HCSS x ENR: Building the Data Foundation That Makes AI Agents Work

Hello, and welcome to this webinar, building the data foundation that makes AI agents work. This event is brought to you by Engineering News Record and sponsored by HCSS. Hi, I'm Scott Seltz, executive director of BNP Media's construction sector and publisher of Engineering News Record. I'm gonna be your moderator today, and thanks for joining us. Our presenters today are Michaela Halliwell, group product manager focused on platforms, data, and AI at HCSS, and Michael Redino, Director of Information Systems at McGuire and Hester. Now before we get started, please take a moment to scroll down and explore the webinar console. You can download handouts, click on the speaker images to view their bios, and submit your questions or comments anytime using the q and a box. We'll address as many questions as possible at the end of the presentation. Today's webinar is being recorded and archived on enr.com. And now I'm excited to turn it over to Mikaela and Michael to kick us off. Yeah, thank you so much, Scott. And thanks for everyone who joined for carving out an hour. I promise you, we're not gonna give you any product pitches. This is a working conversation between Michael and I, and we're gonna talk about the one thing that nobody's heard about all year in the last couple of years, AI, right? We're going to talk about what actually works for contractors and folks in construction and what really matters about the data underneath that. So as Scott said, I'm Mikayla. I'm a group product manager for HTSS. I deal with our platform data integrations and AI teams. Those are the people I get to work with. My team owns a lot of the plumbing behind our software. So our APIs, the things that make connections to other systems, our integrations, the folks that build those, our data products, how our customers report and keep tabs on their business, and then also our AI portfolio. So I spend a lot of my day asking what has to be true for the customer I'm dealing with to get the most value out of our systems and what needs to be true about their data specifically for the AI that they're working with and that we offer to actually click and make sense and be valuable. So that's what we're unpacking today. Michael, you live on the buyer side as a customer of HCSS, which is why I'm really glad you're here. Could you introduce yourself and give people a feel for who McGuire and Hester is and what that looks like on a given day? Sure. Good to see you again, Mikaela. Yeah. Mike Rodino. I am the director of information systems at McGuire and Hester. McGuire and Hester self perform contractor out here in the Bay Area Bay Area, California. So we get the honor of having all the big AI companies in our backyard. We are celebrating our 100 year anniversary. So founded in 1926. Hopefully represent, you know, a typical contractor, mid sized, but one that's, really looking forward to the future and especially adopting, Hugentic AI. You know, my specific role is to build the corporate mind and to start to enable all this hopefully wonderful stuff that's coming. Yeah, fantastic. A hundred years is really an incredible accomplishment for a construction company. And so congratulations to you guys. That's truly amazing. I hope you guys do get a chance to really celebrate. Here's what we're going to do for the hour, y'all. The shape of the hour. First, we're going to talk a little bit about what data in construction looks like, what's broken, what are the data problems that maybe quietly kill AI efforts. Second, we're going to talk about what does good look like. If there's a problem, maybe there's a solution. So what does that look like? And what are the traits that you can implement to underscore against your data and score your own systems against? And then third is how you get there. And that'll include a little bit of how Michael talks to his vendors before they actually get a seat at the table, how he's able to weed through and see what might work for them. And then we're going to hold time at the end for questions as well. So if you have any questions now, obviously, of course, drop them in the chat as they occur to you, but we'll save some time and carve that out at the end as well. If one's too good to wait, go ahead and throw it in. But first, before we get anywhere, I wanna know who has joined us and who's in the room. Be honest, there are no wrong answers here. Where is your company on their AI journey right now? Got a couple of different options here, different maturities. Do you have AI tools in your production? Are you piloting a few actual AI experiments? Are you on the more nascent journey piece of that where you're wondering, maybe we don't even really know how to get started. This is still in discussion. Go ahead and throw your votes in there and we'll take a look at how that is spread out. Generally, I'm expecting to see that there's a certain spread here, but never know, might be surprised on either end of that. So give it a couple more seconds and go ahead and vote, and then we'll we'll take a look at those results. Alright. Let's let's take a look at what we got on the results side. Go ahead and close the poll. Alright. This is what I was expecting to see. It really ballooning in the middle here. A lot of people piloting, exploring, not so many in production. Then good to see that there aren't very few that haven't started or that are asking what an agent is. If that's okay, if you are on that side of the scale, because we're glad you're here today because we're going to go over a lot of that. And so you won't be able to say that at the end, hopefully. So watching, learning, this is kind of where the spread is across construction and where I would expect to see things. Really, in the last The reason for that that I'm not surprised is really where we have seen AI evolve over the last few years. And I want to give a quick definition. You're going to see the word agent. You're going hear the word agent a lot. It is probably the most abused word of the last couple of years. It is not simply a chatbot that answers questions. It is not just a copilot that makes suggestions. An agent is something that acts. It is a a an entity, a workflow that takes an action. It reads your system. It reasons over what it finds, and it works on your behalf. So the progression of you going from point, click, point, click, point, click behind your computer, Now an agent has a job to do, and it can go and execute that point, click