background and views on the construction and architecture industry, but we also would like to talk about the recently passed housing bill and also the Kimi K3 news of the week and broadly Chinese open models and what impact that's going to have on data center construction.
Kanav, why don't we pass it over to you to introduce your friend KP?
Kanav Hasija (01:09)
I met him last year, guys, and, uh, been reading about him, and we have talked like a couple of times after that. It's very rare to find a civil engineer who knows technology, software technology, and who doesn't mind speaking his mind together. So that's good to have that blend with KP, and I love his personality for that. He's a civil engineer, uh, worked, uh, with Gehry Technologies, which is one of the innovative firms back in the day. He founded Shadow Ventures, a venture arm for investing in construction tech and prop tech startups. He has two hundred of those in his portfolio, so he has seen a lot of them. Now he's, uh, opened up, uh, Zero RFI, which is a very interesting thing we'll talk about, you know, today as well. So KP, welcome to the house.
KP Reddy (02:01)
Oh, thanks for having me.
Kanav Hasija (02:03)
So KP, let me start with, uh, the first question, which, I've never asked you before, so let's ask on the pod. You wrote that famous book, BIM for Developers and Owners, fifteen years ago, which became pretty famous. And then nine months ago, you wrote a very nice Substack blog article, BIM is Dead and So is Software. Tell us about that journey. How did you go from BIM is amazing to BIM is dead?
KP Reddy (02:29)
Yeah, I think, um, a lot of it is like most tools, right? Um, our industry has a really good way of not using them correctly, misusing them. You can take the most powerful tool, like whether it's BIM, whether it's before that, CATIA, like all these tools, and if you misuse them, you're not gonna get any of the benefit, right? And so I think fundamentally what we thought 15, 16 years ago about BIM was that BIM was gonna be the document of record, so to speak. We were going to build off the BIM. We're gonna collaborate in BIM. We will do all these things in BIM. And all BIM did is become a very expensive version of SketchUp, which is kind of free, I think.
So that's one thing, right? And it's interesting, I had this argument with, um, Arvind, the CEO of IBM, a couple years ago, on stage actually. And I was like, "CRUD is dead," right? "CRUD is dead." And he's like, "No, no, we're gonna adapt." So when I saw their earnings and all the other day, I was like, "Well, I mean, I told you two years ago. Why didn't you listen to me?"
Jon Wright (03:38)
Explain CRUD to our audience.
KP Reddy (03:41)
So basically, when you think about any software, web or otherwise, it's just a bunch of tables, right, assembled together to a user interface, right? In layman's terms, a bunch of Google spreadsheets with a pretty interface in there, with some triggers and actions, right? Which in the spreadsheet world we'd probably call macros, so when you think about that, um, and go back to first principles of compute, why are LLMs so powerful? It's because we built all these layers for us dumb humans to interact with the compute. And what's really happened with LLMs, the breakthrough is now the compute understands our dumb language, which is English, Spanish, or whatever, right? And so what that means is there's a collapsing of all the things in between, which means we don't need all these interfaces, we don't need any of this because it's all just tables. So I always tell people, "What's a Gantt chart? What is P6? It's just a bunch of tables. Just a bunch of tables with relationships." Nobody ever asked like, "Hey, I need to see a Gantt chart," 'cause that's how I think. It was because we had to create a dumb human interface called a Gantt chart so that we could understand mounds of data.
But ultimately, what you really wanna know is, is my project on time? Is it off schedule? What do I need to do? What decisions do I need to make? That's ultimately all we care about. No CEO has ever asked like, "Give me a Gantt chart." Who cares? Nobody cares about Gantt charts. Nobody cares about project plans.
So if you get back to first principles, it's all just tables. So if you do that, then you say, "Well, we never really got into BIM in an appropriate way." So the end deliverable in design is called CDs which people get wrong. It's not construction drawings, it's contract documents. Contract documents are the plans and the specs.
Here's what I'm gonna tell you, here's what my design intent is. Because unfortunately, most architects and engineers don't actually know how to build a building. They know how to talk about building a building. So in the world of BIM, we used to talk about it's a design intent model, which means it's for the architects and engineers, right? And then literally what's been happening with BIM is they use the BIM, create this plan, a set of 2D drawings, and then the contractor has a VDC department that goes and rebuilds all these things, which has created a nice business for all the people in India to do BIM drafting of the 2D. But the reason you have to do that is because the original artifacts were not built for how do I build the building? They were designed for how do I present to the customer, the owner and developer, their building. Here's what you're buying, right? Biggest purchase made, um, sight unseen is a building, And then how do I communicate to the contractor what my intent for the design was? It's up to the contractor to understand build intent.
