Latest / Engineering Influence from ACEC / How NVIDIA Sees AI Transforming AEC
Transcript
- 0:01A big thank you to Autodesk, the sponsor of Engineering Influence at this year's convention.
- 0:06Autodesk partners with the AEC industry and its leaders to shape a more connected,
- 0:12data-driven future, empowering engineers to work more intelligently and to make
- 0:17confident decisions to drive meaningful impact.
- 0:20This enables firms to evolve, stay competitive, and deliver stronger business
- 0:25outcomes, all while designing a better world.
- 0:28So thank you again to Autodesk.
- 0:31Welcome to the Engineering Influence Podcast from the American Council of Engineering
- 0:34Companies, coming to you from our 2026th Annual Convention and Legislative Summit.
- 0:39One of the issues that has been really burning for a couple of years now,
- 0:43we were at really the outset of it.
- 0:45Actually, I take that back because I was at FIDIC in Mexico City in 2019,
- 0:51and we had a former Silicon Valley exec who said that if you're at an engineering
- 0:55firm or you're at a data firm, you just don't know it yet, that was 2019.
- 0:58And now I could say that if you're an engineering firm, you're actually at an
- 1:01AI firm, you just don't know it yet.
- 1:03I'm joined by Sean Young, who runs really the AEC component of NVIDIA with the
- 1:10Institute on its Firm of the Future initiative. Sean, thank you for joining
- 1:14the program. Yeah, thanks, Jeff.
- 1:16I mean, you know, it was funny back then. They said, well, you know,
- 1:19if you're at a data firm, you just don't realize it. And everyone's like, oh, OK.
- 1:24And from 2019 to about 2020, now, 2026, now it's shifted completely to now it's
- 1:31like you are now at an AI firm. You just don't know it yet.
- 1:35How, it's not really a question of if, but when. From your perspective,
- 1:40you know, how is AI really changing the landscape in the AEC industry?
- 1:46You know, it's, what's interesting as you're talking, I'm thinking what you
- 1:51say can be applied to every company. Yeah.
- 1:53Cause like, it's like, how did you operate before you had a computer?
- 1:57How did you possibly, what you use a fax machine and telephone?
- 2:01How, we're going to say that about AI. How did you operate before AI?
- 2:06And AI is not, in the near term, something that's going to be spoon-fed.
- 2:12It's not as simple as straightforward as the evolution from paper to computers with CAD was.
- 2:20Because that was just like, go buy a computer, the software company will tell
- 2:24you what computer to buy, and you double-click on the installer and installs
- 2:28your thing, and you just read our manual. well, here's the things you can do
- 2:31and here's the things you can't do. Yeah.
- 2:32AI can do just about anything. Yeah. Which is where it gets complicated because
- 2:36it means you have to be the investigator and the arbitrator of that knowledge within your firm. Yeah.
- 2:44And there's a lot of knowledge out there. The good thing is we can all use large
- 2:49language models, chatbots to help us navigate all that knowledge so you can
- 2:54start using AI to help you figure out how to use AI. Yeah.
- 2:58Yeah, you can kind of like help it build the ship while you're trying to sail at the same time.
- 3:03And everybody kind of focuses on the large language models that are out there
- 3:07and pervasive, of course, in everything that we're seeing now.
- 3:11But it goes beyond that. I mean, it really is a question of how do you take
- 3:15all this information that you own as a company? How do you leverage that information
- 3:20into an AI model that actually produces the desired outcome?
- 3:24One thing I was really interested in talking about this issue,
- 3:28you know, a couple of years ago were how few firms in the AEC space actually
- 3:32had people in the C-suites who were actually focused on data.
- 3:36Either a chief data security officer, chief information officer.
- 3:40I mean, it goes, you know, everybody had an IT department, but nobody was actually
- 3:44thinking holistically or universally about how their data is maintained and
- 3:47security of it. And then how do you actually take advantage of it?