for you, come up with a bunch of answers, and make suggestions. So your role as someone behind a computer doing a workflow in construction goes from the doer of that workflow, whether that's looking for information or whether that is submitting a bid or working on a takeoff to now you are the approver. You are the reviewer of an agent that has already gone and done that work for you. I like to think of it a little bit like an agent is a junior employee or an intern that has just started. And the more information you give it, the better you train it. It's going to go and do a lot of tasks for you that maybe would have taken a lot of your time before. So that's quick definition for you. But the agent moment, I think, Michael, you've probably experienced this even at McGuire and Hester where AI comes on the scene and everyone's wondering what they're doing, what we're doing with this, maybe using it to write emails, answer chats, whatever it might be. And now that progression, this whole wave of technology that we're going through, especially in this industry, tell me a little bit about what that fever has looked like for you guys and where it's landing in Heavy Civil specifically. Good question. Yeah. You know, it's it's really, the timing of it is crazy because it's really just only been in the past six months. If you wanna know why Anthropic and OpenAI are gonna be worth a trillion dollars, it's not because of the chatbots. It's not because of the LLMs that you've been using. Frankly, you know, I think that's now universal. I think every employee probably has access to one of those, mostly because Google and Microsoft and the rest have been kind of forcing upon you. But really what's gonna change is when these agents start firing off and that's when they're gonna be really disruptive. We're trying to get ahead of it. We're trying to be proactive. We're trying to figure out where we can augment all of our people. Right? We want to take certain department and 10 x their workload, so we can start become becoming more competitive. We can start geographically expanding, find all those projects that we didn't know existed, all the all the things that a computer can do extremely well. Yeah, absolutely. And like you said, it is really in the last six months that a lot of this has started to heat up. And the reality of where we are in heavy civil is that I think of any industry, construction and specifically heavy civil brings in more data than maybe any other industry in the world, specifically precon and estimating. And that data capture pretty it can turn out to be pretty paradoxical because the adoption of these tools is accelerating very fast, but our data can still be very shallow. I think that even in the top ENR contractors, AI has roughly tripled in the usage and adoption among the top ENR contractors. And so that fever point with document review and scope generation and estimating is where a lot of people are starting. Where does that reality sit for McGuire and Hester? And where do you see the biggest problems, I guess, in the industry and the biggest opportunities for contractors to really step in and use AI in their workflows? One of the biggest universal problems is going to be just the fact that there's so much out there, right? You can be hit from a multitude of vendors. We have thousands of agencies that we deal with just here in Northern California. They all have their own systems and they all involve you know, a human going in there getting information, sending information. And as more and more of these agents are able to take over this test because it, you know, you don't need to develop a formal relationship with those systems and those agencies. The AI can configure it out for you. You're going to start to see all that all that work go away into where your your view of what's out there, your view of the information and construction is going to be universal. You're going to see everything and you're going to be able to act upon it. Once it comes into your organization and you're actually managing projects, you've won these opportunities, right? All that information from the project should be again the same thing. These agents can act instead of a human being collecting this information. You know, if you just start theorycrafting about what we could do, you could you could throw a drone up there today and and start tracking production data based off of what the computer sees instead of your foreman. Right? And allowing your foreman to get back to the work where we where we know that's where they wanna be and that's where we want them. So, yeah, I mean, these agents are gonna just take that data, multiply it times 10, and then hopefully make it actually useful. Yeah. And I mean, that would really be the dream if that is all achievable and seamless and consumable by the contractors. One thing I've noticed when I run into talking to people in the industry is that when I talk to contractors, I rarely find that there's a bad data problem, but I do often find that there's a disconnected data problem where estimating might live in one system, field in another, accounting in another. And all of those answers are needed by maybe one or two personas where they're having to jump from system to system. So be honest with me, where does the data really break down? Maybe not even just at McGuire and Hester, but across multiple vendors. Where do you see that disconnect problem? Well, I mean, all still exist in silos beyond the kind of ones we need to keep silos for HR and legal purposes. It's usually just a reluctance in the vendor to really adopt what's coming and that's the open ended relationships with your customers, right? We need access to our own data. We needed access in the format we need it. We need access using the tools we want. We want to use your tool with our AI. That's what's coming. We're gonna invest in what we want, and we just need the ones and zeros that reside on your servers. Yeah, 100%. And lot of those systems too are do you find that they are offering those ones and zeros, as you put it? Are they gating that? Is there a cost associated? Like, what makes that difficult for a contractor to actually get the information from multiple systems into their AI? Yeah, you definitely see still vendors are reluctant to stay on the forefront when it comes to this. Obviously, there's some good ones out there. You know one of the tests I put out there is am I going to get access in this way? You know up until like I said a year six months ago