In the world of AI, what do we call that? We call that inference. The contractor has to infer what the architect was thinking in order to do their job. Then the subcontractor has to infer what the GC is thinking, which is not a whole lot, because most of them don't self-perform, so when you think about that, then why do we need BIM, right? I'm gonna ask my, uh, AI, right, "Hey, create me a 3D model of my building," 'cause that's all I care about. I wanna see my building. Oh, give me a camera view, that I can fly through the building. Cool. The leasing office needs a cool video. Like, I use Runway or something and just generate it, right? So ultimately, if your core data set is just tables, and then you have a layer in there that allows you as the human to really gain an understanding of what's going on and present it to you the way you wanna present it, then what do we need BIM for?
Kanav Hasija (07:41)
I was gonna just ask this. I mean, I'm the dumb general contractor here. Playing that part, right? So we are, we're so used to having the CDs to help us build the building, which supposedly come from a 3D model which is coordinated.
Tanmaya Kala (07:55)
I feel some aspect of that, I mean, maybe long-term what you're saying, I can see that happening, KP. But I think right now some aspect of that still has to be delivered through the 3D model for a contractor to have a set of documents to build things with some water lines. I know we have a bunch of constructability issues that we still need to resolve.
I don't know how we bridge that gap based on where AI is today.
KP Reddy (08:20)
You start over. I mean, what we've gone from is like BIM slop to AI slop, right? So look, the fact that an architect gives you an artifact, a BIM, that they will have no accountability and liability to. What's the point? What is the point? It's just like AI slop. Like my team, we have a process where we have a corpus of knowledge where they have to human verify it to be true. And if you put AI slop in there, you are fired, and I will shame you on LinkedIn. That is not allowed, right? The human condition is you should review and be accountable to your words, actions, and writings, right? And I think that's the problem, right? But even if the architects and engineers cannot be accountable within their scope, I mean, they have to be accountable within their scope of work.
We can't expect a structural engineer in BIM to create shop drawing level BIM. That's not their scope. It's not their job, right? However, within the scope of work, it should be accurate, and because there is no liability associated with it, nobody's doing it.
Kanav Hasija (09:34)
The way to see that TK for me, right, is construction industry moves slow. They're gonna see this change slowly. Let's see which industries move fast. Software. Let's look at software, okay? A product manager used to write a PRD, a spec document on what the product will look like. A designer used to design some mock-up prototypes, uh, used to go to engineer. The moment it used to go to engineer to build that software, that design used to be outdated. It was like, "Hey, we need to add a few more things." So the design of software was always outdated, out of sync with what the actual build out of the code was. Then came Figma, which was the collaborative design software, which was like, "Hey, you all can collaborate. Engineers can give inputs to designers because designers don't know how to build. You both can collaborate together on Figma. You can give comments." The design would be more, you know, battle-tested, if you will. Great. Then AI came, and Claude Design said, "You don't need to spend time on doing those prototypes. You can just prompt to Claude, and Claude can help you design a prototype, which can go to engineering right away." And so now you don't need that system of design record, which is Figma in this case, in software, right? So software moved away from that already or moving away as we speak, uh
KP Reddy (10:57)
Yeah, and I think there's something else to understand, right? Like we always say, like I've been saying this for 20 years, the construction business is a lot less like manufacturing and a lot more like the movie industry. We show up, we do some stuff, and then we go on to the next project. And sometimes we work together again, sometimes we don't. Sometime the script is really tight, but sometimes the actors improvise, and sometimes that improvisation is magic, and sometimes it's not, right? And that's really what we look more like. And I think to the point of like back to like the human is getting... I mean, I have a four-year-old. My four-year-old will never need to learn how to use a keyboard, nor should they. That is a made-up interface to the compute. We don't really-- It's not natural for us. But now we'll just talk to computers, right? I have one guy on my team that writes all his code via voice. I think it's weird, but it's also magic, right?
Kanav Hasija (11:52)
Yeah.
KP Reddy (11:52)
So I think when you think about it, we all have to focus selfishly on our own workflows and processes. So if I'm the general contractor, I'm gonna use technology innovation smarts to constantly improve my process within my system, within my company, and that is all I care about selfishly. But if you really wanna execute true project success, you have to have two things, and this is against everything, whether it's construction or anything else. You have to have empathy and you have to have understanding. That drives project success 100% of the time. Now, what I'll tell you about empathy, that is a company culture thing. I cannot help you with that. Right? But understanding requires that I don't just have an understanding of the project for myself, I have an understanding how my decisions affect other parties. That is the big break. That is the big disconnect in our industry, right? The architect's not saying like, "Oh, if I move this wall, man, that's gonna make it really hard for the drywall trade. How's he gonna get up there? That he's gonna require scaffolding and have to tie off. That's very dangerous. Do I wanna do that"? Like, there's no thinking and understanding of how my actions impact others. And we all know if you wanna drive projects... We're all in startups, right? If all we did was yell at our people all day, which I'm accused of sometimes, but and had no empathy and no understanding of their point of view and how my decisions affect their lives, "KP, thanks. I'm, now I'm working all weekend," right? Those type of things. We would never be successful. So we know that success requires these two ingredients, and in construction, like I said, empathy, that is... And there's some great culturally, like in engineering and construction, some great, great companies that drive a great culture.
