Agentic AI is the Future of Construction. Is Your Data Ready?

Well, great. We're at two pm New York time, so we'll get started. Before we get started on our webinar, just a few quick housekeeping items. I am Aaron from AGC of America, and if you have any questions today, free to drop them in the Q and A section on the right hand. If there's any technical issues, send us a chat and then we'll troubleshoot from there. We will share a copy of the presentation by the end of the webinar. However, I highly encourage you to stay so that we can ask questions live here. You'll also receive a recording in three to five business days after the webinar. And I do see a hand up, so if you have any questions, drop them in the Q and A and we will get to them. And yes, if you're here for getting your data ready agentic AI for the future of construction, you're at the right place. I'll hand it over to our speakers, Michaela and Michael. Thanks, Aaron. All right. Good morning, everyone. Thank you, Aaron. Thank you everyone for carving out an hour to come and chat with us. This will be a working conversation today between myself and Michael to talk about what actually works, whether AI actually works for contractors and the data underneath it and how that dovetails together. So quick introductions, and then we'll we'll get into the meat and potatoes. My name is Michaela Halliwell. I'm the group product manager at HCSS for our platform data integrations and AI. My team owns a lot of the plumbing with our software, so our APIs, our integrations, our data products, and our AI portfolio and studio and agents. And a question that I get asked probably on a daily basis from half a dozen different directions is what needs to be true about a contractor's data for AI to actually be useful? And so that's what we're unpacking today. Michael, you live on the what I I consider the customer side, but really the buyer side of this market, which is why we're really happy to have you here. So introduce yourself, tell people a little bit about you, and give people a feel for what McGuire and Hester looks like on a given day. Sure. Yeah. I'm Michael Ridino. I'm the director of information systems here at McGuire and Hester. McGuire and Hester's self-formed contractor. Hopefully, mimics some of you out there. A hundred year anniversary this year out here in the Bay Area in California, right across the street or excuse me, right across the bay from some of these AI companies. So we take it pretty seriously and happy to be here and share some of the successes. Appreciate that, thank you. So here's the shape of the hour and how we're going to be approaching this. We're gonna talk about the industry and what's actually broken, what we need to be fixing and the problems that you guys see on your end. And then second, those traits look like, the four traits of AI ready data and how you can score your systems against those. And then third is how you actually get there. So including the questions that Michael asks vendors before they actually get a seat at his table. And then we're gonna hold fifteen minutes at the end to show you a checklist of items, give you time for you to ask questions. But as Aaron mentioned at the top, please drop anything in the chat if you have questions as they occur to you. If one is too good to wait, we'll hop right on it, we'll take it in a moment. But before we go anywhere, I wanna know who's with us and who's in the room. So there should be a poll launching on your screen here in a second, But go ahead and go ahead and click that. The question is, where is your company on their AI journey? And be honest, there's really no wrong answers here. If you pick e, what is an agent, you might be the most honest person on these calls. But Michael, before the results land as people are responding to this, a year ago, where would McGuire and Hester have voted on this and where would you vote today? A year ago, I think we would be right in the middle of exploring. Obviously, this industry is going incredibly fast and changing almost every week. So compared to today, we have quite a few pilots, one or two active AI tools being used. So we're closer to the top, but it shows you just how quickly these things can happen. You are setting me up for success on my transition here because that that is very much the point. You guys can see the results here. We expect kind of that bulge in the middle. Right? A lot of people still discovering, still getting into this. Some folks on on the even more nascent and I wouldn't say immature end, but earlier end of this, figuring out what this means and and how this impacts them. Then other people who are well on their way on the graduation path, trying trying some new tools, experimenting, starting to ask the right questions and other people who are just ten toes down AI AI tools and production across multiple systems. So really really what I kind of expected to see. But like I said, setting me up for my transition here because we've had a bit of an AI agent moment in the last, I don't know, what, three, four years, but really an agent moment in the last twelve to eighteen months. And quick definitions because agent is really an abused word in twenty-twenty-six. When I'm referring to agent on this call, I'm not talking about a chatbot. I'm talking about the user is no longer the point and click of a task, the doer of a task. They are an approver or an orchestrator of many tasks across multiple systems. So an agent is an LLM tying in, an AI tying in, and accomplishing workflows on behalf of a user. It reads from your system's data. It reasons over things that it can find, and it does work on your behalf. And that progression from a lot of folks being familiar with ChatGPT in the beginning, Gemini copilots. Now Claude is very, very much a front runner in these LLM models and these frontier models. That progression from just going to, like, a question answer turn based wave to now where AI is acting on your behalf as a user, and it is making intelligent decisions and accomplishing things on your behalf. It's really been kind of a fever over the last couple of years for us on the tech side and I'm sure in the actual industry. Where have you guys felt that at McGuire and Hester and what has that maturity path looked like, Michael? Yeah. I