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.
TK: data centers are great for blue-collar workers because it's creating jobs. the blue-collar workers are probably the last ones to get impacted by it. it's automating mostly white-collar jobs.
But I don't wanna minimize the energy consumption side that these data centers will have. there is other infrastructure that needs to be coupled with the data center build-out. we need to have a better infrastructure to allow for goods to move from point A to point B faster.
Jon Wright: All right. Welcome to the Punch List podcast We're going to cover the oil crisis and how it's gonna affect the construction industry, AI data center build-outs, will the music stop? We're going to talk about robotics and adoption on the construction site, and you won't leave the last one, the best one for last, AI Corner with Kanav to talk about the latest,
TK: Yeah
Jon Wright: I think starting from
Kanav Hasija: hot topic of today, we are hearing that Aramco has very few days left of-- oil reserves, and the market's gonna into twenty twenty-seven. Saudi is raising the alarm bells that there's gonna be oil pricing shocks, continuing till twenty-seven. Which means it could impact a lot in construction because construction materials, the manufacturing of that, and then materials international borders, all dependent on oil. So it's big news that the industry should be aware of. So let's it off. What do you guys think?
Jon Wright: Yeah, th-this is obviously gonna be a major impact. everyone knows this, that the oil is the, fuel for the whole economy. everything is so globalized, whether we like it or not, everything comes from overseas. So when fuel prices go up, all of these things are impacted.
TK: And then in some of the other countries, it's even more existential because the way they produce things, like if you're getting stuff from China and India, China and India are very fuel-hungry countries. even though these countries are looked at as manufacturing and services hub for the world, they don't have oil capacity.
Russia and US and, Middle East is blessed with oil, so they have that, but not these countries. So it's very hard for them to manufacture things as well, which can also lead to lead time increases, I remember during COVID days, it took us around 19 to 20 months to get a generator for a building we could build in 16.
I could have a building ready with just wind blowing through it for two months. it was outpacing construction. And it won't surprise me if we're back there here
Jon Wright: No, unquestionably, this is going to have an impact. what I think about is if we look back at the last decade, we've had four or five of these big geopolitical events that whether it's, tariff round one with Trump, COVID, it was tariff round two, and now we're at this. if you're planning a project and underwriting a project or buying out construction, the question comes, what are we going to carry in terms of potential increase due to these inflationary pressures and material cost increase associated with this ex-insert geopolitical event? so you're either gonna put it into an allowance, or you're gonna give, some type of contingency to the contractor, or you're going to carry it as an owner contingency. And what worries me about this is what I call the silent padding that happens up and down the chain. So the owner level, we're putting in, a contingency.
And then at the general contractor level, they're thinking about it in their, morning staff meetings. And then the subcontractors, and then the material suppliers, and then the subs to the subs. So you get this padding effect up and down the chain. And, what happens is If the event is transitory in some way and it passes, those prices just seem to stay where they were, when we inserted this.
And so I think an owner, and developer, you need to really have a deep conversation with your contractor and your subcontractors to think about what we really want to do here. Make sure that anything we add into a budget or an underwrite is very precise, targeted. We think about the building materials and what we're building here and how this is going to affect it, and try to avoid that silent padding up and down the chain, which I think is probably solved through communication. you let everybody know, "Hey, this is how we're carrying this right now. if circumstances change, we'll deal with it then. But, we don't need the extra pad, because it may even kill the deal at times."
Kanav Hasija: every pre-con room, every budget room discussions are gonna start with this. Adding 10% for unknowns. That's what's gonna happen in every meeting. think to avoid that, what's important to John, what he said, let me emphasize this even f-further is Can we substitute building materials the ones that are built locally? and can we look at the materials that are either heavy in weight, which is gonna have a huge impact on transportation, or have a dependency of energy its manufacturing, which is gonna again impact the raw prices. Can we look at those and just add some padding to them everything? So you need to be more surgical to check for price escalations and allowances than just having a blanket allowance.
TK: o-o-one thing I would add, just, being worked as a GC, for such a long time, I hear the silent padding argument all the time. my issue with that is,
Most GCs are operating at a razor-thin margin. they don't have the capacity to absorb the impacts of these things going up. there is a real issue where they do want to protect the meager margins they have.
Subcontractors have bit more room to play with, but it's not huge. if you're talking numbers, GCs can be anywhere from 1 to 6%, especially medium to large. subcontractors would be around 10. but again, these price increases can gobble all that up. Every GC has a horror story, every sub has a horror story where, as John says, "Oh, let's have open communication.