But it's the lack of understanding for all the parties involved on the project that's really required. And BIM, we thought that. We thought BIM was gonna be, "Oh my God, we're gonna do IPD, and we're gonna be in the model together, and it's gonna be kumbaya, and so amazing," until we said, "I'm not sharing the model with you."
Tanmaya Kala (14:11)
Hmm. Yeah.
Jon Wright (14:12)
When I was, like, the owner, developer, manager of projects, my goal was always to try to compress and integrate as much as possible all of these elements into one thing. And yeah, it never felt like, BIM was this, the magical social media that it was supposed to be. It was always like, how do we, as quickly as possible, iterate and get to that understanding or empathy, uh, real-time? We'd have a meeting, and then we'd have to have the follow-up coordination meeting and some prep that had to happen, but it never happened real-time. And I'm just trying to understand what do we think LLMs are gonna do, to help with this to move faster? Because as the owner, we just wanna move. You just wanna iterate and get us to that outcome of like, what's this impact gonna be and how much is it gonna cost, and can I do it as fast as possible?
KP Reddy (15:05)
Yeah. For our industry, LLMs are such a minor part of the equation of the AI stack, right? I think that is the human interface thinking around it. But I always try to tell people, one, we're not in the business of building buildings. We're in the business of building society.
We just came out of the World Cup games. None of that would have existed without these amazing facilities and getting people in. And I mean, that-- this is what we do for a living, and it's very important that we think about that, and that is, that has to be everyone's North Star. We are shaping the fabric of society, full stop.
Now, how do we do it is a whole nother thing, right? We do it through change orders. But besides that, I think to your point, TK, is that, um, how we make these decisions, I always say like visualize a job site, right? Visualize a job trailer, and the superintendent rolls out a set of plans, and every party is looking at that set of plans. What are they looking at? The drywall guy's like, "Man, I don't know how many crews I'm gonna need. How much mud am I gonna need?" Right? Everybody, it's the same exact artifact, that everyone's trying to gain an understanding of. Selfishly, what, how does this, where does this matter to me? So I think in the world of LLMs, if you think about it being the translation tool, much like a language translator, if you're the owner, what do you care about?
Do you care about how many drywall screws are being deployed? No. What do you care about, right? So I think the ability to interrogate a project corpus of data around what you need to understand is what's important. Because, you know, it's like you ask people questions. I was just on a call with a big client this morning. People ask questions, and then they answer, not incorrectly, but without context to the person asking the question. I think I published something on Substack the other day, which was, you know, I went through a divorce and a friend of mine's like, "Oh, you gotta read The Five Love Languages. It's gonna rock your world." And then I ended up doing it, right? And so I wrote like, what are the five love languages of construction? Because the whole point of that book is you need to speak to your counterparty in a way that they listen, their language, I kind of feel like we have the same issue in construction. Like, if you're the developer, I don't... Don't show me the sheer wall detail. I don't care or understand that. Tell me how it's gonna affect X, like visually. Like it's a school, I'm worried about it being, you know, the students and like, whatever it is, right? Whatever you're, what's important to you. So I think that's where LLMs start to crush all these layers of nonsense in between to really get you the right information at the right time that is important to you to make your decisions. And with an understanding of how you make your decisions will affect 12, 20 people, could affect the safety. I mean, you could make a decision in your office that ends up killing someone. That is the harsh reality of our industry. When you look at construction being the highest level of suicide rates. Just know that when you make that decision that drives someone to work a late night, this is the net effect of your decision.
You thought you just changed a detail in your drawings. That's not what you just did. You just made someone work late. And that person now work late is stressed out because he's trying to get to his kid's ballet recital, and is tired and all of that, right? So I just think that's where we have to think about these things.
Kanav Hasija (18:53)
One thing I wanted to say, KP, I love the five love languages thing. So Jon and I worked on a project together where Jon was my client, and we've been through situations where I've slammed the door on him. Like, we've like we've been through the whole ordeal and, um, I guess we are okay now.
Tanmaya Kala (19:11)
I don't know if other industries are like that, but construction tends to get kind of emotional during a project. You're, you know, dealing with so many things and that can happen. But to a more serious note, one thing I did wanna pull on, like, on the trade-off is, and I think this is what you're, what you're saying, right? I think of preconstruction as this journey for a owner having an idea to getting to construction document. I'm just talking preconstruction, right? As a contractor, to me, we only make money if you're building the building. Preconstruction is kind of means to an end. Within preconstruction, we never would get a BIM model without someone making me sign a piece of paper saying, "You can't hold this against me. You have to use the documents and all that, right? Someone would make us sign that piece of paper anyways. I always looked at BIM as well as a means to an end to the construction documents and not the end itself. I think what you're hitting on is, and I don't know if this is what you're saying, as LLM progresses, the journey from idea to construction documents, which allows at least the contractor the infrability to build a building, could be made through prompting LLMs.