think it's exactly as you described. You start out with everyone starts asking you all these questions about AI and, hey. Can I can I get access? Can I start doing this? And you eventually give in and get it approved and and kinda hand hand some people some tools and see what they can do. Usually, that will revolve around, like you said, the chatbot asking information, uploading plans and specs, going through an enterprise level chatbot where any employee can now start to get information about the company. You know, what's our dress code policy without having to make a phone call to someone. I think that's that's kind of the base level. But that's slowly or quickly gonna transition into, okay, we're going to incur some costs and we're going to start to drive an ROI. So how can we get these AIs and these agents to actually start to augment our employees and and start perform some of these tasks. So that's where we are right now. Trying to supercharge all of our employees, you know, across all our different our different work streams at McGuire and Hester. So whether it's estimating the resource department out in the field, back office accounting, almost anyone at this point theoretically, you know, you give them an AI and hopefully you can ten x their output. I mean, I know it's definitely ten x mine. I I had the thought yesterday of I don't know how I got so much done ten years ago. I would have had to write all of this stuff, or I would have had to take a lot more time to think through how to organize something or I would have would have had to go and manually hunt down from six different systems to find the answer I'm looking for. Whereas now I can ask a quick question, and there it's in front of my face, and I can strategize and move forward. So I I know that that is probably felt across the room and for the those online listening as well. The you mentioned the estimating and the, like, preconstruction workflows. I I kinda see those as the entry point because there's so much document review and scope generation where estimating is where a lot of the work maybe not begins, but a lot of that work is kicking off, right, with takeoffs and the gains are very dramatic when you have a purpose built tool that is able to drop your work from thirty to forty hours to, I don't know, thirty to forty minutes in a perfect world and identify risks upfront. Does that track with what you are seeing and what you expect as well? Yeah. And in fact, it's one of my favorite spaces to kinda match these tools up with our estimating staff. You know, I can walk through the halls and see them all pouring over plans and specs and TEMs like, that would be a perfect opportunity to send this into an AI. So that's that's been at least recently. One of my main focuses is trying to unlock some of the time for those guys. Trying to get to them so they're having more intelligent higher level conversation about our pursuits about our proposals, and less, you know, scanning through papers and trying to find individual data points. Yeah. And speaking of data points, if if precon and estimating is one of the entry points I mean, talk about operational data here for just a second. I it's something that I find genuinely frustrating because construction generates more operational data every day than I think any industry on earth. Every crew, every shift, time card, every telematic from equipment, every fuel dispense, every production quantity log photo, and a lot of it evaporates or sits there and there's nothing done with it. And we're an industry where we are building infrastructure and we have an underbuilt data infrastructure, which I think the irony of that is so so interesting, and it's a gap that I think a lot of these agents are going to start to expose. So in talking about that agent moment and what the reality looks like in heavy civil, can you speak a little bit to that gap and and what it is that breaks down at McGuire and Hester or what it is that you find frustrating? I know we got a couple things on here that we've talked about during our pre call, but can you speak a little to that about the amount, the volume of data we have access to and where that has a potential to break down? Yeah. I think access is the key point that like, you know, all your systems need to be accessible and they've all been designed for the past thirty years to be accessible to human beings and now they all need to be accessible to AI. They need to at a minimum have some sort of connection point where another piece of software application could connect into them. And then hopefully an AI can can start to talk to these systems and digest all that information. Because you're right, we do produce so much and because it is construction, know that project you built might not be relevant to anything else built out there. And so all this information you produced just kind of will sit there and eventually someone will hit delete on it and it'll be gone. For me, know the real value I can provide is what I call closing the loop and it's taking all that operational information and turning that back to our estimating our proposal departments and providing them with information to then go win better jobs. So closing that loop is really where my focus has been on the operational side. Because I believe in the future that construction companies, the ones that are successful, are going to really nail that loop. And whether it means that you're just operating with a super intelligent AI to bid your work and and turn it over to operations, or you're just augmenting your your your people out there. We'll see. Yeah. I mean, that's kind of the my thesis for the hour is really in an an era where we have all these commodity AI tools. The and I'm speaking from the software vendor side, so it's a little bit of like a gym telling you that you really have to exercise. But the the importance and the people who are going the vendors who I think are going to win and enable the contractors to be able to do better are the ones that actually have that connectivity. You you mentioned that there's this this data disconnection and access problem where estimating might be in a system, field might be in another, accounting is in another, but the answers live in all across all of those. Right? You said something