Latest / Engineering Influence from ACEC / Inside the Numbers: AI Risk in Engineering with Jim Compton and Joe Bates
Transcript
- 0:05Welcome to the Engineering Influence Podcast from the American Council of Engineering
- 0:08Companies. And today we're digging into a new study from the ACEC Research Institute
- 0:13called Managing AI Risk in Engineering Firms.
- 0:17Research built on interviews with firm leaders, public infrastructure owners,
- 0:20technology vendors, insurers, attorneys, and licensing authorities.
- 0:25All trying to answer the same question, which is what actually happens when
- 0:28AI enters a profession where the standard of care isn't optional.
- 0:32To help us unpack that, I'm joined by two guests. If you're any fan or follower
- 0:37of the program, the man needs no introduction.
- 0:40Joe Bates, our senior researcher at the ACEC Research Institute,
- 0:44a longtime returning guest, and has been instrumental in the Institute's work
- 0:48on the AI series of research they've been doing, the Firm of the Future.
- 0:52Larger initiative with the Institute and all of its economic research.
- 0:57Joe has been guiding the industry research for the Institute for a number of
- 1:03years now, And it was great to have him back on the program.
- 1:07And joining us with Joe is Jim Compton.
- 1:10He's the Chief Technology Officer at ESP Associates. That's a multidisciplinary
- 1:15engineering, surveying, and planning firm.
- 1:17Jim's a licensed professional engineer with more than 35 years of experience
- 1:21across engineering, IT, and business development.
- 1:24And he leads ESP's technology strategy, IT infrastructure, and cybersecurity
- 1:29initiatives, which makes him a great person to talk about what AI governance
- 1:34actually looks like inside of a working engineering firm.
- 1:37So Jim and Joe, welcome both of you to the show. Thanks, Jim.
- 1:42Well, let's go back to the start here.
- 1:46Joe, if you can kind of give us a kernel of where this research came from, the specific focus on,
- 1:53really taking it away from just the technology and our firm's adopting technology
- 1:58to the actual implementation and governance within firms.
- 2:02And what risks did you expect engineering firms to be most concerned about?
- 2:09And was that validated through a lot of the interviews that you did?
- 2:14Yeah. So we started out really trying to identify what the risks of utilizing
- 2:23AI are across the firm from a variety of different perspectives.
- 2:28And we expected at the beginning that we might come out with a list to sort
- 2:33of document various risks that are out there.
- 2:37As we progress through the research, we realize if we identify risks today while
- 2:43we are conducting the research, those could be different six months down the road, right?
- 2:49So we took a different approach and we created a framework for how firms can
- 2:57identify the risks. And we do identify some risks within the research,
- 3:02but more importantly, we're going,
- 3:04How can firms take an organizational approach to managing the risks associated with AI?
- 3:11Because we also learned very, very quickly in the research that AI is not just a technology issue.
- 3:18You could have a lot of different AI programs, software agents,
- 3:22and if you don't have a governance structure, if you don't have a strategy,
- 3:27then you're not going to be able to effectively use AI properly.
- 3:31And you're not going to be able to effectively address the risks that come and
- 3:35go as we progress through the mature life cycle. Yeah.
- 3:40And I want Jim to come in on this because both of your experience in technology,
- 3:46and as a firm leader, but starting with governance.
- 3:49You know, if a firm is using AI today, and I think we can all assume that if
- 3:54you have a firm that has a connection to the Internet, someone's using AI in some form or fashion.
- 4:00But let's say a firm is using AI but does not yet have a mature AI governance structure.
- 4:06What specifically should it put in place first to reduce the initial risk and
- 4:11kind of tourniquet first, you know, triage before it starts building out that
- 4:16larger, mature structure? What would some of those controls,
- 4:20what would they look like?
- 4:30The first thing that we do i don't think people should do is name a person to
- 4:34be in charge i mean you need someone to own this and it can't be a group it
- 4:39can't be a committee it needs to be somebody,
- 4:42a specific person that's actually knowledgeable about ai they're excited about.
- 4:50They genuinely care about the growth of the company, but they also care about limiting risk.