Is that where you're heading?
KP Reddy (20:25)
No, I think buildings are too complicated if just in the world of pure LLMs, right? Um, I remember my first job out of college as a civil engineer, I would get mad that my clients would not respond with feedback. And my boss came to me and said, "Go over to their office, roll out the set of plans."
I'm like, I couriered them to them. Like, why?" And he's like, "Do you think your client actually knows how to read plans?" Like, "The reason they're not responding is they don't know what to respond with. Go over there and explain exactly what you did,"
KP Reddy (21:01)
why you did it, uh, et cetera," right? So I think more so is just the ability for people to have a voice and to understand what they're looking at. My book on BIM was to make the point of, like, this is about the owners and developers, this isn't about us. This is about the end customer that's really doing the things, and how do we use BIM in a way that benefits them? Which has never happened, right?
I mean, I helped create the GSA BIM standards, the US Army Corps of Engineers BIM standards, and they quickly turned it into kind of like an IT exercise. Check, check, check, right? Did you create this file system? Like, that's not what... That was never the intent. Another thing I like to say is it's practically criminal that we design and build buildings without being able to tell an owner how much it's gonna cost to operate and maintain it. Could you imagine going to a car dealership and saying, "Hey, how many miles per gallon?" "I don't know. Tell me how it goes," right? "Well, what's the maintenance schedule?" "We don't know. Tell me how it goes." Like, that... It's insane. You know, when you talk to civilians outside of our industry, they're literally like, "Oh my God, you guys are morons." Like, that is insane.
Kanav Hasija (22:23)
So this actually the same point, right? You guys talked about, uh, empathy, understanding. All of this is coming because over decades, fragmentation happened. There's a fragmentation between design and build. Then within design, there was fragmentation in design and different engineering fields. And in build, there was like, many GCs don't self-perform many trades. So much fragmentation happened from three companies when Empire State Building was built to Salesforce Tower took, like, forty-five companies to come together. You guys see with AI or with any incentive design change, this fragmentation coming back to consolidation? Or what is the new structure of the build environment? And KP, tease us within that, your Zero RFI story as well.
KP Reddy (23:15)
With your RFI, right, I've been running our venture firm. We still run it. I've got a great team there. And really it all started with a conversation with Hemant and Paul Kwan over at General Catalyst with the idea that like, "KP, all this technology, all this stuff, it's going nowhere. There has been no deflationary effect to construction. It's not gotten better, some would argue worse. You know, why do you think your portfolio is gonna really make it any better? And what's the real question?" Right? What's the real thing?" And so I walked away and Hemant, if you haven't spent time with him, Paul also like super smart, um, just very deep understanding of things, right, of the world.
Paul's like, "Hey, go write me a theory of change for construction. What's your theory of change for construction?" So I wrote like a five-pager, the history of the master builder. Why did we get fragmented, what happened over time in the fragmentation, and why are we here?
And the theory of change is to go back to the master builder, and that's when they sent me a term sheet. I always say if you want a great Silicon Valley story, you write a five-pager, you get a term sheet, right?
Kanav Hasija (24:26)
That's typical Hemant and Paul's though, but yeah. Yeah. Yeah,
KP Reddy (24:30)
I give them-- I always say they're like more like my co-founders than investors, right? They're my inactive co-founders, so to speak. But I think that's where we're heading. Now, what I would say is that's just not my point of view. That's actually everyone's point of view, and everyone is approaching it differently, if you look at DPR, Suffolk, and some of these other companies, they're now starting to buy architecture and engineering firms. They're now starting to buy self-performing, so fundamentally, we think it has to get back to the master builder. My approach, um, which, you know, y'all know, like I know all these CEOs who have had this conversation with them, right? They agree that's what I'm doing, right? I'm creating the master builder with Zero RFI, except that we all get to keep our companies, I might consolidate some stuff.
We've bought three companies. We'll keep buying companies. My team wants me to go buy general contractors. I'm like, "I'm not interested in that right now." Might buy some subs, I don't know. But generally speaking, we don't think we need to create the next Autodesk Procore.
We don't believe that, we believe that it's the orchestration layer. So our phase one of orchestration layer with Zero RFI is to be that lingua franca across all the processes. So right now, if you go into our system and you're janitor, and you wanna understand how this design is gonna affect my work, it will tell you.
I can tell you as the janitor, like, "Here's how many mops you need," "This is how this is gonna be," right? So what we've built is this understanding layer. I mean, to me, like, object detection, uh, like all these estimating softwares, I don't understand how they're raising money. Don't get it. But I just don't think object detection is important.
It's not that it's hard. We've been doing that for fifteen years. Um, but I think what's important is the context of how these plans, specs, and other artifacts affect me. How does it affect me? If I'm the drywall guy, how many screws and how much tape and how much mud do I need to go on this job on this day?