- 3:50Yeah. What have you seen in the industry over the past couple of years about
- 3:54how that's changed? In this industry, not a whole lot.
- 3:58In other industries, there's data scientists.
- 4:01And so firms have hired data scientists to organize all the data.
- 4:04Because in this industry, very services based, the project's done, you archive the data.
- 4:11In other industries, you learn from that data. So the data, like in healthcare,
- 4:17the data is informing all sorts of research and discovery.
- 4:20Well, it could be the same with AEC.
- 4:23You just need to harness that. So what I would say is now, you know, your data is your IP.
- 4:29It's the way you do things. The data is in all sorts of different files.
- 4:33Some of those files are software vendors files and some of it's Excel or PDFs or whatever.
- 4:39But that's okay. It used to be that you would have to get, I mean,
- 4:43used to be like the ancient history about a year ago. Yeah, exactly.
- 4:45I have to go get a data team of data scientists to figure out how to crack all
- 4:49the different data formats, normalize all the data into one homogenous data lake, blah, blah, blah.
- 4:54But now you can just have agents go and figure it out.
- 4:57Just tell an agent, you go, you know, get, here's all of my data at NVIDIA.
- 5:02We have tons of data repositories, but we have one, well, actually I have many
- 5:07agents, but all of those agents have access to all the data.
- 5:11And it just goes, it doesn't matter, I don't know what format it's in.
- 5:15I just give it a prompt and access to all of the repositories.
- 5:20It goes and fetches all the data. Well, I found this and this in Confluence
- 5:24and I found this in GitHub and I found this and here's what you got.
- 5:29And that, I think, is a good transition from looking at, like,
- 5:32okay, everybody knows about the large language models, and you kind of prompt
- 5:35something, you'll get a result.
- 5:37If you have a closed data set, you can train it and create a project that actually
- 5:42has those parameters to say that you must search within this data set with these limitations.
- 5:48That's one part of the process.
- 5:50What we're looking at in the future is really the spread of agentic AI.
- 5:53And let's boil that back for people who aren't focusing on it as much as you
- 5:57and I are. what's an agent,
- 6:00And how can it be employed by an ADC firm to actually do work faster?
- 6:06Okay, well, everybody, I'm assuming, is familiar with ChatGPT at this point.
- 6:11So ChatGPT is a form of a large language model.
- 6:15It's a model that was trained on very generic data, all of the internet,
- 6:20and it understands languages.
- 6:22One of those languages could be coding. One of those languages could be English, et cetera.
- 6:27In that language understanding, there's also an understanding of all the context
- 6:30of history and different subject matters.
- 6:34So, but the method of you interacting with ChatGPT is you type something in, which is a prompt.
- 6:40It processes that and gives you a response, which is, by the way,
- 6:45what it's doing is inference. Yeah.
- 6:46And it generates tokens. Just side note, generates tokens.
- 6:51And tokens are really the.
- 6:54You know, the currency of AI. A token is essentially your action within that model.
- 7:02The token is the unit of compute that the inference model is generating to give you a response.
- 7:11So your response may be one token, depending if the inference requires some
- 7:16reasoning or some thinking, it may be multiple tokens.
- 7:20And each token has a cost. And as things develop and things get better,
- 7:25the price per compute is going to essentially lower, and then you'll have,
- 7:30yeah, greater flexibility in time and theory.
- 7:34Yeah, it also depends where you get your computer. Exactly. Because if you're
- 7:36running your models locally on hardware that you own, then you have infinite
- 7:41amount of tokens until, I mean, you run out of compute space. Yes.
- 7:46Okay, so back to what is egetic. So in ChatGPT, you type in a prompt and you get a response.
- 7:53Then you learn something and you type in a prompt and you get a response. Well, what if you're...
- 7:59Question was how do i use my cad program to to do this thing.