- 4:54And so you can't just pick whoever's available. So once you've got someone in charge...
- 4:59You need to create a small cross-functional group of some engineers,
- 5:03some IT, some legal people, and put this AI owner in charge of that group.
- 5:09And then they're going to be able to write a policy that addresses all the different
- 5:15issues in the company. But the policy doesn't have to be something huge to begin
- 5:18with. I mean, it could be short. I mean, you just need an acceptable use policy.
- 5:24You can start out with one or two pages. It doesn't need to be 50 pages.
- 5:27No one's going to read it.
- 5:28So you need to cover, you know, what tools are approved for use,
- 5:33what data can go in and what can't, who do you call or who do you reach out
- 5:38to if you want permission to get a license? You know, who approves a project
- 5:42if I've got a project I want to do?
- 5:43Those kind of things need to be in there so you can get some control around
- 5:48what's happening with AI. But before you can approve a project,
- 5:52I mean, you sort of need to know how much time it's going to take to create,
- 5:55the project and how much time it's going to save you.
- 5:58And initially, you're going to be guessing at this stuff. You're not going to
- 6:00know, but your guesses will get better over time.
- 6:04And, you know, is it a one-time task somebody wants to automate or is it something
- 6:08you don't in every project?
- 6:10So the projects that are available early on, the decisions should be pretty
- 6:15early, should be pretty easy because there's so many things that you can do with AI.
- 6:21But then you got to get in front of your project managers and tell them all
- 6:24about your policy and try to get your hands around the thing.
- 6:31Because at some point, someone's going to come to you and ask you,
- 6:34you know, you approved me to do this project. What do I charge my time to?
- 6:37Where do I put that? You know, how do we track this stuff that we're working on?
- 6:42You know, is there a project number I can charge it to? And I would recommend
- 6:45if someone is going to do larger projects that they do have a number to charge
- 6:48time to for every one of them.
- 6:49Because otherwise, you're never going to be able to track it.
- 6:54So just start with some basic stuff. And looking at and taking kind of a wider
- 7:00view, Joe, I mean, the study kind of looked at eight.
- 7:05Risk domains with AI, and it's governance, professional liability,
- 7:10workforce, data, the business model itself, risk to leadership.
- 7:16Of the eight risk domains in the framework, which do you think engineering firms,
- 7:22and their leaders need to most understand now, like right now,
- 7:25you know, as as they address this issue?
- 7:29It's a good question, Jeff. And that the answer is a little more complicated
- 7:33because all of these eight domains are connected. They're interrelated.
- 7:38You can't focus on one area and ignore another area.
- 7:43Yeah. So some of the overarching themes, though, that really come out is that
- 7:48you need to maintain the engineering judgment.
- 7:52That's probably the most important thing, right, is with AI,
- 7:58you can't let AI make the decisions. You can't not check AI.
- 8:03So you have to be able to maintain the judgment of the engineer and keep that,
- 8:10sort of people first is a lot of the terms that I heard when I was conducting these interviews.
- 8:15So it's really important to focus on that. But again, they're all interrelated.
- 8:19You have to also focus on workforce. How are you going to train people how to use AI?
- 8:24And as Jim was saying, if you have an acceptable use policy,
- 8:29you got to make sure everybody knows what that policy is.
- 8:31Then as people are learning about AI, you have to make sure there's a feedback
- 8:36loop, right? So that people that are experimenting, that are learning how to
- 8:41create tasks, they have to teach the AI to me.
- 8:46They have to teach the AI, but then they also have to make sure that people
- 8:49in the firm understand, oh, we made these improvements. We can now do this.
- 8:54We can save time on that. I did this and it worked really well.
- 8:57So there's that workforce issue and feedback loop.
- 9:00And as we already talked about governance, I mean, you have to have governance in place.
- 9:06That's that's really the first thing to do. But again, I think everything's
- 9:10interrelated. So you have to address it all together.
- 9:14Yeah. And I think that's important because it's also running a knife's edge
- 9:19to an extent here because you need to have a governance structure in place.
- 9:22You need to have policies that are clearly defined and also enforced.