is I want an API. I want to be able to connect directly and have some guys on my end develop some ways to communicate between systems. Now I wanna see an MCP. I wanna see a connection directly from the AI that McGuire and Hester is adopting. Yeah. And so you bring up a couple of good points. And so I wanna pivot and talk about how we're thinking of this, not only at HCSS, but really what good looks like. What does AI ready data look like? And this comes from HSS's long term data strategy. This is not from a marketing deck. Data goes through a life cycle. It climbs a ladder. It starts as a record. It's something that happened at its event. It's an event that occurred. If you think of a construction site from an aerial viewpoint and imagine you're a drone or a bird and you're looking down on a construction site, there are hundreds of pieces of data happening at a given moment when a piece of equipment moves or when it turns on and off, when a truck comes onto the site, when a person leaves, when different events happen, when different materials are laid out or put back. There's millions of data points across the world happening on all these sites, and they have to feed into systems like HDSS or like any estimating software or accounting software or project management software. And so when that occurrence happens, say it's someone just coming in and logging in, we have a time card, piece of time card data there, that turns into a record into a software system that is consistent, coded, comparable to yesterday, comparable to last year, consistent or not consistent, and it exists and turns into a connection point and intelligence. Right? So now because of that event, an estimate, a job, an actual can inform on each other and other things. And so that intelligence turns into infrastructure, which is something we build our business processes out of. The questions that we ask on a given day of like, did so and so show up? How many people are we short on this job? What does this equipment look like? How am I behind over here? Why did this specific safety thing happen? All of these data points and occurrences lead to that intelligence, record asset intelligence and infrastructure. So the things that we see on the slide here are really how we climb that ladder. When I talk about we're going go through each one of these. But when we talk about granular, granular is the amount of detail, the scope of that piece of data. Michael may have logged in and clocked in on the site, but how granular that data is, where was he, what time was it? It's how detailed that data actually gets. Structured gets you from it being a record to being an asset. And then connected is asset to intelligence. And then governed, that last piece, is what makes AI ready to be read by an AI or an infrastructure where you can actually trust it and score it against it. So Michael, when you see these four and you hear these four, which is the hardest for McGuire and Hester or the industry at large to overcome? Good question. Mean, personally, the connection piece is what I'm most concerned with just because it's, in my job description. But I think most companies, especially our size, it's that transfer transfer into a structured system. And I think the importance of that is gonna become clear as we do switch over to the agentic AI world in that that's gonna define what we do. Both the fidelity level and the detail are gonna determine why McGuire and Hester is successful. And we wanna put that into some sort of structure that so then we can feed back into our estimating system, our operations and be more profitable. Yeah, 100%. I mean, and that's the reason we're even talking about it. We want to make sure our crews are safe. We wanna make sure our businesses make money. We wanna make sure that the people we're sending home come back safe and that we're giving the answers that we need. So let's talk a little bit about those four traits starting with granular. Granular meaning the detail under the roll up exists. So the excavator costs $40 last month can answer some of that, but it can never answer why. So roll ups are what answer why, right? The details answer why. And agents can learn and keep asking why if they have those details. Michael, do you have an example of maybe how granular data actually improved what was happening on a job or unlocked a piece of intelligence? And can you walk us through what that looked like? I'm not that close to the work, so I can't pick a specific example. But, you know, granularity when it comes to this stuff is always a battle because, I think most people would love to just be able to roll up everything into a line item called construction and run with it there and base all of your metrics and everything. But the reality is is because construction, we don't build the same thing over and over again. Right? All of our projects are different. So all we can do is slice and dice those projects up into the pieces that do relate potentially in the future. So every time we dig a trench, we can pretty much say that we're gonna dig the trench in the same way on the next project. It may be for a different size pipe or a different connection, different soil type, but we need to be tracking at that level. When it comes to granularity, I'm definitely a proponent of more. Know, obviously don't want to overwhelm their field with what they have to track, But we need to be able to take that information and relate it to our next project. And the higher you get, the less detail exists in the in the data, and you really lose out on those opportunities. Yeah. I mean, like you said, you don't wanna overwhelm the field, but even a consistent cost code or phase code structure could help give you that information for an LLM or an agent to pick up on later. Expect that generally this is what it costs for us to be able to do this specific type of work. So let's talk a little bit more about coding actually. Structure, your code book being your ultimate guide, right? Structure is what an AI is able to ground itself in in the field instead of just guessing from pros. There's an actual structure where a time card that says, Joe worked on the culvert forces a model to guess. But one that says, Joe worked on this Costco for this many hours, for this many days, on this day, now it can actually compute. And by the way, this is the difference between how an hallucinates and one that actually cites a real record. So one that makes things up because it doesn't have the information and one that's