Let's be, like, all open about where things are at," and you expose the numbers, you sign a deal, and once the deal is signed, if the numbers go up, no owner will come back and pay you extra money if they have transferred the risk to you. That's a fact. Yeah. And that's business. So got to let contractors and subcontractors also do the business.
unless you say, "You know what? Give me the most precise number where you think it's gonna be at. Do-don't do 10% for unknowns, do 2% or 3%, and if it goes beyond that, I will fund the difference." If we can work something like that out, great. But I've not seen that happen.
Kanav Hasija: so tell me, that's interesting. Tell me one lines on both sides, Jon being a owner's rep TK being a GC. You guys have worked on one common project as well. when the NAFTA thing was picking up and the tariffs was increasing, back in the day, explain in one or a few lines, Jon, were you navigating with the GCs and TK how are you with the owner's rep?
Jon Wright: Yeah, TK and I did a project that spanned out three of these; tariff one, COVID, tariff two. I'd say I approached it with some skepticism, about what request coming from the subcontractor and contractor community to add. and, probably got there over time. But,
advocating for precision and a little skepticism is, I think, an important message to send, before you just accept the increase. Because at the end of the day, we think they're adding it anyway.
TK: Yeah.
Jon Wright: talk about it, the silent padding is hard to avoid
TK: think from our perspective, we had a lot of conversations. Some of them were, very, energetic, lot of back and forth. what was interesting about that job was we planned it during the tariff one days. The execution when contract and everything was signed happened during the COVID days.
so if I were to break it all down, it ended up being a very successful project and coming actually on budget, so actually we did end up with what John was saying, a bit of an open communication.
Again, we, and we weren't on time though, to be very honest, because COVID physically shut the job down. And that's an interesting conversation too, because COVID shut us down for six weeks, added, I think 10 to 12 weeks, and Jon's "What the hell?
why is it adding more time?" But like construction projects are like a ship.
it takes time. You need to get the resources back, you need to get the labor back.
It's just to turn the whole thing on. so that was one intense conversation we had where it was like, Yes, I get it, it's day for day, but why is it more than day for day, right?" So there, there is that aspect. but I think one thing we did well when we were planning for tariff was by having open communication, we created some allowances which were open book, to deal with tariff one issues, but also deal with there was a timeline client really wanted to hit.
So once we created that open book bucket, COVID had its pros and cons. Obviously, the con was shut down, but the pro was timeline was less important. We were able to tap into that to, deal with COVID, which is why the budget didn't change even due to COVID.
Jon Wright: All right. AI data centers. Will the music stop?
Kanav Hasija: let me set some context first the listeners. we are seeing both sides of the coin. three sides now. one is Every AI lab, every big foundation model company is saying, "If we had more compute, we would have sold more tokens." because the world needs more AI tokens. case in point, Anthropic went from one billion to thirty billion of ARR in the last six months. let me say it again. Anthropic went from one billion to thirty billion of annualized recurring revenue, in six months. could have gone 100X if they had more compute. That's what they say. so every formation lab company supply constrained. And guess what the bottlenecks are? the data center build-out There's two bottlenecks there. One is the citizens in, in some cities are opposing because they think, data centers will be bad for electricity prices in their, localities, or some rumors about, baseless rumors about water constraints, which we'll talk about in a moment. But that's one side of the coin, which is they're constrained by that. They're also constrained by, guess what? Electrical workers. There are very few electrical workers out there who can get electricians who can get the energy to the data centers, right? So that's a big constraint.
That's one side of the coin. The second side is, what did you just say? You wanna add a trillion dollar of CapEx, to these data centers to do what? To help us with AI? Are we even seeing enough value in AI? Yeah, it might have some cute demos, and it can write poems, and it can, write emails, but is it a big productivity booster to us that it boosts productivity north of two trillion dollars? That's-- There's a skepticism there. so will the music stop? so these are like the two sides of the equation, we're dealing with. And the big question is, data center is the biggest growing vertical market in construction. fueling all the top GCs the question is, will the music
stop?
Jon Wright: My, response to that is the music will not stop completely, but it's gonna get a little lumpy. Today is the 22nd of May, and it's graduation season, and the headlines you saw from last weekend were several commencement speakers who mentioned the word AI got booed. I think that's pretty interesting and probably indicative of where,a lot of America is right now.