- 9:27You can't just write it and put it on the shelf. You need to make sure that
- 9:31it's understood by the employee base and that it's actually being followed.
- 9:35But at the same time, you don't want to shut down innovation and you don't want
- 9:40to, you know, put a blockade up against good ideas.
- 9:44So, you know, Jim and first and then Joe, I mean.
- 9:49The risk of unmanaged AI use, you know, the shadow AI, built in tools,
- 9:54custom proprietary agents, workflows that kind of start as experiments,
- 10:00but then get into the project delivery cycle.
- 10:02Those all have to be guarded against, and governance does play a role in that.
- 10:07But now, you know, the innovation side, how do you keep that innovation cycle
- 10:12going when you start putting controls? What do leaders need to do so that they
- 10:17can still stay on the leading edge, but at the same time, they're protecting
- 10:20their business and their standard of care?
- 10:23Yeah. I mean, I want to back up a little bit there, too. I mean,
- 10:26you were talking about being able to enforce your procedures.
- 10:29I mean, you need to edit your QAQC procedures to talk about what has to have
- 10:34a sign-off and what has to have a review. It doesn't matter if it was AI created or not.
- 10:38And you have to draw a hard line there on those activities. And you have to
- 10:41hold people accountable.
- 10:43I mean, there's got to be consequences if people are not following the review,
- 10:47procedures because your company's reputation is on the line.
- 10:51I mean, you've got to do that.
- 10:53So, I mean, the liability isn't in AI making a mistake.
- 10:56It's in somebody not reviewing it and catching a mistake. But the shadow AI
- 11:01is a huge problem. I mean, we didn't think it was that big of a problem.
- 11:05But, I mean, it's basically people using public tools.
- 11:10But mainly because you haven't given them an approved alternative.
- 11:13You haven't told them, here's the company version. And so they've got a chat
- 11:17or a cloud account, and they're just out there doing things.
- 11:20You don't know what they're doing. You don't know what they're uploading.
- 11:23And so when we started working with AI a couple of years ago,
- 11:26we were wondering how much AI traffic was on our network. And so we started
- 11:29looking, running some tools to try to find out.
- 11:34And we had eight, this is several years ago, we had 80,000 AI domain hits to
- 11:40720 different AI tools in a four-week period.
- 11:45And, you know, we didn't have any approved tools. People shouldn't have been
- 11:49doing anything, right, according to the management, because nothing was approved.
- 11:53So a few concrete things you can do is buy the enterprise version of the AI
- 11:59tools, not the consumer versions. They have access controls in them.
- 12:04They let you manage who can buy tokens and who can't, how many tokens they can
- 12:08buy, if you know how many accounts you have.
- 12:11And if you don't, there's a lot of horror stories about people spending tokens,
- 12:16people spending $10,000, $20,000, $30,000 a month in tokens.
- 12:19But even the basic subscriptions are $20 a month.
- 12:23If people are out there putting them on their credit cards and 2,000 people
- 12:26do that, that's $40,000 a month.
- 12:28You have no idea where it's going or how to account for it. So we set up an
- 12:33AI governance team, and the AI owner is in charge of that team.
- 12:40And that team, we have an AI collaboration site to encourage people to use AI
- 12:45and do things with AI. They run that site. They talk with managers. They talk with groups.
- 12:49So they're out there interacting with people, encouraging them to use it.
- 12:54If somebody builds something complicated, they're trying to make sure it gets documented.
- 12:58How is it built? How it works? In case that person leaves, somebody's got to maintain it.
- 13:04So it's a process, but...
- 13:08You've got to get people around to experiment. We do that with policy and restrictions on files.
- 13:16We use the data exfiltration software. I can talk about more later to,
- 13:20you know, to keep confidential stuff from getting uploaded. But you've got to
- 13:24encourage people, but you've got to put guidelines in, too.
- 13:28Yeah. I mean, let's go into that because, you know, one of the things,
- 13:32we've had a lot of programs on AI, right?
- 13:35And from the early days when, you know, the LLMs were just hitting public and
- 13:40people were all talking about ChatGPT back in the days when,
- 13:44you know, the image generation would give you five, you know,
- 13:49more than five fingers a hand and you can tell what was fake and what was not.