like, yeah, no, I already have that information. So Michael, I know you told me you're a structure guy, work breakdown, cost breakdown, resource breakdown. Which one do you think contractors are sleeping on and what does enforcing that really take? Yeah, I mean, the work breakdown structure is gonna be the key. I mean, that's kind of because we do individual projects, right, that's the one that can be variable, that's the one that exists on the project. They can roll up into your cost breakdown structure but those are more stagnant, right? That's going be related to your corporate structure and all that. So definitely the WBS and defining what you want to get out of it. You know, every project is going to be different but that doesn't help a corporation because again, we want to be able to take that information and win the next job with it. So you got to go through that step and turn it into what you have defined as your WBS. You know, the effort you put into there is going to unlock the ability for potentially the future when everyone out there has all the same information, right? If we all have cloud subscriptions and we can all log in and you can upload a set of plans and specs and say how much is this going to cost, right? We're all going to get the same answer at some point. So what's going to set us apart? Well, it's going to be this structure. It's going to be this information that we have that no one else has that's going to show. Oh, we're really profitable doing X, Y and Z And, you know, here's where we can make money and here's the opportunity. And so that's why that structure is gonna be fundamental. It's gonna define what your company is from a from a data standpoint. Yeah. It's it's also gonna give your code books or what are what are going to give that system a common language. Right? It's going to make sure that whatever agents or AI or MCP, as you mentioned, speaks to each other from a common language. So let's talk a little bit more about connected, which is the one I probably want to spend the most time with. So connected in one breath, if we think about the life cycle of a job from where we decide we're going to bid on the job, Job becomes a job, an actual, the actual feeds the next bid, and then every job you finish should make the next one smarter. There's some sort of flywheel effect to this that compounds the longer the history and the better it joins with the information it has access to. Once that loop closes, there's an interesting question that moves upstream from how should we estimate this to should we pursue it at all? And I'll leave it there because the specifics about how contractors wire the loops of what exactly they want to estimate on are probably not what we want to get into directionally, but without giving away the playbook, Michael, what is the litmus test, I guess, that you put different vendors or different tools that you work with through this piece of will they actually feed into that lifecycle, into that flywheel seamlessly? What are the things that you ask when you're going through vendor acquisition or you're looking at different softwares for your company to use? Yeah, it's definitely a question. At one point it was actually a question for HCSS too. I've definitely asked that to some of your folks. But, yeah, I mean, they they have to at this point have some sort of strategy around how they're going to handle, AI. Like I said, the the past six months, I wanna see the MCP connected to the big boys. Right? I wanna I wanna be able to fire up Claude, connect right in, and use that information along with everything else that I have developed, seamlessly. So that's gonna become a pretty strong gate for me, right? If you don't have that, you know, if you're still living in the world of your installed on prem and it's, you know, it's just sitting in a server somewhere, God forbid in someone's desktop underneath their desk, less likely to to get our business. In fact, I can say you're not going to. You know, we're not a huge company. We don't have a massive IT department or really structured way we go about vetting vendors. But my vision for the future, I need to see that. I need to be able to connect. Not every company is there yet, but at least they need to understand that that's what's coming and they need to be developing for that. 100%, I agree with you. And not having open APIs that are consumable, that are readable, not just by people anymore, but by agents and have enough context, MCPs very quickly becoming table stakes. Yeah, it's a no brainer to be able to close that loop, right? Because you're not able to be in all places at once, but perhaps with those things, you can pull the information in at once. None So of it works if you can't see it and you can't trust what you can't see. So speaking of trust as well, we're talking about governed data. One rule that we have at HTSS is that it's this concept of RBAC or role based access control where an agent will inherit the permissions of the person that is asking and never more. There's tenant isolation and role based access control. Those aren't features to us. They're prerequisites for every agent. If an agent can surface something to a user that the user otherwise couldn't have opened themselves, that's a problem. And it's not just a problem, that's a legal issue potentially. That's a breach. And so where I would push everyone on the call very strongly is with every single one of your vendors demand and ask exactly how their AI decides what a given user is allowed to see. Michael, where do you guys draw the line at McGuire and Hester? And maybe tell the audience a little bit, give them a little bit of a homework assignment on what they should be asking and what framework they should have going into any conversations with vendors. What's the gut check there? Yeah. You know, assuming that you're on one of the major platforms, either Microsoft or Google, you know, you can fire up Copilot or Gemini and ask it something like, what's the CEO's driver's license number? Right? And it's gonna try to find that and if you have access, know, that that that's gonna pretty quickly tell you if you have a problem. You know, again, we're not a huge company. Don't have a governance department. We don't have, you know, an even an individual that's concerned specifically around governance, especially with our data. So we rely a lot on on, you know, partnering with one of the the big guys. So we here are Google Shop, and we just assume that Google's going to have those protections