So if you're developing a data center or you're working on a, pre-con on a data center, I think that you can probably, count on a macro level, these happening. It is imperative that the United States builds these data centers here. We have to do it. we're in an arms race, if we don't do it, hyperscalers are gonna have to go to other places, and that's not good for us. But these data centers are going to be in front of city councils and state utility commissions, and they're encountering a ton of political resistance. It's just, it's happening. so if you've got your, plan A develop... If you're a developer, you've got plan A, you better have plan B and plan C in terms of location of these things because you could encounter some local resistance, you could encounter a state utility commission because the fear from ratepayers and citizens about the electricity going up is real.
And I think we've seen some evidence of that. I also think we see a lot of kind of fear-mongering out there, but, that's real. Those cash flows are gonna be lumpy coming to the industry because there's gonna be start and stops. The other thing that's gonna happen, I think, is OpenAI's gonna go public.
Maybe we have some other IPOs, and then you're just gonna get the kinda typical corporate gyration that happens on a quarterly basis or a headline basis. If something happens, it's gonna pause a job. that doesn't always make, big news, but, if you've got hundreds of people working on something and then, corporate finance wants to pause it for your earnings call because something bad happened, that is huge.
It adds cost and time and uncertainty for everyone. So I think it's gonna be lumpy here for a little bit. What do you think, TK?
TK: I think data centers are great for blue-collar workers because it's creating jobs. the impact of data center is the blue-collar workers are probably the last ones to get impacted by it. it's automating mostly white-collar jobs.
But I don't wanna minimize the energy consumption side that these data centers will have. And, I was wr-reading a metric that, China added as much power in a year, like talking about arms race, last year, as all of United States has, there is other infrastructure that needs to be coupled with the data center build-out.
we keep forgetting infrastructure. There's a tendency to look at China and say,how do we compete with them?" one of the simple answers is we need to have a better infrastructure here to allow for data center build-out, to allow for goods to move from point A to point B faster.
So that's the construction and b-blue-collar perspective. zooming out, I'll just take a historical perspective because I'm a student of history. Every revolution that has happened has started by bourgeoisie, right? Whether it's Russian Revolution, French Revolution, and that's your white-collar middle class right now, which these data centers are most impacting.
This is-- I'm talking about outcome, not the construction piece now. and so it'll be interesting to see how that plays out because, history has shown when you go after the bourgeoisie, it can turn into, major changes within the society itself.
Kanav Hasija: there are these three data points to cover, right? If you look at the GDP growth of United States and you exclude data centers, it's a flat economy.
It's only the data center
TK: idea.
Kanav Hasija: that's fueling the whole growth of United States, right? Coupled with that, there's a huge opposition, to slow down the data centers. Anthropic was supply constrained. Anthropic has been on the meteoric rise. What did Anthropic do? They signed up a deal with Musk saying, "You have Colossus One, which is only 15% u-utilized. Can you give us the rest 85%?" the S1 filings just came out for SpaceX, so we know the numbers. SpaceX just rented out 1.25 billion a month, for its data center to Anthropic. So Anthropic is spending about billion to Colossus One a year, because they are supply constrained. they were lucky they had an empty data center lying around, which SpaceX did not use a lot. but
TK: because government's not using Grok
Kanav Hasija: I'll not reply on that one. But, I think it's Grok is too early, the ARS. it might catch up. it's still not there yet. But the thing is, what many, middle class white collar, or lower middle class white collar is not understanding is For every dollar of dataset, for every dollar of AI compute, The hyperscalers need to invest about three to $4 in CapEx of that three to four dollars in CapEx about $1 is construction.
Jon Wright: So give us some gross numbers on that for the listeners
Kanav Hasija: if you picked up a Claude subscription for $20 a month are e-essentially same $20 will go back into construction. Why is that? Because for every $20 you spend to Claude, Claude gives about 10 to 15 back to hyperscalers. For those hyperscalers operating cost of 10 to 15, they need to spend about 60 in, in capital expenditure. so they recover the money in six months after they build on the data center, right? But they need to spend that $60 in the capital expenditure. Of that it's chips, it's electricity, it's all that kind of stuff. But if you remove all the chips and the electricity, we just include construction, of that 60, construction is a, is about 20. So for every $20 you give to Claude, $20 actually going back into construction of building data centers. so that's how the economy is fueling with the build-outs. we saw, $100 billion of commissioned this year for data centers.
we could only build like 60 of that because the rest got opposed by city councils. but st- 60 is still being built out, and the demand is just quadrupling. It's not slowing down.