I can tell you that, right? I can tell you that. And so the idea is to really create this model of understanding that serves all parties involved, including the owner, right? I had an owner today that's like, "Hey, we wanted to move our ribbon cutting." And I was like, "Well, let me tell you, why don't you ask our model? And it might tell you that the ribbon cutting needs to be at the signage at the front of the building versus in the parking lot, versus actually in the building. Depending on the schedule, I'm not sure where you're gonna put the ribbon cutting." And so I think when you think about these things and what we're doing with Zero RFI is just this idea of being the orchestration layer.
So that's phase one. I 100% believe we will have automation on the job site.
So as we build this orchestration layer of, um, orchestrating what I call meat robots, people, that extensibility into actual automation. If we can't get people to orchestrate themselves and understanding in a deep way, right? In a two o'clock on Tuesday, how many drywall screws are you using? If we can't do that, we will never be able to drive motion control and orchestration of robots on the job site. So that's kind of our phase two approach of how we look at stuff. But ultimately, if you start saying that, well, if the robots can do much of it, why do I need a GC? Maybe we do become a GC. Maybe we become a design builder. Maybe we become design build, operate, and then transfer. Maybe we... I mean, I don't know. All I know is if you keep knocking down these big problems, it just keeps opening up capital markets. That's how that works.
Kanav Hasija (28:02)
TK said this to me, one year ago. He's like, when mobility came, it didn't add productivity to folks on the field. It added slop. Were sending them hundreds of top pages of documents. They just need to know what do I need to do today at 2:00 PM. That's it. Not those 100 pages of BIM documents or construction documents.
KP Reddy (28:22)
Right. No, 100%. And I think, look, I think these automated systems, right? Name your flavor, the cognitive load of automation that's happened, like people don't know what emails they need to actually read and so what it's now come to is people just don't read them. They wait for the phone call. If it's that important, they will call me.
Tanmaya Kala (28:47)
I mean, uh, KP, to add to that, right? The reason I was telling Kanav that was, uh, I had a motto at our job sites that we are not USPS. So our job is not to get all this information from design team and architects and then transfer it to the foreman to, "Hey, you sift through the data," right? Our job is to filter it, to make it accurate for what they need to do that day.
And yeah, that's hard and that takes time. And my biggest fear with AI is actually that, and I think you hit on that too, that let's not generate a bunch of slop that we leave it to our frontline troops to like sift through, because that doesn't allow us to build a better product.
Jon Wright (29:25)
Think you're spot on here. It is the end product. It is for the owner, whoever that may be, or the user, that's interfacing with the building, if it's an office occupant or whatever. Being able to query a corpus of information, a project file that has everything in it, and have an orchestration layer tell them their simple question like, "How am I gonna run my conference calls?"
You know, uh, and it be able to say exactly what that's gonna be, uh, that's very important. And taking out the layer upon layer of middle management that has to translate that down is-- that will be very efficient.
KP Reddy (30:12)
Yeah. I wrote a piece last year, about the joy of construction, and if you're a school system and you just get approved to build a school, so much excitement, so much enthusiasm, right? Like, "Oh my God, we got approved to build a school. It's gonna be amazing for our students. It goes to be amazing for our community."
And then they have their first meeting with the architect. Still excited, right? 'Cause they're watercoloring and creating all kinds of animations like, "Oh my God, amazing." And then they get to actual drawing production and having to approve things, and then putting it out. Like, the joy of these projects just quickly declines.
And it's such a shame. It's such a shame. It should be fun, right? It should be fun.
Jon Wright (30:54)
Yeah, this is not going to, as you talk about it, you know, in this context, AI is not gonna eliminate the roles. You're still gonna need those subject matter experts and engineers and designers. Maybe it eliminates some of the drudgery and the coordination that ends up only happening with small groups and like, you know, that's suboptimal with not all the information.
It is. I will tell you, these big projects, they end up becoming just drudgery and not as fun after the concept plan.
Kanav Hasija (31:24)
Jon, what you said like, you know, reminds me of this conversation I had with this pre-con guy in Miami, when I met him in person. Came from ops, like building buildings to like, you know, estimating buildings. And I asked him, "Why did you move from ops to pre-con?" And he's like, "Well, ops is gratifying.
I can tell my kids I built that building, but that's only like three buildings in 10 years, that's the maximum gratification I have. In pre-con, I can estimate like hundreds of buildings, right? So I wanted to see more of that. That's why I moved to pre-con." I asked him, "How is it going?"
He's like, "I'm estimating the same multifamily apartment again and again, and it's like crazy. But new thing came in last week where this owner wants to... You know how the Miami market is. Owner wants to have his penthouse, residents to be able to lift their Ferraris from the ground floor all the way to the penthouse, they want a car lift with the glass exterior. Now that's fun. Now I want to estimate that new stuff, right?" So those are few joys that these people get.