in place, and that works for us. They have the same thing where, you know, you you as the user of the AI only have access to what you would normally have as a in your workspace account. So, that covers that. But what you're gonna expose is all the stuff that didn't follow your own compliance. So, you know, someone in HR stored a file they weren't supposed to in a certain folder. Well, it's quickly going to become clear. So it's actually a good test and it's a way to be proactive so that you don't get stung later on. Yeah, 100%. I got another poll for the room here that I'd like to see what the split is. Can your core systems talk to each other today? Core systems being project management, estimating, the tools that you're in every day, your accounting system. Do those systems talk to each other today? Michael, any predictions on the split here? I think it's gonna be a bell curve right in the middle. The options being, what is an API? Not sure who has APIs among my vendors and who doesn't. No. We move data between systems manually, partially, some connected, some black boxes, and then, yes, they're all connected. I'm gonna predict a bell curve too. Give it one more more second here and we'll see how that turns out and go ahead and close the poll and let's take a look at those results. And you were right. Pretty good bell curve. I would love to know which ones are all connected via APIs. That's fantastic. Partially, some connected, some black boxes. That's pretty much what we what we see as well. Moving between systems manually, man, I feel for you because I know that that's a pain point. Not sure who has APIs and who doesn't. What's an API? API is an application programming interface. It is how computers talk to each other and get information. But, yep, this is pretty consistent from what I would expect to see. So let's talk a little bit about what procurement looks like in that vendor test. We've talked a little bit about it throughout. Michael, you mentioned you have a question you ask on every vendor sales call, ones with HTSS. What is that and what happens to vendors who cannot answer that call? Yeah, I think it's right there. Are you embracing AI? I think we're quickly turning the ship around when it comes to whether or not you actually look to bring some of this stuff internally. I know we're all lucky to be in construction and close to physical work, but if you're in the computer science world, you know, your opportunities are are quickly being shut down. So, you know, personally, I see a lot of very intelligent, a lot of very eager eager people becoming available. And, you know, maybe maybe it does make sense to start up your own spot. But at the same time, we're a construction company and we're not a software company. So that has always been my take. But, yeah, I need I need my vendors to be ahead of me when it comes to AI. Right? I mean, it needs to be core to their existence. They can't ignore it. They can't think they're gonna just run their little app and it's gonna do its thing and and that's it. Like, I I need it to be bigger than that. Yeah, absolutely. That excites me so much that you said that because that's really where I feel the energy when being from HTSS. And I want to build on what you said a little bit with an uncomfortable truth. As someone who builds construction software for a living, companies like HTSS, we're not anthropic, we're not open AI, and we're not pretending to be. But the software companies and vendors who are pretending to be that you should worry about because there's a lot of AI being bolted onto their little apps. So now it's more than or it has implications beyond what a word can appear on a marketing slide. When you evaluate your software, you need to actually ask, is this AI connected to my data? Does it have the right permissions behind it? Or is it just a chatbot wearing a hard hat? Our job as a vendor, as HTSS, is different from a frontier lab like OpenAI or Anthropic. And I would argue it's probably more durable because it is the open structured governed data layer that intelligence, like a contractor like McGuire and Hester, needs to be able to act on. So being ahead of where contractors are in their AI, yeah, that's our job. The AI that we build into our products will be good because it sits directly on structured, governed, connected data. Others will try and standardize that with Claude or Chatchi BT and expect their systems to all be reachable, but everything has to work together. It has to have that connection, which is why we invest in APIs and in MCPs and an agent someone had said recently, I can't remember where it was, a magazine or an interview, but 10 agents sitting in 10 silos just gives you 10 faster silos. So openness and connectivity needs to be a line item in procurement processes. Put those into RFPs. Full API access to your data without a tollbooth in usable formats, a permission model that's respected, and support for those open connection standards. Ask to see the API documentation Right? Before you watch the demo, do you have API documentation? Can I see that? The vendors who have it are gonna be so excited that you asked. I promise you. There are API nerds behind the scenes being like, asked to see our dev portal? Yes. The ones who do not have it and do not you about it or tell you, oh, it's on the roadmap, those would be red flags in my opinion. Let me give a quick spiel and show you what taking that looks like taking that seriously looks like from the HTSS side. On one slide, I promise this is the this is the one the one pitch I'll give you. I really want to show you a breakdown of what HTSS does. It's really a picture of our philosophy. We don't have estimating living in one system and field and fleet on another. Our unified platform comes out of the box with analytics, with AI, with integrations, with reporting as a foundational first layer. And that placement is very deliberate. Our AI and integrations are features are not features. They are the foundation. It's why when a bid becomes a job across our system or a job becomes an actual, the actual feeds the next bid without you having to move anything manually. It is just part of that intelligence. So Michael, you're a customer of HTSS here. So greatest honestly, where does this openness actually work for you? Where do you want to see more from in the future? Because that pushes where we want to be headed, right? Yeah. So I mean, I said it last time, but I really like to see dispatcher have an API so I could start moving