- 8:05And the response would be here are the 30 steps or here are the 100 steps or
- 8:09here are the 1000 steps yeah and you'd be like all right i'll copy and paste
- 8:13that into a document and then go do those,
- 8:151000 1000 things now i could do this cool thing i never knew how to do before
- 8:20and it's customized to specifically what i wanted to do and i didn't have to
- 8:23read the manual pretty awesome except you still have to do the 1000 steps exactly
- 8:27So you're not that far ahead.
- 8:29Well, where Agentic comes in is now you say, all right, do that for me.
- 8:34And now the computer, the Agentic AI is going into that application and doing it all for you. Yeah.
- 8:42And you just skipped a thousand, and you can go to sleep. Exactly.
- 8:45And that's why I think one of the big takeaways in the Firm the Future Research
- 8:48and the report that is just now out that essentially AI is not a question of replacing engineers.
- 8:53It's going to be allowing engineers to do things faster, more efficiently,
- 8:57and to free them up to do the things that they really should be doing by leveraging agents to,
- 9:04do all these sub-operations at a level.
- 9:10A multiplication effect that was impossible beforehand. So you could have a
- 9:13firm of 20 engineers that now becomes unlimited by the number of agents that
- 9:17you're actually allowed or can produce to do the work. That's right.
- 9:20Every engineer now has a team.
- 9:23Resources are unlimited as long as you have compute. So you have a team of 10 people.
- 9:29Before I start my day now, before I go in and check my email,
- 9:32I want to get my agents rolling. I want to get at least a couple agents rolling.
- 9:36So like they're doing work while I'm doing work. because once I get into email
- 9:41and meetings, I mean, I have to get them rolling.
- 9:43But I have some, I just had a team call. Some people on my team have automatic
- 9:47agents, so they're just working all the time.
- 9:50They don't have to get them started.
- 9:53So, and those agents could be, I mean, they could be doing a lot of work for
- 9:58you, but also helping you iterate because there's experimentation involved in
- 10:02the process. There's creativity.
- 10:04Well, I'll have this agent try 10 things this way. I'll have another agent try
- 10:08another 10 things the other way, and let's see which one wins.
- 10:12It allows firms that don't have this. It used to be that you'd have to be a
- 10:16very large firm with a lot of different capabilities to be able to do the amount
- 10:20of iteration, the amount of,
- 10:22experimentation that, you know, there's no way a small firm could do it.
- 10:25But leveraging agentic AI...
- 10:29Setting it up properly, it really does completely level the playing field that
- 10:34you can now engage in the same amount of testing iteration that any of the large ones get. That's right.
- 10:41And like, you know, in terms of replacing people, I mean, that's not high value
- 10:46work, in my opinion, pushing buttons in a software application.
- 10:49Like the high value is bringing your ingenuity to bear and applying your creativity and making decisions.
- 10:56But like have the agents go and do the boring stuff that is tear-pulling.
- 11:02Yeah, it's empowering the engineer to do the work that they were trained to
- 11:06do. Yeah, that they want to do.
- 11:07Now, looking ahead, I mean, from NVIDIA's perspective, I mean,
- 11:11both from just the development of AI, both in hardware and then the implementation,
- 11:16I mean, where do you, I mean, you're engaged with the firm of the future.
- 11:22It's fantastic to have a company like NVIDIA part of this process to help guide the future.
- 11:27What is, you know, NVIDIA's.
- 11:29Entrance into the EEC space? I mean, where do you see the growth opportunity there?
- 11:35Where is it from your company's perspective on what the future holds in our
- 11:40industry? Ah, it's a really open-ended question. I know.
- 11:44The sky's the limit. The sky's the limit. And where NVIDIA fits is we're enabling
- 11:48the ecosystem to do better, to do more with less, to accelerate.
- 11:53We work with all the software companies to help them at the low level with these
- 11:58computationally complex processes that can be accelerated on our technology stack.
- 12:05So that's AI is a great example.
- 12:08We have, yes, all these applications hit our hardware, but we have tons of software
- 12:14libraries for developers that they implement and integrate into their software
- 12:19and use to develop their AI,
- 12:22and that gets rolled out on our interviews.