Jon Wright (32:28)
Well, you can ask, here's the five phases of a project typically. Enthusiasm, disillusionment, search for the guilty, punishment of the innocent, praise and honors for the non-participants. On a multi-year project, you'll have about three or four cycles of this actually. And always tell my teams, those first three elements, enthusiasm, disillusionment, and search for the guilty are always going to happen and there's gonna be multiple cycles. We just don't want to be the innocent that gets punished. So, do everything we can to be really hygienic and really responsive, uh, in terms of our project planning so that we don't get punished.
KP Reddy (33:13)
And Jon, it's funny you say that. I used to teach at Georgia Tech, um, and I used to tell my students, like, "Here's the process: design, build, operate, litigate." That is the four phases of a project. And they used to all laugh at the last one. I'm like, "Don't laugh. That is reality." Like, that is how it is, right? So and what that means is when we all, like, build an experience level and start to understand these, like, things, we operate, we design and build differently because we're optimizing to litigate, right?
Kanav Hasija (33:50)
Talking about litigation and law, let's switch to the next topic.
Jon Wright (33:54)
Yeah, absolutely. So, a bipartisan bill has, you know, passed, the 21st Century ROAD to Housing Act. There was just broad consensus in the federal government that we needed to have a bunch of reforms and spur housing. Housing and demand for housing and supply of housing has become a crisis across the country, and there's real consensus across both parties that we need to do something. It's really driven by our aging housing stock, millennial family formation, et cetera. The first compromise version of this bill had, like, 99% good stuff and 1% really controversial stuff. The first version out of the Senate a while back, essentially banned single family rental housing providers, large institutional providers from operating, because it said that if you had 300 single family homes that you were holding out to rent, uh, you were not gonna be able to sell those anymore to other investors. Also, that version of the bill said that if you were building things to rent, what we call build to rent, actually, I develop single family housing build to rent, that if you built those, you had to sell them after seven years. So the whole marketplace for that, um, really reacted. All of the industry reacted. The practical effect while this was being negotiated is the capital markets froze, until it was sorted out. So that was kind of the controversial one.
And the final version of that, of the bill with respect to that element, the compromise was that the 300 house cap remained in place, but the industry can continue to operate if it does two things if you're a large... You know, Invitation Homes is a large REIT, for example, owned by-- It's been owned by Blackstone. It's been public. It's been private again. If you, uh, wanna sell parts of your portfolio, you have to comply with two things. You have to, uh, report rent to the credit bureaus essentially so that people get credit, for-- and build credit through rental payments, which I think is good policy. Um, and the second is, if you're gonna sell the home that someone's living in to another investor, you have to offer to the current tenant a right of first refusal. A lot of the major REITs were doing that already, but now this codifies it and is good policy. Um, so industry continue to operate. That's kind of the controversial part. Let me talk about the most um-- kind of the biggest parts of the non-controversial part.
So that would be with, uh, manufactured housing, and I wanna share my screen for a second.
So manufactured housing or mobile homes, as they call it, um, since the late '70s, the regulation around that has been codified in what the industry calls the HUD code, um, and has required a steel chassis on the bottom that can have wheels on it so that it could be mobile in a sense.
And it was not a real property interest, it's a chattel interest, and that has defined how the industry is operated. Modern mobile homes or manufactured homes are typically not mobile. They really rarely ever move. They're on a pier foundation. Versions of them actually are on permanent foundations these days. And so this bill allows for the removal of that permanent chassis. And why that's important is there really is a stigma in the marketplace, with manufactured housing when you see that element. And consumers, you know, consumers and home buyers are reminded of what in the early days of this industry was really subpar product. And so why that's relevant? Since the 1950s, the sale of manufactured housing, we had a large bump in the '70s, then you had the enactment of the federal legislation. And aside from some peaks throughout the years, if you look over the last 20 years, those starts, uh, those shipments have largely remained the same.
And if you compare shipments of manufactured housing to overall US housing starts, it sits around nine percent a year. So for the industry, it's a pretty resilient. Um, so, you know, that's what attracted Warren Buffett and Berkshire Hathaway to buy Clayton Homes. It's a very resilient, great cash flowing business. But, like, there's kind of been a ceiling. And so the industry believes that the changes in the regulations that are beyond just the chassis requirement, there's also great changes around financing with the agency financing in the government, uh, for manufactured housing, that's gonna lift those shipments up over time. And that not just manufactured housing, but also modular, other factory-built housing is going to be increased through this. And so there's just great fanfare there. And what is appealing about, um, manufactured or factory-built housing is just it's more affordable. And with construction costs where they are today and the modern code requirements for building housing, it's just become difficult and people can't afford it.
Kanav Hasija (38:55)
And are there any restrictions to import them from China or somewhere else, or just you have to build them in here?