information there, but I get it. So no, HCSS has been great. They've definitely, you know, embraced what's coming and and adopted it, especially for heavy civil. Right? I mean, they're still focusing on what matters for us, which is, you know, how our foreman operate in it, our estimators heard all that great stuff. But they know it's coming and I really like what I see. Yeah, thank you for that. A couple takeaways and then we're going get into some questions here, everyone. So if you have any and you're thinking through anything, go ahead and throw it in the chat. We'll take those. But the four things I want you to walk away with. First, there is no intelligence without APIs. Openness has to be a procurement requirement. Two, you need to score your systems against a granular, structured, connected, and governed data layer. Three, close the loop. Your history is the moat, but only if it compounds. So make sure that whatever systems you have are talking to each other, you're bringing that knowledge from other systems. Four, Govern before you scale. Wall off payroll. Wall off HR. Run tests against that. The query test that Michael mentioned about asking that AI, can you tell me these details about my CEO? If you actually get legitimate answers back, you need to raise that with your IT department immediately. That's a breach. Michael, one last word before Q and A. Any one final takeaway for someone watching this who may still be resistant a little bit towards AI and where things are headed? You know, I'm I'm just grateful that somehow I chose construction as a path, I've been doing it for now twenty years. I see a ton of fantastic things coming. I see, you know, McGuire and Hester hopefully being on the forefront, especially in, you know, heavy civil, is tends to not worry about a lot of this stuff. So for that, I feel extremely grateful. I feel grateful to be partnered with HCSS. You know, I've always been a tech guy, so it's kind of my time to shine with some of this stuff. So if that is your role and you're on the call, stand up and start finding some of these solutions because they're out there. Yeah, phenomenal. Scott, I think it's back to you and we can get into some questions now. Yes. Well, thank thank you both for a really great conversation. And, yes, we did get a lot of good questions, so let's go. This viewer is asking, what software companies is Maguire and Hester seeing that are being opened with API MCPs? Yeah. Well, we were talking to one right here. So HTSS is they got they got asked last presentation. And, you know, you're gonna see a ton of startups. If you're familiar with Y the Combinator, you know, a couple of years ago construction tech was like the number one. And so all these startups are out there and they're gonna be promising you a ton. So you can evaluate them and they may serve a purpose, but a lot of them don't understand the whole construction life cycle. And and so that's where actually HSS is our is our fundamental. You're gonna struggle on the ERP side. A lot of the legacy accounting systems just from a tech stack, it's gonna take them a huge effort to get to where they need to be. Great. Thank you. Another viewer is asking, if I upload civil plans into an AI system, is there any AI systems that can provide a detailed quantity takeoff of all civil items? Yeah. So I'm gonna focus on the civil part of that because I think for the most part, the rest of the trades are pretty close. Civil is the one that's difficult. You're talking to, you know, what's under the ground, three d, all that stuff. So I have I have yet to see one that really nails it. Good. Okay, this is a good one. Zhiro is asking, I hope you talk about security. Why would I trust an AI agent to not share my company's proprietary information with another AI agent or the AI AI agent's parent vendor? Yeah. He this viewer has no confidence whatsoever that this won't happen. Your thoughts? Oh, yeah. I'll I'll I appreciate that one so much. I started in cybersecurity, so I feel the trepidation and the pain there. I really do. That's why that fourth piece, governance, is extremely important for everybody on this call to ask their vendors. Do you have role based access permissions? Do those permissions respect what a user can already do? How are those users that have access to the AI? Who is giving them access to the AI? And is that a an admin that's across the whole platform? Is it an admin that's specifically for AI? There's a lot of layers and nuance to that. And how agents are governed and how you specify a permission for an agent might be different from how you specify permissions for a general chat conversation like, you would with, you know, an Anthropic tool or an OpenAI tool like ChatGPT or Claude. So the the IT departments behind this and the vendors taking this seriously are going to take care of you. However, I am with you on that. I have I I know for a fact it's already happened. So and not within HHS, I know for a fact across software, this happens maybe even on a daily, where where people are not taking this seriously or they're not taking governance as a piece of connected and secure data. They're not taking that seriously. And so, yeah, that's gonna happen. You're gonna see it. So it's very important to have conversations with those vendors and make sure you understand how are these things permissions, how are they rolled out, and am I what data it does do those things have access to as well? Because this is concept in cybersecurity of least privilege of giving the least amount of information to something. We should take that with agents as well. Great question. An excellent answer by the way. This viewer is asking what was meant about not feeding AI tools for incident data? I wouldn't say that you should not feed it, but how there is personally identifiable information and incidents. You should check your company's policies on that. You should check with, your country's, policies on that. You know, there are, there are different policies in the EU that we have, from Canada, from The United States. You should make sure that whatever incident data I don't know what that means. Some sometimes that could be personally identifiable, information, or information that, is private. So I would just be cautious about what you're feeding as far as safety data goes, to an AI tool. Thank you. This viewer is asking, what's the smallest practical first step a contractor can take to map where its key data is disconnected today? It's a good question. Yeah. You know, coming