- 13:53You know, we've come a long way in a short period of time. At that beginning
- 13:57stage, a lot of the conversation went around that question of organizational,
- 14:03knowledge, understanding data,
- 14:06understanding where really that transition from,
- 14:11you know, an engineering firm just being an engineering firm to being a data
- 14:15firm that does engineering.
- 14:18It's a question of how is that data stored? How is it governed?
- 14:23And then what controls are on those? So, let's go into that a little bit more
- 14:27about the importance of understanding good data hygiene, good data control and
- 14:32implementation within a firm.
- 14:34Explain that from your experience.
- 14:38As far as trying to get your data usable, I guess, in AI is what you're looking
- 14:43for, your knowledge management aspect of AI. I mean,
- 14:49that's a tough issue, right? I mean, everybody's got plans, you've got specs,
- 14:53you've got checklists, you've got lessons learned. A lot of that stuff's in
- 14:57paper, and you've got to get it so that it's usable by AI.
- 15:02And that's a boring task that nobody wants to jump into. But if you don't have
- 15:08that data usable, then...
- 15:13You're not going to get the full benefit out of AI if you don't have the data for it to learn from.
- 15:20I mean, you can do a lot of projects. You can do a lot of things.
- 15:22You can accelerate a lot of things, but you're not going to get the full benefit out of it.
- 15:27I mean, there's the only standard that I'm aware of out there.
- 15:31There's ISO or GIS 30401. I think it is. That's the knowledge management system
- 15:38standard, but it's really generic.
- 15:40I mean, it has a lot of good stuff in it, but it's very generic.
- 15:43There are pieces of software out there like Guru and Document 360 and stuff
- 15:47like that will help you put your knowledge into format.
- 15:53We don't use one. We haven't been using one. You know, we're in the early stages
- 15:58of trying to get that stuff done.
- 16:02I don't know that there's any one best way to do it.
- 16:06You know, honestly, we struggle with trying to get people to cooperate to get it in there.
- 16:10No one has the non-billable time to be loading old projects into the system.
- 16:15So that is a challenge. But the firm that gets their stuff organized and into
- 16:21AI so it can learn from it, I mean, they're the ones that are going to get the
- 16:24value. Everybody else is going to be left behind.
- 16:27Yeah, I agree with that 100%. I think the differentiator is going to be the
- 16:32firm that invest in the teams to definitely manage and control the data or the
- 16:38ones who are going to have a competitive advantage with their temporaries.
- 16:43I'd also say that, you know, one of the interesting things that, you know,
- 16:46getting into a conversation with a couple of firms in previous interviews is,
- 16:53kind of shocking compared to some other industries of how few firms have positions
- 16:59like chief information officer or chief data security officer.
- 17:03And the need for the industry to kind of.
- 17:08You know, take pace with other industries and have those positions in the C-suite
- 17:14that are in those executive leadership positions and have the impact that are
- 17:19empowered to take control of that data.
- 17:23You know, what's your what's your view on that, Jim?
- 17:27I agree. I mean, you've got to have someone. I mean, the data governance is
- 17:31a corporate decision. You know, it doesn't, that decision doesn't belong to
- 17:34IT or anybody. I mean, it's a decision the corporation has to make.
- 17:39And if the data is going to be the most valuable thing that you have in the
- 17:43future when working with AI, then someone needs to own it.
- 17:47Someone needs to be responsible for it. And someone needs to be challenged to
- 17:50get it in a format that can be used.
- 17:53But they also need a budget to do that, right? I mean, you can't just give,
- 17:57you know, you got a 2,000, 3,000
- 17:59person firm and you get one person to do that. That's a lot of effort.
- 18:03And, you know, as you're growing the AI tools, I mean, you have to treat it like anything else.
- 18:09You've got to pilot a test on it, see how it works, review it,
- 18:14and then scale it up. I mean, you don't want to go in and try to digitize all of your data all at once.