to me during our prep call about how you don't wanna hire bodies to move ones and zeros between systems. Can you can you speak a little bit to that and connect to that to where your head's at and where things are headed and what good actually looks like. Yeah. Moving ones and zeros by humans is just really expensive, especially here in the Bay Area. You know, I think today in twenty twenty six most applications should be at a minimum using APIs exposing that stuff. My job as the person making these decisions to only bring in vendors that enable that. You know, because the quicker, the less friction we can have in the system, the quicker we can turn around answers, get it to the people who can make those decisions. It's also becoming like you said commoditized. A lot of these applications we see fundamentally aren't that special, and so you can envision a world where you have some fairly intelligent people on your staff. You can build out your own tools. It's all built on the backbone of the internet. It's an open APIs, they can start communicating, you can just see this thing take off. Yeah, one hundred percent. Problem is maintaining it from there, but That is always gonna be a problem. Gotta make that You're one hundred percent right. If the access is there and the consumption and the skill, makes us a whole different ballgame. I would almost qualify, like, I love your statement about you know, asking if organizations have APIs. Do you find that having APIs is enough, or is it consumable, readable, easy to work with APIs that aren't gated behind payments? Is there more to it just than than just having an API? Yeah. I mean, you really wanna embrace it. You know, you saw it a couple years ago with marketplaces and people partnering with other applications. I think that's all been blown up and and, you know, thrown out there to where I want it to be connected to the latest and greatest AI. Right? That that that's what most people want. The farther you slip behind that, the farther you're gonna be behind your competitors because a lot of a lot of companies are just embracing that right. Because it's coming from the users right now. They're all have their personal subscriptions to cloud or chat GPT or Gemini and that's those are the tools they want to use. They know they they see it on the news. All these crazy things they're starting to do. And so being on those frontier models can really pay off any friction you put in there. Like you said, any gatekeeping to opening up that data is gonna be, to me, a not lead to success in the future. Yeah. It's just one more friction point in an already technical utility of software APIs are already technical enough and over complicating them by not making them usable consumable, a little bit of a side passion to find. So don't let me get on a soapbox about like, consumableity of APIs, but here's how we think about data at HCSS at least. And this comes from our long term data strategy, not from a marketing deck. There's a lot of thought behind There's a lot of conversations and hours of conversations on what good actually looks like. If we think of data and you take a sort of aerial view of a construction site and then pretend you're a drone for a second and looking down construction point and a construction site, think of all of the points of entry in which data goes into an Internet of things, an IoT system. And you've got, you know, trucks coming in with with different production quantities, with dirt, with whatever your your guys are bringing on the site that day. You've got drones sending aerial photos. You've got people punching in, punching out. You've got equipment going off the site, on the site all day long, and all of the things that I mean, cameras on-site, infrared, whatever it might be. All of those data entry points and how they come into a computer become a system of record. Right? So data starts as a record. Something happened in the real world, and it gets written down into a computer. That then turns into some sort of structured digital asset, whether it is consistent, coded, comparable to something else. That connection then has the potential to become intelligence because now your estimate, your job, the actuals can inform on one another. Right? And so then there is governance, a governance layer that your tech companies are responsible for. Something that turns intelligence into an infrastructure where it is reliable enough for you to build business processes on top of and make future decisions about how you build, how you build power, how you build water, how you build bridges and roads and airports and what have you. And then that record asset intelligence infrastructure, that's how we climb the ladder. That's how we get better at this. So AI ready data in h from an HSS perspective has these four traits. It is granular and structured to get you from record to asset. It is connected to get you from asset to intelligence, and then it is governed. And that's what makes the infrastructure that you can actually trust. You can score your systems against and know exactly where you stand, exactly how to fix it, and how to move forward. So my question, Michael, to you after that mouthful that I I I just said, when you hear and see these four things, which of these is the easiest to resonate with and which is the hardest for McGuire and Hester and and Michael specifically? Honestly, like, which one would which one do you pick as, you know, the easiest to connect with and which is the most unattainable currently? I mean, the easiest is granular. People like to store information, especially when there's no cost to it. Right? It's uploading gigs and gigs of data to a job folder. It really there's still really a huge role for number two in having it structured. If you point an AI at unstructured data, it tends to start to cause a lot of errors and actually becomes very expensive. So if you can keep things structured, you're going to see the payoffs. For me, number three is the hardest connected, right? I need to work with multiple departments. I need to work with our infrastructure people to physically open up our systems, you know, firewalls and all that. I need to work with vendors to make sure that they're pursuing the same goals. And so, you know, as my role as information