into a new org, the first thing I do is just print out 11 by 17 sheet of paper and start physically connecting things, seeing what talks to what, what requires a human. It used to sit on my door. I I recently took it down. But you're gonna need a system map. You're gonna need an architecture. I mean, there's guys like me with that in their title. They're literally building the system. I mean, we're in construction. We know what that means. So that's where I would start. Cataloging, drawing lines between them, figuring out which can be replaced with an API is a huge step. I mean, you you may be taking away hours of of an employee's time where they're spent just grabbing information from one system and putting it into another. I I'm gonna, piggyback on that and say drawing it out, having a visual, that is probably one of the first things I do. If I ever get to a new software company, it's one of the first things I do, is ask for an integration map of the internal systems and then externally as well. What does that look like for us and what pieces of information are being exchanged? From there, you'll you know, as experience comes in that company, you'll be able to see where those blind spots are, because just having that visual or go finding that architect, it's a good step one, but then you also don't know what you don't know. It could look like a lot of information is going back and forth if it's a really busy data map, but it also could have a lot of things missing from it. So, experience and then, yeah, having that visual is just really helpful. Very good. Excuse me. This viewer is asking, I have records of every completed demolition project we've completed. How can I use an agent to cross reference that data to compare against bids I'm doing? Yeah. Good question. Okay. So there's some layers to that. You go ahead first, Michael, and then I'll Oh, I was just side. Take That might be a multibillion dollar question in that you know can you start to augment and potentially replace your estimating efforts based on your own historicals? Yeah I would just start trying it you know Claude can do amazing things and so pay for a subscription so you're protected you know all that good stuff and then see where it takes you. So depending on what format you know, when you say you have records of it, you know, it's probably a bunch of PDFs and then some database somewhere. You know, if that database is locked to you, you're gonna have to find a way to to be able to feed it. But you can start with simple queries. I mean, it will do its best, and it will do a damn good job of trying to answer that. Even just saying this is the type of contractor I am. This is the type of work we do. Here's the project. What do you what's what's the number? Great answer. Yes. Agree with all of that. Claude is amazing. Give it all the knowledge you have, but also in a couple weeks, if you are you are an HTSS customer, we'll be releasing our agent studio where you can do the same thing, and use all the information that's already in your system under a governed agent, to be able to give it the context for those things so you can reference and cross compare across the different systems you already use. So more on that. Keep an eye on the website on the launch notes. We'll be coming out with that in a couple weeks. Nice. I thought I heard it to dumb there for a minute. This viewer is asking, do you have any advice for cleaning, normalizing, and categorizing data so that it can best be used by an agentic estimator? Good question again. Yeah. The easy answer is just to have an AI do it for you. That's that's a little easier to say than than actually accomplish, but it should be part of the discussion. I mean, there's not too long ago, we were hiring interns to do this exact task. But if you can get that information and you can get it to a super intelligence and then have an export that you can upload, I mean, you're you're gonna cut that down by 90%. Yeah. A lot of that could be done manually. I mean, you wanna make sure that cleaning probably makes, like, fixing errors. Right? Removing a lot of duplicates, scaling. But if you if you throw that into an AI tool, it's gonna do it a lot faster than an intern might do that. But if you just give it the prompt of like, hey. I I would like you to clean, normalize, and categorize this data so that I can have an agent, a guestimator, use it, ask me any questions that you have that are outstanding before we get started, and have it ask a you couple questions so it clarifies and then send it off to the races. That's what I'd do if I was in your shoes. Very good. This viewer is asking, what's the next step for a company who how some connected API, but haven't integrated AI into any of it? I would say your your your role today is just to be aware of what's coming. I don't know if there's too many people out there who are actually hitting home runs with this stuff yet, but, you know, design it, keep it in your mind as you, you know, new problems come up, new vendors knock on your door, solutions. You know, that's the stage we are at today. It's moving so fast and changing so frequently. You're gonna have to find you're gonna have to wait for those opportunities and then and then pounce on them. And I'd say too, like, it's great that there's connected APIs. I would go a step further and ask if those APIs are consumable. And what I mean by consumable that is that they're readable. They're understandable. They're not behind a paywall. And then there's that second piece of integrating AI into it. Wanna make sure that the AI has the context because an API can return fields for you, but not meaning. So MCP is important to ask about. And then what kind of AI? I would be curious about that as well. Is it an AI that, like we've talked about today, that is acting on connected, governed, structured, clean data, or is it just, a a GPT wrapper? There's there's a lot of nuance to what that even means with AI. Does it understand your workflows? Does it actually provide value, or is it brittle? Is it hollow? Does it hallucinate and you end up arguing with it? There I would kind of get a feel after a little while of whether or not you trust that or you're getting good results out of that. So couple couple different layers to that question. Make sure that that API is and the AI is of actually good has good quality tech behind it. Great. We have time for a few more questions. Is there a workaround of how we can use agents if different systems are being used? You know, we