Jon Wright (39:01)
Well, the-- where that restriction is going to come, uh, in terms of import is at the state inspection level. So if you're building off-site, a lot of the states have a regulatory regime where they have to have an inspector there or a third-party inspector, and that's typically been all domestic.
And so, for us to open that market, I think there's kind of consensus that people don't wanna do that, you're gonna have to develop a regulatory regime that allows for inspections in Korea, or Vietnam, or China, or wherever it's being manufactured. Do I think that's on the horizon potentially?
But I think we got a lot of room for growth domestically because right now these manufactured housing providers, whether they're providing what they call a HUD code product, which I showed you with the chassis, or a modular product, which is a state code, they just can win out on cost and time on a simple single-family house or a small multi-family duplex, for example, to site build.
There's also, uh, in this bill lots of changes to the federal environmental review, regulatory scheme NEPA. Um, so if your project touches that, there's exemptions for in-residential development. And so now we need to see companion exemptions across the states 'cause California just passed a, for example, California Environmental Quality Act residential infill exemption last summer, uh, Buffy Wicks did.
Industry's happy with this. Uh, everyone is cheering and lauding this. It is-- bipartisan consensus and passage of this bill, which is very rare these days. And so I think the rub for me as an operator right now is it's gonna take like twenty-four months for them to implement the regulation.
We wish that it could just happen now. Um, but I can tell you from my inside information on just the large manufactured housing provider, they're already designing and engineering a product without the chassis that they can perhaps stack, um, and gonna be ready to take advantage of this as soon as this happens.
And I know the Trump administration is gonna prioritize certain elements of the regulatory scheme here to make it go faster, for example, so.
KP Reddy (41:04)
Yeah. So it's interesting. We've made a lot of bets in this market. My first investment was a company called ICON that has 3D-printed houses. It is doing really well. More recently, we invested in a company called AUAR that does on-site robotic panels. Um, in our fund, actually, we have one of our-- Berkshire company as one of our LPs.
We also have Builders FirstSource. So we've been spending a lot of time, um, in the residential space. And I do think, like, to your point, like, let's see how it ends up. The bill passes. It's always about implementation, right? And I'm sure there will always be some implementation confusion 'cause that's just how it is, but, um, yeah, I've spent a lot of time with the folks at Clayton Homes and all of them, um, through that Berkshire network. And I think everybody's really excited about this. And fundamentally, you know, I've always said, you know, twenty percent of a construction job ends up in the dumpster. If we reduce that by half, everybody gets a free house.
And so I do believe in the world of manufactured housing, you know, and if you go way back, you know, um, when you could buy houses from Sears, it's a pretty efficient thing, right? I think it's gonna be some really great entrepreneurial opportunities 'cause now you have a lot of systems that you can buy off the shelf, whether it's cold rolled steel, whatever it is, right?
Um, that I think there's gonna be a great opportunity for entrepreneurship to build the components that go into this main... 'Cause you can't ship this stuff all across the country, right? So there has to be a, uh, local distribution model. And I think that's gonna be really cool for people. Um, I'm excited about the entrepreneurship angle.
I think, uh, if this continues to get implemented and implemented well, I think you're gonna see a lot of the younger generation that now wants to work with their hands, apparently. Maybe they all become like industrial home builders.
Tanmaya Kala (42:59)
They start buying houses from Sears again, which came with IKEA style instructions and yeah,
Jon Wright (43:06)
I think it's a great idea. I think there was a lot of genius there, whether it was the Sears model or whether it was, um, the model we mirrored, which is we call the GI Bungalow. The FHA back in the '50s had a minimum property standards that if you met that, you had access to financing. It was very simple houses.
And those houses are well-built. They still stand today. People lived in them forever. They didn't call them starter homes. They were three bedroom, one bath, or two bedroom, one bath. You could add a basement, you could add an upstairs bungalow, unfinished, or a detached garage. And that allowed people access to home ownership.
People stayed in there for years, raised children, everything. And so if we, you know, at least my development, we're kind of evoking that and hearkening back to those first principles. I think we need more and more factory-built and industrialized construction to ship houses faster and cheaper.
KP Reddy (44:04)
I think there's also this other secondary effect of all of this, which is, um, from an economic development perspective, when you think about workforce housing, 'cause I grew up in the South, right? So Kia builds a plant in the middle of nowhere. Everybody's driving an hour and an hour and a half to this plant, um, versus being able to develop workforce housing that's maybe two hours from a city center for a factory, and you can actually build a real community there, right?
You know, the best definition of a community is always where people can, like, everyone can live and work there. And so I think, um, there's gonna be some secondary effects of how we think about, um, you know, that data center in the middle of nowhere. Like, why are people driving for an hour and a half each way to go work at it?
Whereas you might be able to deploy some workforce housing, uh, in a much better way. That's nice, right? That's nice. Where unfortunately, like if you're in the fracking business, there's workforce housing, but no one's gonna bring their kids and wife there.