systems director, like, I am I am looking to make sure our architecture is gonna meet that connected environment. And then governance is kind of like a corporate level, higher level discussion on on who we want to see what, where do we want to store this information. Again, remember we're we're a we're a self perform, you know, mid sized contractor in the Bay Area. We're not a massive national player. We don't have a huge team of risk people and and and those folks looking at that. We're utilizing a lot of vendors and and trust and Google, right? We're we're going to assume that they have the best governance and and all those policies in place. But so, yeah, for me personally, the easiest granular people are gonna collect as much data as you give them, and the most difficult is making sure everything's connected. Let's talk a little bit more about granular. You said, you know, it's easy to collect granular data. If you give the people the access and the opportunity, it's gonna come in. Granular meaning the level of detail under that roll up existing, you know, this excavator cost forty grand last month can answer the what, but it can never answer the why. So can you speak a little bit to how granular data gets analyzed and dissected and how the level of detail matters, when an agent has to earn their place in answering the why. Yeah. I mean, if there's not that level of fidelity, then then there's no chance of you having the correct answer when it comes to a human being or or an AI. So we need to make sure we're at the level of detail that we want to to answer those questions. So if you, like our in our example here, want to know what it takes your cruise to dig a trench a hundred feet, then you have to be tracking at that level of detail. What does that mean? That usually means, you know, it has to start from the very beginning. Your estimator needs to have a have a a budgeted cost for your digging activity, and that need to makes it make its way to the foreman so that they can actually then track that activity. Yeah. Consistent code books, consistent data collection, and it only pays off if it lands, consistently, which is the next point, which is structured. Let's talk a little bit about that structure. When I say structure, I'm saying that is what allows an AI, an LLM, an agent to ground its answers in real fields instead of assuming regressing from the pros and context around. So a time card that says Michael worked on the culvert forces a model to guess, but one that says, this employee did this cost code for this many hours on this date for this job at this location can now compute. Right? So this is the difference between an agent being deterministic and nondeterministic as well, which we all know if we go put in the same question into ChatGPT in two different chats, it's gonna give us two different answers or across ChatGPT and Anthropic tools or Microsoft tools. We're gonna get different answers. So that's their nondeterminancy. But if you give it a lot more structure, if you onboard that agentic employee in a more consistent structured way, it's going to result in less hallucination and one that actually cites your actual data and your records. So, Michael, you mentioned on our pre call that you're a structure guy, you know, work breakdown, cost breakdown, resource breakdown. Which one do you think that contractors in this this industry sleeps on, and what does enforcing structure really take? From what I see usually the resource breakdown structure is the loosest, but probably the least important when it comes to what we're trying to do here. Again, going back to closing the loop, it's going to be all about your work breakdown structure because that's what's going to feed back into your estimating side. I think at its core, know, as a as a competitor in the space, your work down your break work breakdown structure is going to be one of those keys because that's if you are self performing, you know, it's it's where you make your money. Are you better at digging trenches? Are you better at laying concrete? Are you better at doing landscape work? And that's going to be reflected in the structure. So if you can build that up, if you can start to collect that information at that level, once you turn that over to the AI, once you turn that over to your super powered estimating teams, you know, you're gonna be able to really focus where you can make some money. Yeah. Agreed. Let's talk about that closing the loop a little bit, that connected. Right? Connected. So when I say connected, I mean in one breath. If a bid becomes a job, a job becomes an actual, an actual feeds the next bid, and all of that syncs through your ERP and accounting as well. Every job you finish should make the next one smarter. That's that flywheel effect that we will talk a lot about. I mean, you'll hear a lot about it industry wide and it compounds. So the data that feeds the AI, AI is then going to produce more data, which is then good. We're gonna go circularly around this is gonna compound. And the longer the history, the better it joins, the harder you are to compete with it. And then once that loop actually closes, the more interesting question moves upstream, right, which is how should we estimate this? Should we pursue this at all? And I'm just gonna leave that there because the specifics on how contractors actually wire that loop are exactly the type of competitive advantage that everyone's gonna take a different tack on. But, you know, Michael, without giving away the playbook of McHugher and Hester directionally, what steps do you take to close the loop on the job and within how your systems work? What questions do you go into with vendors and say, you know, here's what I'm going to figure out if it works for us or not? Yeah. So I have I have a couple of requirements that if I get my way are dictated when we do make a vendor selection. So a big one is how connected or how open is your information. If you expect me to fire up a server on prem and install something, you're you're less likely to get a contract with us. You know, I want things open. I want things out there so that I don't have to pay someone to move a data point between a and b so that I can hook into our internal AI that we're developing to close the loop and get that information