talk about APIs, and the a is important in that in this discussion. But what you're starting to see is that these frontier models can just act as a human, you know, potentially just moving your mouse around and clicking things. So, yeah, there always be workarounds whether that that efficient in the long term plan, no. But, you know, I can go fire up a bot and have it log into a website and pull information. Yeah. 100%. That that's exactly what I'd say. Well, interesting that you mentioned frontier models. This viewer is saying nobody outside of a few companies are building frontier models. So why would a smaller nimble startup with a good idea that leverages a frontier model as part of the stack be a problem? I don't know that it would necessarily be a problem. We use the frontier models as part of our stack. I think that we might see more companies adopt these, sort of untrained frontier models against their own data to where we see more nuanced frontier models and LLMs come out of different industries. But, yeah, I agree with you. It it costs billions and billions of dollars. Even HTSS spends an exorbitant amount of money, maintaining and building software. So, I don't think that a small nimble startup with a good idea, needs to pretend to be a a competitor with this frontier model or is necessarily a problem. Good. Another go ahead. I'm sorry, Michael. I just said well said to Mikaela. This viewer is asking, talk about an AI agent will be selected from other competing agents. Will AI agents be sold as specifically trained for construction related work? I I could definitely carve out something for construction due to the fact that these projects are novel and that they tend to be different. Right? That's why we have a whole architecture and engineering side of it. So I could definitely see a lot of companies just having specific construction like construction clot or construction GPT or or something like that. Also because of, you know, if you've seen a huge PDF of plans and specs, like the average context window struggles to handle a 900 page set of plans. And so it's actually something that I've been looking for someone who really identifies in the nails that aspect about construction. Yeah. Well said. Good. Another viewer is asking, most of our current vendors are way behind on tech and AI. How can we implement AI with older tools and tech? I mean, you're gonna have to decide, right, whether you continue the relationships, you know, whether it's worth it or not. You know? It it it's going to get expensive. I mean these these bots will start to you know add up and and become something that you need to manage. Whether it's worth it to replace something that cost you $500 a year who knows That's up to you. But I think long term, the more you invest in companies that are going to be focused on this and take it seriously, it's going to be better. Good. This viewer is asking, how do you suggest we protect ourselves from some of the overcommitment into AI as a new technology that will have some failures as it develops? Yeah. That's a great question. Michael, I'll let you answer that too. But I if it was me sitting in your shoes and not on the vendor side, I would say, ask for road maps and ask for when is the next give me a date for the when the next thing you're doing is releasing. Right? If if you're getting a little bit more like, oh, it's down the road. It's later. I would question that a little bit. If you get concrete dates, probably a good chance that they've got that in the plan. They're working really **** ** it. If they give you multiple dates too, you know that this is a long term strategy. Right? I can think of five different dates off off top of my head right now where I know we're putting specific things out because we're working on it, and it's it's part of our every breath every day and what we're working on. So if they're not giving you specifics, I that would give me a question mark in my head. Yeah. I'll add on, you know, if you're able to silo things into a pilot and try things out, that's great. But also just be be prepared to fail and be prepared to fail fast, right, and move on. If you do see something there and potential, like, embrace it and actually become a partner. These companies, especially the ones that are here in the Bay Area that know nothing about construction, they really, really, really like to talk to you and understand because they might be coming from a world that's just an idea that they learned at Stanford or Berkeley, and they got some money from a VC. And you have a great opportunity to partner with them and drive some of their decisions. That is sage advice. I think we have time for a quick answer for a quick question. What goes into building a data connector if the system has an API, if it doesn't have an API? I don't know, Mikaela. I've never built one. Have you? The what? A data connection or an API? Yeah. Okay. So if a system does not have an API, you wanna ask about SQL access. You wanna have some sort of middleware whether that's I mean, there's a a dozen tools out there. A lot of our customers use Boomi or Power Query. There's Workato. There's a lot of different, middleware connectors out there. So you wanna actually have either SQL access or API access, oData connection, that would come with APIs. I mean, there's there's a lot of different ways to hook into a piece of software. So I would ask a vendor if they don't have an API because APIs that's where I would start. APIs are gonna be the best way to get data in and out of a system. But, if they don't have an API, I would ask them directly, how do we get data in and out of your system? How do we integrate with other systems? What's your connection strategy? Great. Well, that's all the time we have for questions today. Please join me in thanking our presenters, Mikayla Hallowell and Michael Rodino, as well as our sponsor, HCSS. If you have any additional questions, please email them to webinarsbnpmedia dot com. We'll share those with our presenters so they can respond directly to you. Now, when you exit the webinar, you'll be taken to a brief post event survey. We'd appreciate you taking a moment to share your feedback, as it helps us to keep improving our programs for you. Please visit enr.com/webinars for the archive of this presentation, to share with your colleagues, and to find out more information about our upcoming events. We hope you found this presentation to be a good investment of your time. Thank you again for joining us, have a great day.