- 18:19And, you know, I was talking to someone recently, and they were talking about,
- 18:22well, you know, we think we can double the speed of this process that we're
- 18:26doing. That would be fantastic. If you do that on every project and you can
- 18:29double the speed of it, that's an amazing improvement.
- 18:32How do you build that project? They're like, oh, we build this all hourly.
- 18:35I'm like, well, I don't know you've done yourself any favor if you're doing
- 18:39it half the time and you're building it hourly.
- 18:42You know, so the first projects maybe you need to start looking at are things
- 18:45you build lumps on, not half in your income from doubling the speed of doing an hourly job.
- 18:52So it's, yeah, somebody needs to own it and make those decisions.
- 18:57And that person needs to be the AI owner in conjunction with someone who's in charge of the data.
- 19:05And I, you know, we could have an extended conversation just on this topic.
- 19:09It's another reason, read the study and go into it and do, you know,
- 19:13follow up interviews. I want to hear from the audience too. Like,
- 19:16what, what do you want to dig into? Because there's a lot here.
- 19:19I want to take a, a, a kind of zoom out a bit, Joe, and, and talk about something,
- 19:24which is, you know, I always kind of laugh at,
- 19:28to an extent, I mean, you know, you always see the futurists and the ai guys
- 19:32out and the west coast are talking about how work is going to become,
- 19:37irrelevant in a couple of years and we're all going to be replaced but the fact
- 19:41of the matter is i'm married to a to a lawyer who chafes at the idea as well
- 19:46which is that ai does not replace,
- 19:49the the the,
- 19:52the need for a professional like an engineer to be working on a project it does not,
- 20:00essentially lower the responsibility, the professional responsibility of the work.
- 20:07If anything, it just kind of puts a greater focus on that.
- 20:12Let's go into that and dispel that myth that, you know, AI is going to replace
- 20:16the judgment of a licensed professional trained engineer in the near future,
- 20:21because it just isn't going to happen.
- 20:23Yeah, we don't see that happening at all, Jeff, and this research points to.
- 20:30The opposite, in fact, that engineering judgment becomes more important,
- 20:35that it becomes more important for you to,
- 20:39ensure that the outputs that AI is giving you are accurate and then to interpret it, right?
- 20:46You have to interpret the information that the AI is giving you and you have
- 20:52to become that consultant to your client.
- 20:54And so I think it actually puts the engineer more in the center in terms of
- 21:02providing decisions, using their judgment.
- 21:05And then this has implications down the road because there are tasks that will be automated.
- 21:11And there are things that engineers have done for a long time,
- 21:14sort of in a more manual way and in a more hands-on way.
- 21:19And that's how they got the judgment and experience that they have,
- 21:23right? you have your senior engineers.
- 21:25So the real challenge then is how do you,
- 21:29change the education, the professional development process of younger engineers
- 21:36who may no longer have to do certain tasks that engineers used to do that helped
- 21:42to give them that experience, right?
- 21:44So again, it becomes more important, not less.
- 21:47And we have to focus on how we're going to develop our workforce with this new
- 21:51tool and that's going to change the entire firm,
- 21:54yeah this this this is the evolution that you know from from the the physical
- 21:59drafting equipment and the drafting cables to CAD from CAD to.
- 22:05Full digital design from digital design to AI assisted design,
- 22:09it is the same story told you know it's the next chapter for AI it's going to
- 22:15have probably a larger impact But it does come down to that,
- 22:20trusted advisor relationship that professional engineers have with their clients.
- 22:25And the study I thought was interesting that the clients really don't really
- 22:28care about the kind of AI that's being used, right?
- 22:30They're kind of agnostic on AI, but they have zero qualms about saying that
- 22:37we have to make sure that the people who are doing the work maintain that level
- 22:41of professional responsibility and judgment,
- 22:44regardless of what technology is.
- 22:46Yeah, there's a standard of care that you maintain, and that will not change.
- 22:51As Jim said earlier, AI by itself doesn't create the liability.
- 22:56The liability is if you have not checked AI and have not made sure that the
- 23:03outputs are correct or you've used your judgment to say, OK,
- 23:06I like this, but I don't like that.
- 23:08Yeah, I want to wrap it up there. I think that, you know, I do want to do a
- 23:12call out for the audience listening.
