Victor Augusteo

I am the CTO at Boone AI.

Boone AI is a company that builds a pre-construction platform for a company that builds buildings and hospitals and apartments.

So pre-construction means it's the part before you actually build the building. So it's the preparation phase. So imagine Mr. George has just won the lottery. You want to build a hospital for charity. So you shop around and for an architect you have a drawing. And then after that, you want to get a builder to give you an estimate of how much it's going to cost or how long will it take. You're not just going to ask one builder, you're going to ask 20 of them. And each one of them is going to have to tell you how much is it, how long it's going to take. They can't charge too high because you're not going to go with them. You're going to go with a cheaper one. They can't charge too low because then they're going to lose out on money.

So they will have to give you a very accurate assessment of the actual building and what it is going to cost.

Exactly. And in record time. So the winning bid usually here means that you want to bid accurately, but also you want to bid quickly. And quickly means that when you are looking at a plan for a building, for example, this building, it will have a lot of steel foundation and structure. And they're going to give you a list of plans for each of the level. There's going to be a lot of steel beams and joists and the columns and everything. And right now, a lot of people are doing it manually, meaning that they will actually use a software to go with a mouse to click here and click there. And that's how long the beam is going to be. And then they'll do that for every single beam and joist. If you just look up and you'll see the beams, there's so many of them. So it will take them hours and hours to just do one floor.

We help them to. There's a lot of things that we're building. So one of the biggest things that we think is the killer use case for AI in this industry is speeding up this process that we call the take off process. So based on the plan that the architect has drawn, how many lights do you need to buy, how many kind of length of cable that you need. So all that kind of stuff is the thing that all the contractors need to know when they are deciding whether to bid on this building or not. And then they have to give an accurate estimate of the prices. So that's like one part of it, what we call the take off. And then after the take off, after they do the take off, they will submit the bill, the kind of the bid. And then they will, for example, if they are granted the process, granted the bid, they will then have to deal with what we call the bid leveling process, which is another process where, let's say you are the general contractor who builds a building. You don't do everything yourself. You will get a electrical subcontractor to come and do the electrical. They will get a mechanical subcontractor to do the hatchback, the piping and the other. The same with landscape and the other. So you're going to be sending out a bid kind of like request to the subcontractor. And the same process happens. That this GC, the general contractor, is going to send out bids to maybe like 10, 15 different subcontractors. And they all come back with different bids. Now we have to figure out again, which subcontractor do I want to go with, because they all give me different prices. They all use just like different brands. And then I have to be able to quickly look at it. So this is another product that we are also using with AI. So we are building this kind of AI platform for the construction companies to just speed up the process, accelerate, automate.

This is extremely disruptive and disruptive in the sense that the companies that are doing this, like you say, construction is a very old company. There are a few legacy players like Procore, Trimble. They have been in the industry for many, many, many years, 20 years. And the software are still there. They are usually with the case of Trimble and Bluebeam. They are kind of like a Windows application that has no connectivity. So they just go there. You have to export an Excel file or something and there's no connectivity. So then we are bringing this to the Internet and we're adding a lot of AI into it. And actually, some of our customers, like I said, Postel, who did like a still work, they say that, for example, to estimate one page of one level of the still framing will take them maybe seven, eight hours to just do that. With our software, they can just click a few buttons and then it gets a result in like 30 seconds. And then they just spend a couple more minutes to review it and fix it. So we are bringing the time that it takes to estimate the still from like eight hours to only 10 minutes. So that's extremely disruptive.

Exactly. Exactly. So the problem with these construction companies is that the revenue that you can get is gated on how many bits you can release. And how many bits you can release is gated on how fast you can do it and how accurate you can do it. If you only have two estimator and each one of them will take one day to do one plan. If the plan have a hundred page, it's going to take you a long time to do. So if you have the two estimator and you use the Boon software and you're able to bid extremely quickly, you're going to be able to put up so many more bits, thus increasing the productivity and the revenue of your company without needing to hire more estimators.

There aren't that many.

Yeah.

Yeah. So this company was focusing on logistics and tracking like around two years ago. We started and then just around five, six months ago, we decided to pivot the construction. And so far in this kind of five, six months, including the time to build up everything, we have maybe around, you know, 10 customers right now. And several of them has released case studies with us and they are very happy. Like I say, Porcelo is one of the one and we have some other. And we also work with a bunch of kind of consultant in the field as well. So like we are not construction kind of like mechanical engineer, civil engineer. We are software engineer and AI engineers. So we work together very closely. We have daily meeting with these people in the construction industry to just learn about what they do, learn about how we can help them to accelerate. So, yes, we are we are building really fast and I think there is still so much work to do, of course.

Right now we have around, I would say around 40 software and AI engineers. And then we have around 40 kind of construction professionals. So this will be civil, mechanical, electrical, structural, because we need them to help us train the model and everything. And we are operating kind of on a kind of remote and local as well. So we have a bunch of people in here and around our own country and also a bunch of people in Asia as well.

So it affects both of them. Right. So initially we are growing much quicker. But since we adopted AI and cloud code specifically, we have grown much more slowly. Now we are hiring with much more.

We only hire if we really need to because now we are finding that the people that we have can work with a lot higher productivity. So we need to hire less people.

Yeah, what we're looking for when we hire someone is someone who are already comfortable with using AI.