back to people who need it. It's becoming more and more critical. These things are increasing costs and spend each day and so we're working super hard to get the value out of them. But yeah, we're we're not we're growing out of the world of oh so and so is used to this piece of software, so we're going to keep keep it in house and and keep so and so happy and into what makes sense for the enterprise. Where are we gonna get the biggest value from? Oh, yeah. I think you have to ask that question, especially looking at your goals long term and and thinking about a more strategic footprint. Right? And then lastly, the last qualification or the last quality of getting your data ready is that that data has to be governed. Our rule, at least at HSS is simple. If an agent is going to inherit permissions of a person asking or running that agent, then there it needs to inherit the the permissions of the person doing that behavior orchestrating the agent. Nobody should have access to something in an agent that they wouldn't outside of the agent. So we have tenant isolation, role based access control. These are not features to us. They are fully fledged product teams. They are prerequisites, are requirements. So because if an agent can surface something to a user that a user could not have opened for themselves, that is not intelligence. That is a breach. And where I would push everyone listening is to be very demanding with every vendor, us included, HHS included, and ask us exactly how does your AI decide what a given user is allowed to see. There was a question sent in prior to the webinar around, safety data. And, if if there was any advice on the webinar about, how agentic AI can be used by safety professionals for regulation compliance, safety manuals, safety data sheets, hazard recognition, updates, all these type of things. Be be cautious, be careful, be intentional, draw hard lines. The governance is the piece that I can get super nerdy about, but I won't for sake of time in everybody's brains. But take this seriously. There's PII data and safety that you do not want to be leaking. You wanna make sure that the permissions and the structure and the access control for these things are set up in a in a in a conservative way because these are people's lives, these are people's information we're talking about. So that would be my advice. Michael, can you speak a little bit to the hard lines that McGuire and Hester draws and give the audience the guidelines you gave me on our prep call about the homework that you have to do around some of this. What does that look like for you? Yeah, my hardest battle for all this was actually getting past our head of legal and getting approval to allow our our rank and file employees to start using AI. And she has absolutely the valid concerns and and and things that need to balance out my my goal, which providing access. But down there, can see one one of the things that you can take away from right now is I'm assuming most people are going to be on Microsoft or Google or one of the the big players. Fire up Copilot or Gemini and start asking questions what it knows about the enterprise. You'll quickly see how well you are governed. I like to put in things about our higher ups, know, and ask it what the president of the company's driver's license number is. It should not tell you it. If it does, you have a problem, but AI will quickly expose all of those issues. It can also be used to solve them. If you do it intelligently, can can. You can figure out where it is and plug those holes, but at no point should you ever not under not remember that these are human beings and this information can't be shared. There's financial risk. There's, you know, you don't want to come off as one of those companies that plays fast and loose with employee information. So when it comes to safety, when it comes to HR, you know, put those things behind a firewall and do your best to not expose that information. Don't let an AI digest it. Make sure you're if you are conservative around this and I think construction at a whole will be, You're not developing out on those free Chinese LLMs that are out there. You're probably with one of the big players, so you have all those tools, you have those guarantees and you should be safe. Yeah, absolutely. I can't stress enough about how much I see pop up in the industry around this and I started my career in cybersecurity, so this one gets me a little passionate because I know that there is risk out there and I want I want people to be careful. I want organizations to be careful and hold their vendors accountable for how they are structuring and how they're being intentional with building agent infrastructure. And maybe not the sexiest parts of agentic AI, but definitely in my opinion, the most important. So time for another pop quiz, Nobody in the room is safe. Nobody's ERP vendors watching you there probably. But question here, can your core systems talk to each other today? Yes. Most are connected by APIs partially with some black boxes. No. We do everything manually. Not sure who has APIs and who doesn't. And then what is an API? I'll give everybody a minute there. Michael, you do you have any how how would you answer this a year ago versus versus now? Yeah. We're we're definitely with the crowd here in that partially. We still we still have some black boxes, some self inflicted, some just the vendors haven't been playing ball. That's about the same as we were last year. We're working on it. Yeah. I mean, I think that that's probably I mean, that's true for us at HSS as well. I mean, we've got we we could probably answer this poll in a very similar way as well. There there are some vendors with APIs. There are some that have the APIs, but not with the data that we need, or that there are some that are connected and some that are not. And there is definitely some manual moving between systems. Have you answered what is an API? It stands for application programming interface, and it just means the way that computers talk to each other. So if if you imagine yourself in your favorite restaurant and you order your favorite dish and you tell the server or the waiter, hey. I would like this dish. That server waiter is gonna take that back to the kitchen and then bring you back your