Jon Wright: So for the industry,some of this is not gonna be recurring, but it's here now and it's here in the midterm. And so you should be grabbing
TK: on the construction side, the math is not everyone can build a data center. So you got ten to twenty GCs who can build data centers, and when you're talking about these mega data centers, it's probably ten. so it does benefit that smaller, cohort because you need a cer-certain level of expertise, certain level of bonding capacity.
there are a lot of things that go into it. The other thing that is about data center from a contractor perspective, they are so speed-focused, right? So every pro-project has this, speed, cost, quality, balance, And when speed is the focus, margins are higher
Kanav Hasija: and that's very interesting. And I think one more trend with this, I don't know if you guys are seeing it, is, apparently warehousing business has become a big thing now because you can have regional warehouses which can later be commissioned into a data center
Jon Wright: This has been trend for a while too, industrial property near major metro and it had a huge moment during COVID. So and reuse in a data center is another interesting, trend, and one I've seen personally, from corporate users here in the Bay Area.
All right. Robotics. very interesting week.
Figure AI, Bay Area-based, humanoid robot company, for the last week or so had a live stream of humanoid robots, sorting packages. I think they just concluded it today, just to, let the team have a break for the holiday. But, they also had, notably a competition with an intern on number of packages and time and mapped how many times he had to take a break versus the robots.
And the robot, it's not a continuous work with one robot. It was, multiple robots. They relieve each other. They go back and forth. I think this is being hailed as a big milestone. It's being discussed. we've got that. We've also got another announcement from a group that had raised million for an industrial robot,
application on job sites.
And so they're still trying to figure that out. a humanoid form factor, more the spider robot utility. can fasten, can inspect, can, multipurpose. So I just wanted to talk a little bit about it. I'm really curious to hear what, TK thinks about this in terms of when are we gonna see the application and proliferation of more robotics on the construction site for improving efficiency?
And just one stat, in terms of, adoption of construction robot deployments,
approximately 40% of the adoption so far has just been drones and inspection robotics. curious what you think is possible here and what around the horizon, and applications on the job site do you see as low-hanging fruit?
TK: it's a great question. I've been watching robotics very closely. In fact, one of the projects we did, I tried really hard to get robots to put drywall up, but didn't happen. anyways, coming back to robotics, I think even the packaging thing, and I haven't watched the video yet, so I'll go watch it.
I've noticed robots do well in repeatable tasks when they are on stable ground. So a factory setting where the floor is level, clean, you sit the thing down, the motions are repeatable, it does well. Unfortunately for construction, Buildings are not as repeatable. at a micro level it is, at a macro level it never is.
And the surface we work on is not clean and level. That is where I see these robotic implementations stumble a lot. there is no question that they will bring more efficiency, but I think the way we'll have to approach it is it'll have to be a two-part approach. I think we need to do a better job on the planning side to build buildings that are more conducive to repeatable tasks at a micro level.
can we standardize the outlet locations on the wall room by room, can we standardize conference room layouts? Can we just break the building into these chunks that are as repeatable as possible? Not from a prefab perspective, but from a perspective of someone being able to replicate those motions.
Like once I set a robot in a room, is everything at the same distance when they are installing it, right? Thinking like that during planning. We don't think like that during planning at all today. Because we don't need to, first of all, we don't have any robot that does these things. but we should.
and a step before that, and I was talking to Kanav about that, could be to deal with the flooring issue. You rent a large warehouse, as we talked about, there are a ton of them these days. try to prefab things there because warehouses have great floor conditions, and you can prefab things there with robots and then transfer it to the site.
That will be the half step before the full step of trying to replicate it on site itself. So- Yeah
Jon Wright: production with robots first would be a entry point for your perspective
TK: With the thought process of making walls repeatable. Yeah
Kanav Hasija: Yeah, I was doing the math on this, Jon. So I was like, my first reaction was, can I have a big mega factory of prefab materials in middle of Texas and which transports the material to the whole nation? the answer was the transportation cost is too high, to do that. Then the next thought was, can we have micro factories of prefabs all the major hotspots where the development is happening? The problem is San Francisco needed a bunch of buildings five years ago. Now it, it needs a bunch of houses. So the whole landscape of the city is changing.