So for example, I come from Apple before and a lot of the interview at the FAN companies are very algorithmic. Right. They have to go there and interview, but the other structures and algorithm nowadays cloud code can do that in like five seconds. You don't even need to understand it anymore. So a lot of work nowadays that we need to do is just to understand the spec of what is it that we need to build and then sort of have that taste of like how do we build this. Right. And then you have to how do you prompt the A.I. So whichever one they use on generating the right plan and then you have to review the plan and then you have to execute it and then you have to know whether this code is good or not. How do you do that? So the interview process that we have created since we adopted code code is now very different from before. Before we still ask algorithm question. But now we are asking really here's a real use case problem in construction field, which is like looking at a full brand drawing. And then you have to like write a computer vision code to like trace the wall or something like that. And you can use any A.I. tool that you do want. Right. So we are seeing that successful candidates typically really proficient at using those kind of A.I. tools.

Yes. So it's not about specific algorithm that you need to solve anymore. It's about are you able to just take a kind of like open ended problem that we don't tell you how to solve it. Right. We just say here is a problem or here is a software that we need to build. Here is the input and here's output and you sort of have to build it end to end. So the scope is much larger now. It's not just write one single piece of code but it's actually built entire application in like half an hour.

General problem solving and I would say one of the most important thing is the comfort level of dealing with LLN. Right. So I would make it an allegory to this metaphor would be like it used to be in woodworking. You have to use a handsaw and just saw it and you know how slow that can be and it's very tiring. Now we have a chainsaw. Right. But the way to use a chainsaw and the way to use a handsaw is very different. So we want people who already have used chainsaw before which is cloud code or cursor and then they can just go in here and then use that tool and impress us with how you use that tool. So it becomes like one is like how do you think about a problem and then two how do you apply your knowledge by using this new tool.

Yes.

Yes.

Yeah. Yeah. This is this is a very big problem. This is actually the question I was asking before in the front right. And we have automated the problem definition because now we can talk to our customer. We have a recorded and then we have transcription running through the transcription and can understand here are what the customer is saying and then we can file a ticket for it. We can quickly develop it. But then it comes back to your point after we have so many of these PRs and we have this application. How do we know this is solving the customer issue. Right. So right now we are hitting this bottleneck in which we have to have our kind of product designers and product managers just go and try out the software and then see whether it works as a customer wanted. But this is a giant bottleneck because we cannot just have our one or two product designers and product managers to QA the work of 40 engineers. It's just not possible. So we've seen now that it the QA has been kind of hard to catch up. That's why I am currently investigating ways on how do we make this much more scalable with a. Again just like how I has scaled up the productivity of engineers. How do we do the same for the product manager and the product designers so that they can quickly QA. So I think imagining if let's say our product designer and I say Tyler Tyler is really good at knowing and have that taste right of what a delightful software would be. But Tyler is just one person. How can we production eyes Tyler into a C.I. platform so that we can have this Tyler bought to just look at every single PR right now we're not there yet. But I'm looking for ways to get there.

Yeah it is a hard problem.

Yeah.

Yeah lately in the last couple of months and last month at least has been open claw. So open claw has been the hottest topic on the Internet. It's like the highest rising star at GitHub. Right. And Peter was hired by open A.I. and all that. And in fact like today we have GDC and Jensen from from Nvidia just announced they have nano cloud which is the Nvidia version of open. So everyone's just releasing their own claw. It's the hottest thing. And I run open claw at home and make me for my family for my wife and I to plan everything about our family about trips about doctor visits about health about finance investment. So it's been great at that. But that at work it has been extremely transformative. So now we use open claw for almost everything from sales to marketing to engineering to debugging to like product management. And now even our our consultants which are those electrical engineers and estimators and like mechanical and civil. They are also have been using our kind of open claw instance and they have been saying like wow this is amazing. So it has been a really interesting journey to see how we are using the same model. It is just the same Opus model but somehow with the new modality because of the context because of the integration of it. It can do so much more. And this gets me kind of excited as well on what are the next steps of evolution because right now we just talk about how it is still very hard for a model to kind of mimic a person. But with open claw at least we didn't even have to change the model. The model is the same. It's just a harness in which we've run the model that makes it so much more powerful. So because of that I am hopeful that at some point in this year we'll be able to make it work much better for like product designing and product management and maybe up to the same line as the engineering.

I have two young kids, seven and nine. And I do not know what the world will be like and how best to prepare them in this world as parents. I think a lot of people are experiencing the same thing. So it used to be that we know that if you just work at study accounting or study finance or computer science you'll be fine. But now we don't know. It might not be fine. Like we're just talking about whether the software engineering as a profession is going to look the same in 12 months. And this is talking about 12 months. By the time my son who is seven right now go to college and graduate it will be like 13 years until he's 20. I don't know what's going to happen in 13 years. And also like what is the steps to kind of like educate them into that. And then in addition to that there's going to be a lot of people who lose their job and could not compete with machines. Right. And we are seeing with this already with the software engineering right. The people who are very productive with AI are much more productive than the people who also use AI but not much more. Right. So the gap is widening and there is a lot of societal implication in that that I'm not sure yet how we as a society is going to navigate.

I would guess so poorly.

I am always hopeful that we will be able to find some sort of solution to this problem. And whether this is the same problem as the previous problem we don't know because whenever people talk about this we always bring back. Look at the Industrial Revolution used to be that people were working at a textile mill and all that. But then the machines come and people say everyone's going to be jobless. That's not true. The job just changed from manually doing the textile and pulling pulling the car to machines doing it. But not human doing other things. But now is it the same thing. I don't know.

We'll find out.