food. An API is a waiter, a server in between systems. So you say, hey. I want the time cards, and it brings you the time cards. Hey, I want this information about this employee and it brings you the information about the employee. It is just a little moving car in between and how computers talk to each other. So that's quick and dirty and nerdy about what is an API application Programming Interface. All right, let's move forward on our slides here. We got a little bit of bubble in the middle of that, most people saying partially connected and then yes, they still move manual data. So let's talk about the vendor test here. Michael, you mentioned you have questions that you ask on every vendor sales calls, including the ones you have with HSS. So what is that and what happens to vendors who cannot answer it? Yeah. Well, usually they go lower on the list. I can't always say that they go to the bottom. But yeah. I mean, an API at a minimum, twenty twenty six, you gotta you gotta be able to connect your information. I I would like to see even in the past three or three or six months, you see more and more vendors just connecting directly to AIs to the MCP, which is fantastic. Again, it it allows you to use those tools. A lot of companies spent the past two or three years building out their own internal integrated API or AIs, and those are great, but they're always going to be limited. So that's going to be a part of any call. I would say nine out of ten are are on board at this point. There's only a couple of them that really haven't embraced that. I can actually remember being on a HSS call at one point and asking you guys about APIs and it was long ago where I don't blame you for not having an answer, but I'm glad to see the progress you guys have made since then. Oh, thank you. They brought me on as the first API product manager three and a half years ago, so maybe it was your question that got me Well, it coincides exactly with when I made that phone call. Just kidding. I wanna build on that with kind of a maybe it's an uncomfortable truth, maybe it's just a reality from someone who builds construction software on a daily basis for a living. Companies like HTSS are not anthropic and open AI, and we are not going to pretend to be. The software vendors who are pretending to be should worry you because there is a lot of AI being bolted on to products just for a headline moment, a LinkedIn moment, a marketing moment, purely so the word AI or artificial intelligence can appear on a slide. When you evaluate a software vendor, you need to ask, is this AI connected to my data? Does it have the right permissions, or is it a chatbot wearing a hard hat? Right? Our job as vendors are different from the labs and the frontier models. And I would argue maybe even more durable because it is the open structured governed layer that whatever intelligence you choose to act on can impact how we build the world. Right? And some people on this call will use AI that we build into our products, and it will be great because it sits on top of structured data. Others will standardize it on ChatGPT or Claude and expect systems to be reachable from that, but both have to work. That is why we at HSS at least invest in APIs and there's questions in the chat about MCPs. And, yes, we are investing in MCP and these open standards that allow Claude and other tools to connect to systems like ours because an agent built in a silo is only going to give you a faster silo. So make openness a line item in your spreadsheet for a procurement. Put things in every RFP, like full API access to your own data without a tollbooth, data export in usable formats, permissioning models that AI actually will respect, and support for that open connection standard. Do they have an agent marketplace? Are they building one? Ask to see their API documentation before you watch the demo. The vendors who will have that will be delighted that you ask. I promise speaking as a vendor who gets tickled pink if when people ask about our APIs. The ones who do not will will give you the runaround. Right? So let me me show a little bit about what's going on in the HTSS world, and this is my one minute where I I get to have a little bit of a show off moment and be very proud of what we're building. On the slide here, this is really a picture of the HCSS philosophy. So our estimating field fleet all lives on one platform with reporting analytics, AI, and integrations as the foundational necessary layer. That placement is deliberate. AI integrations are foundational. They are not features. They are fully fledged product teams with product managers behind them. It is why bid becomes a job. A job becomes an actual and actual feeds the next bid without out anybody doing anything manually. And Michael, you're the customer here. So greatest honestly, has openness of our APIs worked for you and where do you see the that you would you are hungry for more? Yeah. Absolutely. You know, one of my quick tests when I'm I'm looking at a new vendor is do they have a a dev page? So I don't know. I think it's dev dot hss dot com and having those open and published is exactly what I want to see. Would I like to see a dispatcher page added there? Yes. But for those who know HCSS, it's it may not happen, but for the rest of it, absolutely. You guys have nailed it and I think it is should be at least the standard that you hold any vendor. Yeah. That URL, if you're looking for it too, is developer dot h t s s apps dot com There you go. Is our open developer portal, and he is correct. We do not have dispatcher APIs, but we do have ways to sync into desk dispatcher that aren't APIs. So thank you, Michael. I appreciate that. Couple things to walk away from this with, and then we're gonna open it really up to questions. So if you have some for Michael and myself, I see there's already some in the q and a. I will get to it in just a second, but if you're thinking of it right now, go ahead and throw it in for myself or Michael. We'll be happy to take your questions in a second. Couple things I want you to walk away with here. One, there is no AI without APIs. There is no intelligence without APIs. Make openness a procurement requirement. Two, score your systems against granular, structured, connected, governed. Three, close the