- 12:25A portion of the firm of the future also looks at, of course,
- 12:27the future of the firm in 2035. And, of course, we look at the workforce issue
- 12:31and really what is the firm going to be after we get to the point where AI,
- 12:37energetic AI becomes kind of ubiquitous.
- 12:39And you talk about pretty much the AI, you know, the AI first mentality.
- 12:43It's not a firm that implements AI in the back end, but it's one that actually
- 12:47is built on the foundational model of having AI at its core.
- 12:51Right. You know, how do you see the industry evolving to be an AI first industry?
- 12:58I think the industry needs to first think about itself as a technology-first industry.
- 13:05Because I think it's been like a technology laggard. Yeah.
- 13:10Not taking, I would say, firms in general don't really take accountability for
- 13:18technology. They buy off the shelf. Yeah.
- 13:20Which is fine. You could buy off the shelf, and you should continue to buy off
- 13:24the shelf. There's tons of options out there.
- 13:26But I think you really have to know what those options are and have to have.
- 13:31Particularly now with AI, have to have some ability to use your data,
- 13:38to leverage your data, to inform what those applications are doing.
- 13:42So, for example, if you buy a software application that has AI functionality
- 13:47in it, it's generic. Just the same way Chad GPT is generic. It's trained on
- 13:51the entirety of the internet.
- 13:53So some AEC-specific AI feature that is from a software company is trained on
- 13:58some data set that wasn't yours.
- 14:01It doesn't reflect the way your business works.
- 14:03So the question becomes, how do you then teach that AI to do things Jeff's way?
- 14:10Because Jeff's been doing, Jeff Incorporated has been doing the same thing,
- 14:14building a hospital in this region according to this building code and the way
- 14:19this region or hospital board likes to work for 50 years.
- 14:23And you know what you're doing better than anybody. That is not going to be
- 14:27encapsulated in that generic AI. So you need to teach that AI with your data.
- 14:31In order to do that, you have to have people in your firm to understand all this technology.
- 14:34And that's the biggest thing I think is, you know, even organizationally from
- 14:38AECC, We recognize this and we realize that, you know, we have a lot of data.
- 14:45It's not structured appropriately to be able to leverage AI.
- 14:48You have to get to the point where you actually understand the entire universe
- 14:53of your data and then structure it in such a way where you could put it into
- 14:57a model to actually have it do what you want to do.
- 15:00So that's really interesting because I would have totally agreed with you a year ago.
- 15:05Is that all now outmoded thinking or is it? I think we're getting there.
- 15:08I mean, there's two schools of thought. One is exactly what you just said.
- 15:12You got to go in brute force, get some data scientists, get all of your data
- 15:17in one repository, normalize it all into the same format, homogenize it all.
- 15:24And then all of a sudden you have a single point of view. And if you think about
- 15:27tabular data, you can zero in on exactly what you need.
- 15:31But with agentic AI- You can have it do it for you. Yeah.
- 15:34Just put it in and say, okay, here's the data, sort it by kind,
- 15:38sort it by X, Y, Z, whatever kind of tag you want to put it in.
- 15:41You figure it out. Figure it out and then come out. Yeah, you figure it out. So there you go.
- 15:44So I'm behind the times already because I'm trying to figure out like, you know, everybody is.
- 15:48And that's the thing. This is happening so fast. But the point is,
- 15:52you know, all of a sudden you have to worry a lot less about interoperability
- 15:56if you're not having to brute force a situation.
- 15:58Absolutely. If you have agents that are working around the clock to just go,
- 16:01you know, whether it's import, export, normalize, find some standards and normalize
- 16:06around that, whatever it is, agents are going to figure it out and the agents can figure it out.
- 16:12And you just keep testing until you find like a non-lossy way to do it.
- 16:15So we could talk about this for hours.
- 16:18I think that, you know, the Firm the Future report that came out,
- 16:22you know, this week, I encourage everyone to take a look at it.