The demand is changing. So you can't have a factory built out for one kind of a demand. Then the answer was, oh, can we have Good robots which can do-- a drywall cutting robot which does an amazing job in a warehouse setting, not on site. And if you need a drywall robot in San Francisco today for the next one year, just ship it out. So there will be a of shipping out of robots for specialized use cases the nation. rent a warehouse, do some stuff there, and then, ship the material nearby
Jon Wright: I think the whole concept of off-site construction is a topic for another day, and I'm deep in this. But, my view here is I'm bullish on robotics. just personally, I think that setting in a developer owner standpoint to take a risk on a new technology, probably if your general contractor says, "Hey, use this new technology, on your job, and, it's our first time doing it," probably don't wanna fund that. so that's gonna have to be funded through other means, I think, in terms of proliferation. But, look, I think reduction in labor cost, repeatability, all in. And I would say, I agree, you're designing an office building and the contractor says, "Hey, look, I wanna have repeatable locations on this and that."
it's gonna have design implications for spaces. I think a lot of people are willing to make that trade-off for a lower cost. I'm just curious, TK, when you think about a humanoid application, let's just say we finished interiors or finished the structure and we're in the interiors, your view on maybe a general purpose humanoid application versus say, the industrial spider robot? You talked about drywall. other applications could you see this working?
TK: I'll start with some of them that are already there and very successful, like layout, because the floor is clean, everything is open. the other one I'm seeing more of, which, should be easy to do is painting and taping, when Kanav talks about doing drywall, like when you have to cut framing members and all that, they call it robotic, but it's a big stationary machine that does that,
it's not a humanoid thing like you're saying, Jon. so to me, a general purpose humanoid robot that can do what a carpenter does, seems very sci-fi to me today. And it could be my ignorance of not knowing robotics enough. I just find it hard to believe a robot being able to effectively replicate what a carpenter does.
measure twice, cut on- You know, the-- just like a large machine like in cutting framing member, which is a large machine, you feed, big members in and it cuts it to size. That's doable. I get that. But a humanoid thing walking through the site and being able to do what a carpenter does or what an electrician does with conduits,
I'm not seeing that right now. I don't know if we'll even get there in like next 15 years. But, I'll go watch the packaging video first.
Jon Wright: Yeah. No, it feels like early innings. I would say the other constraint on these humanoids the strength that they can, to move material around, and they're a little bit fragile right now. I think 15 years feels really far out. I think we're in early innings.
I think this is gonna accelerate very quickly. I think these humanoids are gonna quickly become companions to workers and laborers and in our homes. And so it's a natural progression of the job site. But right now, I think they are a little fragile. I think when you add strength to them, you add a little bit of risk, for hu- humans working alongside them. but, I agree. Let-- Think, something to watch here, cost is, definitely something we should be all looking to do, throughout
the industry.
TK: a pushback on that is like car manufacturing industry has had these robots for 20 years. they literally can put a whole car together with very minimal human involvement. And car manufacturing is probably closest to putting together stuff on a construction site.
We still haven't seen that happen. I wonder why that is. and so that's why I think even 15 years is aggressive, in my opinion, for a humanoid style robot to be able to put anything together in a building.
Jon Wright: can tell one use case for humanoid robot replacement would definitely be a project executive
TK: Yeah.
Kanav Hasija: All right
TK: that you don't need a robot. Claude is already there, I feel.
Jon Wright: TKBot.
let's digital twin TK Yeah.
Kanav Hasija: sell that.
Jon Wright: All right, Kanav, why don't you,this segment, I'm most excited about this, and give us a little tech corner on AI adoption and, where we are right now
Kanav Hasija: Yeah. Awesome. Oh, I love the AI corner. I was having a lot of conversations with the AEC leaders in architecture, engineering and construction, all three. And I keep on hearing these views that is scary, it's gonna take our jobs away in five or 10 years.