loop. Your history is the moat the moat that protects your castle, but only if it actual actually compounds in that flywheel effect that we talked about. Four, govern before you scale, wall off your payroll, wall off your HR data, wall off your PII and run these query tests that Michael mentioned this week. Do it today, do it this afternoon. Michael, last word before q and a, one sentence for someone who is watching this and still kind of resistant or hesitant to all of this and this agentic moment that we're in right now. I would say embrace it. Feel lucky that you're in construction. Know, my world is being quickly upset as a information guy, but we are close to an industry that is fairly protected from all this. Very much so. That's great. Alright. Let's take a look at some of these questions we got here. I'm gonna start with some of the earlier ones. Does HHS have Claude connectors already created? No. We do not. Will we in the future? It depends on what you mean by Claude connectors. Will we have MCPs? Yes. We will. Connectors, when you go into the connections tab of Claude, we are we will be working with Anthropic on that in the future. Yes. Does HHS have a published AI related capability roadmap published maybe quarter by quarter or horizon? Published? No. Shareable where I would show you in a heartbeat? Yes. Absolutely. I have a a road map that I would pull up for you. If you get my contact info here, I will send it to you. And it is a now, next, later format, not quarter or horizon. And so when we say now, it means we are actively working on it. When we say next, it means twenty twenty six, and when we say later, we we mean twenty twenty seven and beyond. Does HHS have any high level published case studies using value stream, bid cash development value stream, DBS mapping with agentic AI, HSS tools, and others? If so, has the data been used to show on Versus or DBS maps? I'm gonna be honest with you. I don't know the answer whether we have this or not, but we do have some folks backstage that will link our case studies that we do have available to this answer so that you can see them and go check it out for yourself. Joshua asked, what AI systems are you using to talk to HSS and accounting? Also, accounting software are you using? I think that one's for you, Michael. We are a Google shop, believe it or not. Not very common with construction companies I found, but we have started to expand and allow different use cases depending on the individual if they wanna use cloud or they wanna use ChatGPT. We have a couple of smaller specific vendors as well, one for proposals and TakeOff as well. I won't say our exact accounting software because I don't want to talk negatively about anything, but it's a common one that probably plenty of people on this call use. And they have they are one of those that I do feel is behind and could do better in opening themselves up. Yeah. How do you keep your data safe if you have open info? Leslie asks. I mean, it should be part of any new hires kind of new hire packet they're onboarding and having them understand the potential dangers they have if they do go out of policy. You know we have a policy in place about sharing information both internally as well as information we receive from proposals. You know, if you're under NDA and all that good stuff. So we're handling it kind of at a employee level policy. We don't have like an agent that's watching for this stuff and looking to see if it gets exposed. The other safeguard you can put in place is that most of the big companies are going to guarantee your information isn't shared or used to build their models if you are on one of their paid plans. Alright. Next question is I answered in the chat, but do we have plans to add MCP and HSS? Yes for all products. Does it make sense to invest in structuring data, or do you believe it will be solved with agents in the next two years? That's a good question. That is a good question. Oh, I I am going to say it it makes sense to invest in structuring your data regardless because it depends on how granular your data is. It like, I'm I gave that example. If you do not have context and meaning behind some of these data points, AI is still going to hallucinate, is going to be nondeterministic. I think the more structured, the better positioned you are for the agents to be more informed. It's the same as like if you think of an agent as a junior employee, you you got that intern is real eager, really excited, is gonna be real positive and upbeat on their in their first six months with you. And depending on how well you train them and how well you structure their onboarding and their training versus just saying, alright, I hope you can swim, jump, jump in the pool. You're going to set them up for success. The same is true of an agent. The more you structure your data, the more you prepare it, the better that agent is gonna be from day one. And wouldn't you rather have it working better on day one than playing catch up when all your competitors have structured data? That's my just that's my two cents. Michael, you have any thoughts on that one? Yeah. To me, it's you know, if you ask it today, you upload a set of plans and specs to any of the big players and say, much is it gonna cost me to build this job? It's gonna use what it knows, which is basically what's been published to the Internet at this point. Compare that to if you had a structured data set that you could feed it that says, and here's what we have historically spent to build these like projects. And so having that structured data is going to allow you to get way better results. Agreed. There's a question in the webinar chat real quick that I wanna take. Is HCSS connecting with Bluebeam since you are co owned? For those on the call that don't know, HCSS and Bluebeam are owned by the Nemachak Group, And yes, we are working very closely with our Bluebeam counterparts and friends on that side to build integrations and build connections both on the agentic side and just point to point integrations across workflows that make sense for our customers, those conversations and work is in the works, as we speak. So yes. Let's see. Let's look at a couple more of these. I'm gonna give you this one, Michael. For these agents, are you primarily using third party providers