- 16:25It's up on the ACC Research Institute website.
- 16:28It's the continual research product coming out of this initiative.
- 16:32What's next for the work that the Institute's doing and NVIDIA's work with the
- 16:37Institute on kind of the next product in that firm of the future?
- 16:41Well, what I'm super excited about is, first of all, all this stuff can be done
- 16:46today. And so we talked about ChatGPT.
- 16:49That's OpenAI. Yeah. And OpenAI has an agentic toolkit called Codex.
- 16:55Yep. And you can go and download it and start using it. Got it on my Mac. Right on. And so now...
- 17:01Back to that automation system instead of giving me the 1000 lines of code just
- 17:06tell your codex to go and do it the way it does it is by into with a traditional
- 17:12isv application is by connecting to that application
- 17:15through the applications api yeah all applications have apis which is like if you're a programmer,
- 17:21or a plugin developer you can write some code that accesses the application
- 17:25through the api yeah well now your codex
- 17:27can access the application so once you have that It's connected,
- 17:31and then you can just create the agent to do exactly what you want through that
- 17:34API connection to that app.
- 17:36And essentially, that's a blank canvas for whatever you need to accomplish. That's right.
- 17:41And now, to make it even easier, so the other example of an agentic tool is
- 17:46Claude Cowork. And Cowork just...
- 17:50Last week, announced a bunch of connectors for creative workflows.
- 17:54SketchUp, Autodesk Fusion, Blender. So it's all coming together.
- 17:59It's all starting to come together. So that's the MCP and the skills that it
- 18:03needs, all that to go and do the work for me.
- 18:07Now, I mentioned three applications. Why couldn't you just tell,
- 18:11you know, co-work, hey, go do something with all those three applications.
- 18:15So use all the applications you want. They're all installed on my computer.
- 18:18So on my team, we're testing now, here is the RFP.
- 18:23Generate an Excel version of the RFP with all the key data. Go and use Rhino with Grasshopper.
- 18:31Create a conceptual drawing of the building with some options.
- 18:35Maybe even I'll give it a sketch as some inspiration.
- 18:39Generate a conceptual model. Render that for me.
- 18:42Bring it into a BIM application.
- 18:45Create a structurally sound model based off of all that and create the construction docs
- 18:51one prompt and i was just on a call with my team and we were testing this out
- 18:55and watching it and we're like wait a second put 49 columns how did he get that
- 19:01information is that in the rfp like how does it know to do 49 columns and,
- 19:05and we went and checked the rfp that we didn't there's nothing like that in
- 19:07the rfp so somehow somehow just figure that out and that's funny and it shows
- 19:12how quickly things were developing because I've had conversations with firms
- 19:15that were on a leading edge of AI and they were even talking about the fact that.
- 19:20Yeah, we haven't got to drawing yet. We're getting close to that.
- 19:23But now we're at that point where we can put the prompt in and actually render.
- 19:28And that was not too long ago. I mean, if you don't get on the train now,
- 19:33you're going to be left at the station. Yeah. Here's your prompt.
- 19:36Read my RFP, generate conceptual model, BIM model, conceptual rendering,
- 19:44and full set of construction docs. Yeah. These specific sheets that you're looking for.
- 19:49Well, let's wrap it here because I know you've got another education session to get to.
- 19:53You were at one earlier. It was standing room only. You were on stage earlier
- 19:57talking about this with the Firm of the Future group. I encourage everyone out
- 20:01there, again, look at the research, get involved with it.
- 20:05I'd love to have you on again as this keeps on developing.
- 20:09But for right now, we'll leave it there, Sean Young, NVIDIA,
- 20:13part of the Firm of the Future initiative over here at the ACS Research Institute.
- 20:16Thanks for your time. Thanks, Jeff. Really appreciate it. Thank you. Thank you so much.
- 20:19And again, that has been Engineering Influence, a podcast from the American
- 20:22Council of Engineering Companies. And check out.