Kanav Hasija (45:06)
Let's talk about one last topic, which is, uh, if you're in the world of AI and kind of live and breathe AI every day, last two months, a lot of token maxing was happening, and people were like, you know, spending thousands of dollars a month per person on that.
And then, a new Chinese movement happened last week. Kimi K3 model came out, and it's not best as compared to all the frontier models, but it's good enough to do a bunch of tasks. Not the most heavy task, but it can do a bunch of other tasks. So, one thing is, is China closing on the race?
There has been some speculations of distillations from US models and US government trying to block Chinese models. But the bigger question is, irrespective of that blocking or not, has this at least made the industry aware that seventy, eighty percent of a task can be done by smaller models, uh, or open source models, and then that mix will change, the investment structures will change?
And does that mean we don't need to quadruple our data center CapEx, uh, development like we thought we have to?
KP Reddy (46:12)
I think one of the things to consider with all this is like, I think this LLM movement has made technologists a bit lazy, and we've all forgot how to think about systems engineering and memory and process and all that. So I think, um, what it's doing, which I love because I came from a world of making my own routers, right?
So this is exciting to me, which is now we have to start to think about the systems engineering aspect of how we think about token usage, best use. Back in the world, we used to do like least cost routing, right? You start thinking about least cost routing in the telecom space. Now we're gonna do least cost routing around tokens.
And I think some of this is interesting. It's timely from, you know, the Alex Karp rant a couple weeks ago. So I think that's also kind of, you know, it's an extension of that narrative. And so I think if you're a builder, you know, in the startup space, if you don't understand how this stuff actually works, you're about to get very much commoditized, right?
You have to be able to create some level of your own models. You have to think about governance and sovereignty. And, you know, one of the things that we really have done well at, I would say at Zero RFI is like the auditability of everything, who asked it, why they asked it, and what the output was, and having a full audit trail around everything.
Because at the end of the day, our industry is a very risk-averse industry, so they need to know exactly what happened, right? Who asked the question and what was the response and how was it validated becomes a massive conversation. But that's my two cents.
Kanav Hasija (47:50)
Yep. Oh, thanks, KP.
Jon Wright (47:52)
I think at the end of the day, do we think this impacts construction demand around data centers? That the adoption of these open models, does that change, you know, what we've been saying for the last six months about demand?
Kanav Hasija (48:10)
What this news is saying, it's not saying that the number of tokens will not increase. The number of tokens will still increase. We still have a lot of gap between what AI can help us do and what we're using AI for now. So that gap still needs to be filled up. AI can do much more than what we are doing now.
So number of tokens will increase. The question is, the percentage mix of proprietary or expensive tokens from Claude and OpenAI will reduce. Does that mean Claude and OpenAI will slow down? No. There's still a huge growth curve to them as well. It's just that the mix will change. And as I think KP was saying, we get pretty lazy, as engineers who are building the systems.
Even the users get lazy. The users will use a very heavy model for a very simple task, while they can use a very simple model for that simple task, right? So now do you let users do it, or do you make a system, as KP was saying, smart enough so that it routes itself to a weaker model?
That's where the industry is going. And once that happens, I can see a world why open source models come in. There's-- I think the second point is if the task is predictable and if the task is repeatable, I can use open models or smaller models. But if I don't know what the task is, if I don't know what the outcome is, I still don't know what will come in on the way of this road to answer this question, I need to use a good reasoning model because I don't know what's gonna come up.
But if I know it's a simple task as for like construction, I got thousand lines of invoices or invoice codes want to classify them to my proprietary cost coding system made for this project. It's a very tedious task. It's a simple classification system. You need a very small model. You don't need a big model for that.
But if you're like, "Hey, will the date for ribbon cutting move?" we have to answer a lot of questions to get there. That's a heavier model, right? So that's I think where the world will go. Which means we still need data centers because we still need tokens.
Anthropic grew 10x from last year. Will it again grow 10x? Maybe not. It might grow like 5x or 7x and then grow 5x the next year, and then grow 3x the next year, right?
So it'll still grow and double, triple, you know, quadruple.
But the AI is growing 10x overall. The AI tokens is growing 10x to 20x every year. Just that the open models and the light models will eat more share in that. OpenAI said, "Hey, we are investigating with Hugging Face. Someone hacked Hugging Face." They are now creating these rumors that Chinese might have hacked Hugging Face and leaked the test of these benchmarks, right? So, these benchmarks over time, people know kind of reverse engineer know what these benchmarks want. So models are getting biased towards benchmarks, by the way.
I'm trying to say is that you might score really good on the benchmark, but in the real life you might not score that well, right?
Jon Wright (51:25)
Well, today has been a great episode. Uh, learned a lot. Learned a lot from KP, learned a lot from your views, Kanav, on the models. Uh, and, um, thanks, and look forward to seeing you guys again.
Kanav Hasija (51:40)
Thanks all. No, master builder. Let's do it.
Jon Wright (51:42)
Talk soon.