for the LLM and harness, or are you self hosting local models? How do you handle data privacy concerns with third party providers? So again, you know, remember that we're a fairly mid sized company, know, I'm being honest, there's only three of us in the IT department, so self hosting an LM is not really in the cards for us. And so everything is going to be one of the major players As far as privacy concerns, we just have to put faith in the Googles and the Anthropics and the ChatGPT's that they are going to live up to what they promised us in their terms and conditions. I believe that there's way too much at risk for them to have a breach or have something go off the rails And there's much bigger companies putting faith in these companies than our company. I wish I had a bigger war chest to bring to the the the conversation with these guys, but it's just it's just how we operate. I think that's probably representative of a lot of folks on the call, though. Appreciate that. Alright. Let's see here. Lots of really great questions here. Leslie also asks, we have one software that manages project cost data, including labor, which does our payroll. How do you keep that part behind the wall not AI accessible and the other cost to be gathered by AI? That's a good question. Depending on the software, I mean, if there are user access controls built into it, you can tell the AIs to respect that. You can, for instance, we're Google and Google has a lot built into the G drive. So depending on your role and your permissions in the G drive, you are not able to see any of that information. If that same schema carries over to your payroll application, then technically, could you could continue those roles or continue those rules. Yeah. Tacking on to that one too. That's Leslie, that's why we talk about governance being such an important piece, and and you'll see it abbreviated as RBAC, role based access control. It's not the sexiest part of a piece of software, but you need to make sure that products and even not just products, but features within that product and the level of granularity that people can see are permissioned properly and governed properly because you can turn an you can turn access on and off for as as deep down for every piece of information as that vendor will allow you to. So I would say that that is a great question to ask that software vendor. Take it back and ask them, how are you making sure that my payroll data is not being consumed and being used to train AI, but that our production quantities or our cost data, if that is what you want to be leveraging. How do I make sure that the permissions for those are separated and how are you thinking about governance? I would ask that software vendor. Or even go farther and say what but I do want my payroll people to be able to use AI against this data. So how can you enable that? Yeah. How do I make sure only my payroll data or only the people on payroll that I want seeing this are able to use the tools that I make accessible for them? And what are you providing and how are you keeping it separate? Yeah. Ask that software vendor. Alright. Let's see. How does HCSS handle identity off on agentic tasks? Does the agent act on its own behalf or as the user? What a phenomenal question. Love it. So excited that I have a good answer for you too, Blake. We respect role based access permissions. The the the permission that a user has in a product to do a certain thing will be the exact same access control that they have on an agent. You will not be able to get access to something that you don't normally have with an agent that you would have without it. Does an agent act on its behalf or as the user? The that is that is also a phenomenal question. We take the philosophy at HTSS that a human not only has to be in the loop, but will be dragged into the loop. The loop being an agentic workflow. An agent will not take an action on behalf of a user without a human being approving that action and thinking for themself. So the boring, the tedious work might be solved for you, but there is still a human that needs to make a decision. And it will be logged as a human agreed to this decision. This is how an agent came up with the suggestions. You can deny it. You can all augment it, or you can accept the conclusion it came from. And so there will be complete total freedom to reject an an agentic workflow and agent behavior or to change it or to roll with it completely if the human trusts it. So there will be will be a lot of autonomy in HCSS agentic flows and a lot of protection around who is able to do what. Not just anybody is going to be able to design and set up agents. So we will have access controls even around how agents get deployed, tested, and run-in your environment as well. I think that that is cutting it closer to our time. Aaron, I don't know if we have room for one more question or if you have a couple more things you wanna cover, but I'll defer to you. No. Go ahead. I think we have time for one more question. If anyone has any further question, feel free to drop the main right now. All right, let me find a really good one because there's so much gold. Answer everybody's questions. If your contact information is attached, I will follow-up with you though. Let's see. Let's take Matthew's question. What does time look line look like for HCSS MCP soon? We if you go to h s s dot com slash AI, which I'm gonna throw in the chat here for everyone, we are launching beta for HCSS AI core AI studio and embedded agents in October and fall MCP shortly following that. So keeping a lookout on that. That page is newly launched with a lot of really good go nuggets on what we're planning to do and the workflows, and you can meet the first agents that we're going to market with. So hopefully, that's been helpful. And you see my contact info on the screen. If you have a question that I did not get to, please email me. Please, please, please. I would love to hear from you, and I will follow-up with you. Same with Michael as well. If you have questions for him, I'll make sure you get set up and we get synced. Appreciate everyone for attending. Michael, thank you so much for your your time and your generosity to answer questions with real stories and bumps included. Aaron, thank you for organizing.