- 23:15I mean, tell us in the comments what you want to dig into here,
- 23:19because there are a lot of areas to explore. There are eight areas of risk that the study addresses.
- 23:26Tell us what you want to dig into, and we'll do a follow-up on it.
- 23:29I do want to give both you, Joe, and Jim the opportunity to wrap things up.
- 23:36What do you think is the big takeaway? If there had to be one takeaway from
- 23:38the study, what do you think that would be?
- 23:41You know, from my perspective, it really is that the long-term success of an
- 23:46organization is going to be less dependent on which AI tool they use.
- 23:53And it's going to be more dependent on how they manage the use of AI.
- 23:57That's the real critical differentiator. And, you know, Jim talked about several points.
- 24:02If you don't have your data organized, connected,
- 24:06accessible, seamlessly flowing through the project phase, If you don't capture
- 24:12the organizational knowledge that is not the data itself, but it is that experience,
- 24:18that judgment, as people retire,
- 24:21you're going to be behind. Yeah.
- 24:25And just quickly on the people part of it, I mean, if you are doubling the speed
- 24:31of some tasks your young engineer is doing,
- 24:33and you had one engineer doing the review of a group of five people,
- 24:37and now they're doing twice as much work, you're going to need two reviewers.
- 24:40I mean, it's not going to eliminate jobs. And, you know, maybe that reviewer
- 24:46doesn't have time to do business development anymore.
- 24:48Maybe he's focused on, and you have to hire a business development guy.
- 24:51You know, the seller-doer model may change. I mean, so there's a lot of real
- 24:56implications of how it does speed things up, but eliminating jobs isn't one of them.
- 25:00I mean, you've got to get your data in order.
- 25:07You've got to pilot some projects and then alter your mentoring and your training
- 25:11based on what you learn from that.
- 25:13If you run some engineers through a couple of projects and you're like,
- 25:16they didn't learn half of what I thought they should learn, that you need to
- 25:21change your training in order to deal that.
- 25:24Basically, I mean, AI is an amazing tool. It's getting better every day.
- 25:28The goal is to find some way to use it to improve the quality of service for
- 25:33our clients. I mean, that's the whole point.
- 25:35If you go back and look at a project and analyze it from a business perspective,
- 25:41I mean, how did it actually change how fast you're getting the work done?
- 25:45How did it actually change your work and what you can charge for it?
- 25:50And those kind of meaningful things that hit the bottom line,
- 25:53if it doesn't do that, then I'm not sure why we're doing it.
- 25:56Yeah, very good points to end the conversation with, and I 100% agree.
- 26:00It's not going to kill jobs. It's going to create new ones, especially for maybe
- 26:06the non, a lot more data people, a lot more people who are computer science
- 26:11and are people can work with and organize data.
- 26:15The people who can also help back up the engineers in utilizing the technologies.
- 26:22But it's definitely going to allow for maybe some greater innovation,
- 26:26but it's not going to reduce the world wars. And I think that,
- 26:29you know, that's very important.
- 26:31So anybody out there who's looking to get into the engineering field,
- 26:35probably going to be more opportunities because there are going to be different
- 26:37kind of jobs out there to get.
- 26:40I will encourage everyone to go on the ACEC Research Institute's website.
- 26:44We have the link to the study in the show notes for managing AI risk in engineering firms.
- 26:50It's a great piece of research, very usable from the C-suite,
- 26:54and I encourage everybody to download it and let us know what you want to hear.
- 26:57This is a topic that we're going to be coming back to a lot.
- 27:00Joe's going to be with us in fall conference. We're recording this in the last
- 27:05week of August and October of 2026. We're going to be in Phoenix.
- 27:10Joe, you'll be there talking about some economic stuff, and I'm sure we'll be
- 27:12touching him on this issue.
- 27:15Jim, I really appreciate you joining us as well and your expertise.
- 27:19See you next time. Fantastic. And we'll look forward to extending this conversation.
- 27:24So, again, this has been Engineering Influence, a podcast from the American
- 27:27Council of Engineering Companies. And we'll see you next time.