But at the same time, I hear, " Oh man, AI doesn't do anything good in construction. what's happening?" So it fails. it's like a cute demos that happens, and then once you try in real life, it just crashes. I was trying to make a sense for them, like what works and what doesn't work in AI. So let me share those resemblance,
with all the listeners here, right?
if you haven't seen this already, Anthropic came out with this paper on March fifth of this year, where it said, " Let's look at the theoretical coverage of what AI can do," which is the blue chart here. And let's look at where AI is used today. how much of that is used today, which is the red chart here. As you can see, AI is doing a lot of stuff in the computer and math world. So you see Claude making code for you. You're seeing Excel plugins where it can do a lot of good stuff on the Excel side and do pivot tables and analysis and spreadsheet stuff. good stuff. And are seeing that coverage also.
the red bar is pretty high. There's still a lot of overhang, which is people are not-- still not using it completely to what it can do. So there's a huge overhang there, which is where, by the way, biased for Anthropic. They want more funding, so they got this paper out, right? But, if you look at construction, even the theoretical bar is pretty low, and the red bar is like way low.
people are not using AI, as simple as that, right? There's a reason for that. The reason is wanna divide AI. AI is loosely used AI overall. and for the nerds who have been doing AI for the last forty years, like AI has evolved from machine learning to neural nets to now transformer. If you look at transformers, the AI we use today more generally, divide that in three different types of models. It's language, vision and reasoning When you use ChatGPT or when you use Claude, they're using all of these three behind the scenes for you. So what do I mean by that? let's say you say, hey Claude, summarize this document for me." language, You say, " Hey Claude, here's an image of something. tell me what's in there."
That's vision, And you say, "Hey, AI, hey Claude, analyze these 10 things for me and build a business strategy for me." That's reasoning, right? Claude is using all these different models behind the scene. That's why we call it multimodal. But if you see language, they had a benchmark called MMLU, which is like an SAT for AI.
It's if AI had to take an SAT exam, what would that look like? it used to score 67% in 2022. It went up to 88 in '24, then incrementally went up to 90 in '26 My expectation, it might go to 92 or 94 in the next years. It's pretty good already. So if you experience AI today and say, "Hey, summarize this doc for me, write this email for me, clear this document for me, rephrase it," it's doing a pretty good damn job.
the benchmark That's what, you're experiencing right now in language. When it comes to vision, AI is good today. And to be honest with you, it just started, two years ago. the care that the models took in vision is two years old. So it's getting there. Reasoning is only one to two years old, so it's getting there. if you look at reasoning there, all these, AI nerds came together and said, "Hey, can we make a really tough exam for AI?" So all these PhDs different domains, art, science, philosophy, engineering, what they call humanity's last exam.
So what they're saying is, if AI can pass this exam by 100%, That's AGI for us, right? so AI used to score two point seven percent, now it scores 44. it might get to 65, 85 is my prediction in the next four years. Let me show you a fun fact on vision. a paper got released on May 1st, which says... It's benchmarked the multimodal models on how they understand the architectural drawings. And you scroll down to the f- to the results, The best model is Google Gemini 3 Pro in Vision. It can count doors with a 39% accuracy. I can flip a coin better than that accuracy. so that's where AI is not good. And doors and windows are the basic stuff to count in
Jon Wright: So we're seeing right now these are elements within a drawing that it was scored on ability to identify. So 39% for doors, 34 for windows, 89 for bedrooms, and 82 for toilets. And that's consistently throughout all the models, but Gemini is winning right now on vision.
Kanav Hasija: And let me be clear. if that door had a note somewhere or spelled out in a door schedule, in a table, a language model. That will pick it up. We're talking about here, door was drawn in a plan, can we pick it up or not? a door from a schedule can be picked up very easily from a language model. it's identifying the door as it was drawn in the floor plan,
TK: just one thing on that benchmark, Kanav, I will add. That doesn't include walls and flooring. even when it talks about toilets and bedroom, it's talking about it as a space. So the more complex stuff where the value is to a general contractor or an estimator is, can you identify a two-hour wall vs. one-hour wall versus a privacy wall?
it's stuff like that which is more intricate. Mechanical ducts and electrical outlets. And so it didn't check on that. But my assumption is, or from what I have seen, it's, it performs even worse on that.
Kanav Hasija: So if you want to search long spec documents, which is like 1,000 pages long and summarize them, language is good. Go for it, right? to read contracts and identify risk in contracts, go for it. in drawings, you want to read notes within drawings, go for it, right?
Language will work pretty well. But people get very disheartened when they're saying it can't take off quantities from drawings, it can't do reasoning really well, it can't identify scope gaps. It's not there yet. and when it gets there, it's gonna be really good for construction. the sole reason why I joined construction industry is because it's full of documents and images and plans and drawings and the prior SaaS world could not help it, but the AI world might be able to help it. and that's why we are all bullish about this.
Jon Wright: Yeah, that's fantastic. this has been a great, conversation today and, I think we'll just call it a wrap
Kanav Hasija: Sounds good. See you guys next week
Jon Wright: