Fuqua Insights Podcast: Is AI Replacing—or Reshaping—Jobs?

Kathleen Barrow

Welcome to Duke Fuqua Insights, a podcast where we explore faculty research and the actionable takeaways for business leaders at every level.

Artificial intelligence is reshaping how companies operate, but is it actually making them more productive? New research reveals how AI is affecting productivity and workforce composition, and how this technology wave may be unfolding differently from those that came before it. Drawing on responses from nearly 750 financial executives across industries, the study provides one of the most comprehensive snapshots yet of how firms are adapting to AI.

I'm Kathleen Barrow, I'm an MBA student at Fuqua and a research assistant on this project. And for this special episode of the Duke Fuqua Insights podcast, I'm joined here by Professor John Graham, who directs the CFO Survey conducted with the Federal Reserve Banks of Richmond and Atlanta. Professor Graham, thank you for being here.

John Graham 

It's great to be here.

Kathleen Barrow 00:00:58 

So before we jump into specific findings from the study, I'd love to start with what makes this research unique. This study examines AI through the lens of financial executives within firms. Why is that an important perspective as we think about productivity and the workforce?

John Graham 00:01:12

Well, there's probably two components to answer this question. First, what are we studying? We're studying how AI is used at the typical firm and what the effects of that are on labor and productivity. And second, who are we asking these questions of? That's where the financial executives come in.

So we’ve run this CFO Survey every quarter at Duke for almost 30 years now, last five years jointly with the Federal Reserve Bank, and in the fourth quarter of 2025, just think, early December 2025, we added about a dozen questions about the use of AI. So why are CFOs the right people to ask about the use of AI at the typical company? Well, because they know all these numbers, they know what the spending is on AI, in particular what type of spending, how they expect that to affect labor and productivity. When I use the word productivity, I mean output per worker. Let me say a couple things that we're not doing here. At least one thing: we're not really focused in this project on the big data centers and server expenditures that are being done by kind of a small number of very large companies.

That's super important to the economy, and that needs to be accounted for also. But our study is more at your typical firm. How are you using it and what are the implications?

Kathleen Barrow 00:02:32 

Yeah, it's really interesting to think of these financial executives as having a bird's eye view of their individual firms. So, really important perspective. So, given that this is who we were talking to, what findings stood out to you the most?

John Graham 00:02:43

Well, another qualification before I jump into that, I guess. So, we intentionally focused on a short-term horizon 2025 for those questions we asked. How has AI already affected your company and the main focus really was on 2026, just one year ahead. We think this makes sense because it's always difficult for companies to forecast very far into the future.

And particularly with AI, this is just there's an explosion with AI going on right now. It's really hard, I think, for anyone to see very far into the future. We did ask a little bit about 2028, but let's just think primarily of 2026. So, with that in mind, one of the key takeaways is, in 2026, we're not expecting an AI apocalypse to the labor market.

We're not expecting massive job loss because of the use of AI. Now we ask in our survey at the typical company, but then we can aggregate that up across all the companies in our survey, to get a sense of aggregate effects. So this is pretty important. Now, the headlines might suggest there are companies out there laying off because of AI.

Some of that's probably legitimate at well-known companies, so it gets into a headline. We're, you know, we're asking not.. we're surveying not just the headline companies, but a wide range of companies from small, medium, large, public, private, etc.. So, first effect is we don't expect a large aggregate job loss — a little bit yes, but not widespread and not large, not particularly large.

Now we can talk a little bit more later about what types of jobs will be most affected. In a nutshell there, I think we're seeing that some routine clerical work might be the first jobs to go because of AI. That could start happening in 2026. 

But offsetting that, some companies say they might hire on the technical side, so we might actually have more employment to offset, at least to some extent. Again, we can unpack that later. The second set of main results with this productivity, this output per worker. And here we do see CFOs saying they expect an increase in productivity. Both already experienced a little bit in 2025, but even at a higher rate of productivity growth in 2026.

One thing that's nice about a survey is we can ask specifically, do you, you know, attribute it to your use of AI? What do you think the effects will be? Now, what's a bit interesting is we find what we call a productivity paradox, and we'll unpack that a little bit more later too, but by that we mean when we also ask CFOs, how much do you expect your revenue to go up, revenue per worker to go up, even in 2026, it's a smaller number than the productivity increase. So that's a little bit of a head scratcher. Why isn't all this showing up in revenues yet? We'll get back to that later.

A third element that again we can unpack a little bit later is unlike past technology waves for your typical company, this is not all about spending on hardware. You can spend very little on hardware and actually still benefit from the AI boom. And again, we'll unpack that a little bit later.

Oh well, one last thing. What I personally had expected, but we don't see much evidence of, so I was surprised by this is CFO say this is not about, at least in the short run, cutting cost dramatically or laying off a bunch of workers in the short run.

In fact, if you think about it, you've got to spend money to get the resources to use AI. So this isn't really a cost saving move, at least not yet. It's more about what you produce, producing it better, if you will, and maybe setting yourself up for the long run.

Kathleen Barrow 00:06:28 

Well, I think there is a lot of public anxiety about AI driven job loss right now. So I think it is really important to have this conversation and think about what are the nuances here. So specifically, your data show almost no measurable change in total employment. And large firms and small firms look quite different, actually. So what explains this divergence.

John Graham 00:06:46

So interestingly what we found was that large firms are expecting to reduce employment, somewhat, and in particular for routine clerical type work in 2026. And again, a little surprise to me, small firms are actually expecting, hoping to increase employment a little bit. Again because of AI. For your typical small firm, they might need to hire a technical worker in order to fully implement and, you know, integrate AI into their company.

What about large companies? How is it that they're potentially laying off routine clerical when small companies are not indicating they'll lay off routine clerical? Here I think it's a little bit of a numbers game. Let's say you're at a company with a thousand employees and 20 of them are doing data entry, you know, depending on what your industry is, but just as an example, well, at a large company, you might be able to lay off five of those data entry kind of routine clerical workers and replace their output with AI enhanced remaining employees, so the 15 remaining data entry can now do the work of 20 thanks to AI. Whereas at a small company, maybe you're at, say, a 50-person company, there might only be two data entry people and laying off one of them… that's half your data entry staff. And not only that, it's quite possible these people have other tasks. The data entry person might also do something with payroll every month, and you can't really lay off the data entry task because you need… maybe AI can't replace the payroll task. So at large companies you just have more employees with maybe narrower job focus, and it might be easier to kind of get rid of some of them and still cover all your bases, if you will, and that might be harder at small companies.

And finally, small companies also are more in a growth mode usually, you know, on average. And so the AI, the investing and retaining of employees might be helping them to grow, whereas large companies are still growing? Yes, but maybe at a slower rate, so that AI might be able to keep up that rate of growth while they're potentially reducing employment a little bit.

But let me turn the tables on you for a minute, Kathleen. As you mentioned in the beginning, you were a research assistant on this project. So I have two questions for you. One, can you let us know who the other research assistants were on the project? And let me say they—all four of them—did an outstanding job. It was maybe a bit tedious at times, I'm not sure, but it's a little bit fun maybe to looking through the data and seeing these survey responses and trying to make something out of it. So, one, who were your compatriots there? And second, what's your takeaway? You looked at the data. You know, what's your thought about this job loss part of the equation.

Kathleen Barrow 00:09:33

Yeah, absolutely. Yeah, we had a great small but mighty team that was myself, Ayush Vats, Matt Gilliam and Akshara Bassi. And we worked together to kind of be thoughtful about defining categories of tasks and job types and then work together to analyze all the responses from the CFOs accordingly.

John Graham 00:09:49

Okay, good. And did you have any thoughts about did it seem like there was massive job loss to you? Did it seem like most companies are laying off employees?

Kathleen Barrow 00:09:58

I think something that really stood out to me, even from the beginning of the analysis, was how many CFOs were actually saying, no, we don't expect AI to lead to job loss at their firm. So the question was open ended. It was unprompted. So I think it's pretty meaningful that AI replacing people is not actually the overwhelming view of people who are at the top despite, like we said, what we might see in the media.

John Graham 00:10:21

Excellent. Thank you. All right. I'll hand the mic back to you.

Kathleen Barrow 00:10:25

Okay. So even if total employment is flat, we'll just keep digging in here, the composition of the workforce is shifting. So, what kinds of roles are shrinking and which ones are growing.

John Graham 00:10:34

Right. So I touched on this a little bit. But let me just expand. So, we've divided job types into four tasks using Bureau of Labor Statistics categories. One was routine clerical, one was creative, one was technical, and then the other one was other. The fourth category we didn't get a lot of changes expected in creative or other, so I'll set those aside and focus on the routine clerical.

And again, what we found was that in 2026 and even into 2028, by the way, we do see that the proportion of employees in any given company doing routine clerical work will decrease, by 2028, by maybe as much as two percentage points. So maybe you had 24% routine clerical, now you'd have 22% of employees being routine clerical, by 2028.

And well, that could mean you're losing total employees. But in fact, we did see this partial offset in the technical workers. And so it was the large firms laying off or planning to lay off routine clerical and the smaller companies hoping planning to hire on the technical side. There is still a net job loss, but again, it's kind of small.

The trends that we see in 2026 do continue into 2028. So everything I've said could apply to either 2026 or 2028. It's just the magnitudes are a little bit bigger in 2028.

Kathleen Barrow 00:12:02

So one of the most interesting findings is a modern version of the productivity paradox, which you alluded to earlier in our discussion. Companies report sizable productivity gains, yet revenue-based measures lagged behind. So why is that?

John Graham 00:12:14 

Well, I'll give you the history of that phrase, the productivity paradox. So Nobel Prize winner Robert Solow made this statement back in the early 90s, when we were in the computer revolution. At the time, computers were ubiquitous. They're showing up everywhere. And, you know, economists and others were all saying, well, this is going to greatly enhance productivity. Well, not only that, computers are going to replace people—that was also said a lot at the time.

But Solow is kind of a productivity economist, and he looks at the data and he's like, I noticed that there's computers everywhere, but it's not showing up in the productivity. So his exact quote was the computer age is everywhere, but not in the productivity statistics.

So, meaning even though you see this trend happening right before your eyes, is it actually increasing productivity? Okay, so back to our survey now about AI. And when we ask CFOs a direct question, how much do you expect productivity output per worker to increase, they said 3% as much as 3% in the year 2026 due to the use of AI.

That's a pretty big number. Productivity doesn't usually grow that fast. But then when we looked at the revenue growth attributable to AI, it was a number smaller. About half as large revenue growth due to AI. And if we even did the calculation revenue growth per employee. Same thing, much smaller than what CFOs reported as productivity growth.

So it's a little bit like what Robert Solow was saying. We're hearing all about the promise of AI and productivity. But if you think about it, if you're producing more per worker, well, if you're selling those units, then sales should likewise go up maybe proportionally. Now it's possible that maybe prices are coming down a little bit and then sales revenue wouldn't necessarily go up even if units sold went up.

But we looked at our data. We don't think that's what's going on. Instead, what we think is going on is there's just a delay. So, for example, let's say your company where you work right now, ramped up on AI technology in the fourth quarter of 2025. You might not even produce new units because of that, until the first quarter of 2026, and you might not sell those units until the second half of 2026, depending on how it goes, because you probably still have, in some industries, a human salesperson out selling product for you.

That person has not been replaced by AI, right? And their normal process is still happening. Even if you have a warehouse full of inventory now that you want to sell. Okay. And so in the shorter version of all that is we're hearing more about productivity than we're seeing it in the revenue numbers. Now can we solve the riddle?

Well, it turns out when we look at what CFOs say about 2025 output per worker, that lines up really closely with what they're saying about sales revenue growth in 2026. So at least in our data, it looks like there's like a one-year delay before it might show up in the revenue data. So because of that, we're thinking that maybe this paradox is really just about a delay, and so we start seeing it in other parts of the data.

Kathleen Barrow 00:15:36

That's really interesting. I'll be excited to see next year's report and see if you know that revenue will catch up with productivity. So again, thinking about major technology waves of the past like the PC revolution, these things required heavy firm level capital investment to drive productivity gains.

You mentioned kind of at the top. This might not really be the case now. So curious if you could share a little more about what we're seeing with the AI wave.

John Graham 00:15:59 

Yeah. So again, if we compare to the computer age, if you will, you could not benefit from the computer age, the new age we were in without buying a computer, without buying hardware. Well, that's not necessarily the case with AI. Why? Because there are certain companies out there building the data centers and servers, and most of the rest of us are just kind of paying rent, if you will.

We have an operating expense, a subscription to their services. So there may well be physical capital investment across the economy again, that's really important, but at your typical company, we're not seeing it as much. In fact, when we break down the spending, we ask companies, what are you spending on? For small companies, roughly two thirds of their spending is on what we call operating expenses, or basically subscriptions. It's kind of like when we stream Netflix or whatever, you know, we're paying this monthly streaming fee. We're not investing in the hardware to produce movies or transmit movies or anything like that.

Another analogy might be it's kind of like leasing a car. We're not putting that big down payment up front that you would have to if you were building out all the physical capital. We're just starting to make monthly payments or quarterly payments, whatever they would be. And so kind of the good news is I think it helps more companies jump into AI faster because they don't have that huge upfront expense, because fortunately, kind of for the overall economy, there are these very wealthy companies out doing all that spending, so the rest of us can kind of just jump on their backs, if you will, and ride the AI wave. So I think that's a good thing for most companies. We should see them less likely to feel the burden of AI expenses. But it is important to know that AI is not free. I mean, yes, I can go do a super-duper Google search using Gemini or something like that. That's—at this point in time—free. But for a company who's actually using it in their operations, tailoring it to what they do, expenses are involved there. And so it's not free. It's just not on physical capital.

Kathleen Barrow 00:18:04

That’s taking me back to all my finance classes at Fuqua, thinking about those operating expenses versus capital expenditures. So thank you for that. I'd like to tie it back to something we were talking about before, which is the productivity paradox here. Dive a little bit more into productivity. You're seeing that the strongest productivity gains are tied to innovation and serving customers more effectively, and not just cost cutting.

Does that change how executives should think about AI strategy?

John Graham 00:18:28 

It could. Well. And this is actually, again, something that surprised me when we asked executives, where is this productivity growth coming from? I personally had expected cost cutting to be part of it, that if you kind of reduce the per worker part, then output per worker could increase, productivity could increase.

Again, we're not seeing big reductions in the workforce. So where could a productivity gain come from. Kathleen, as you said, one place is innovation, right? Now you might be able to produce something new that you couldn't do or a much better version of what you were already doing, through the use and help of AI.

And in fact, I'm going to go back to when you mentioned the four people on the team. Matt Gilliam used AI to produce this incredible data dashboard that anybody listening to this, if you go out and find our research paper on the front page, there's a link, you can click on it and get to see Matt’s data dashboard, and it summarizes a lot of the data from the project.

 And he did that without really coding himself just with using, you know, vibe coding through AI. So Matt innovated, right? He was able to do that. And that was sort of a new product, if you will, that he was able to produce. So anyway, companies through innovation—are using AI to innovate.

But equally as important, according to CFOs, is just meeting demand, if you will. We call it the demand channel kind of the economist lingo, if you will. But, making sure the product you're producing—the quality is excellent, and you're really satisfying demand even without kind of innovating outside of the box necessarily, but just delivering really high-quality product.

So those are the things that kind of loaded up when we ran statistical analysis to see what is it that's causing these hoped for and achieved productivity gains.

Kathleen Barrow 00:20:17

Now, I’m just seconding the professor's recommendation to check out Matt's dashboard online. It's really cool. Get to play around and see some of our results. So given your findings—modest job displacement, shifting skill demand and gains driven largely by innovation—what would you prioritize if you were running a company in 2026?

John Graham 00:20:35 

Oh, you're asking a professor how what the real world should actually do. You know, we live in our ivory tower here. So let me try to let me try to reach out here into the real world a little bit. Now, joking aside, I mean, a little bit repetitious here perhaps, but I think a lot of times, you kind of make sure you do what you do really well.

Don't lose track of what your excellence is about. Just do it better. I would say that's one thing. But secondly, yes, try to do the innovation part in the enhancement and improvement part through AI. Which again, we're not talking about cost cutting yet, but for now I would focus on quality I guess is the way to summarize what I'm trying to say. It’s just do what you do really well… you know, look, AI has had some embarrassing hiccups along the way, from the legal profession to many other things where it's not quite delivering fully on the goods, so don't count on it. Delivering high quality immediately has really important quality checks, I think in the short run. 

And always be thinking about your customer. Are you serving them best? Are they getting what they need out of your company? That should, I think, always be your target.

Kathleen Barrow 00:21:44 

So that's great advice for companies that are out there and thinking now about MBA students like myself, who are about to head out into our own careers. What skills or roles look more resilient based on what you're seeing in the data?

John Graham 00:21:54 

Well, I mean, you can look out there at the paper, by the way, to see specific skills and, you know, more specific than I'll be able to remember right now off the top of my head. But honestly, more technical skill set like engineers, for example. That’s  a position—I’m not sure how many MBAs are going into engineering per se—but that's one that's sort of a little bit protected right now and if anything, enhanced.

The good news is, right now, at least, it's more entry level, clerical work getting hit in the short run. But in a sense, I think what you want to do is do what you have to do for your setting to try to be one of the winners from AI, somebody that's not displaced by AI, but who benefits. I mean, I think everybody kind of is already thinking of it that way.

But here's something I really think is important, and I'm not the first person to say this by any means, but, you really have to understand what it is you're doing or what you're producing incredibly well, because the communication of what you're doing and related is still essential. So I'll just give you an example.

Way back when, before I went to PhD school, even I had a regular job. And my job, I took like three days and I made some table up with maybe ten rows and five columns. So 50 numbers in some table. And I ran in to show my boss. And again, it took me days to do this. And he looked at this and he pointed at two numbers on this piece of paper and he said, what's going on here?

What's going on here? And I went back and I was like, how in the world in like two minutes did he pick out these two numbers? Sure enough, I went back and I dug into it. One of them there was like a data entry problem, and the other, there was kind of an interesting explanation, but it merited further look. And I was so impressed by my boss.

But it turns out what that really was just experience, right? Once you are the one looking at those numbers month after month, you're going to see what jumps out at you too. So in today's world, AI could produce that table of 50 numbers in probably about five seconds or whatever. And yet looking at that and understanding, does this make sense?

What's important here? What do you take out of that? What do you need to communicate to your boss, to your client, to the board of directors, whomever. Whoever that person is crystal clear understanding of it, and communication of what your output is, is, is the key. That's where you're going to drive value, still.

And, in a sense, anybody can produce numbers today. It's really going to be understanding them and explaining them. And those, I think, those are going to be the real winners. And so I think this is one thing that, you know, people from a business school degree have an advantage at—that you get exposed to a broad set of classes and topics and issues, and I think it broadens you in a way that hopefully you're going to be able to look at these complicated things and say what it really driving this here and then communicate it well.

The other thing is that I still think interacting with humans is really important. Okay. Your teammates, yes. We talk all about team here at Fuqua, that's super important. But you know, it may be immediately or at least someday you'll be a boss of the people working for you or your boss you are working for… you can't, yourself become just a robot who spews out numbers, right? The human interaction and getting the most out of people is going to be crucial. And honestly, if you read the press, it sounds like some people are getting discouraged about work these days. And you might have some not that enthusiastic coworkers.

So if you can come in and get the best out of everyone, I think that's a skill that there's no way AI can replace that. So at any rate, I'm not sure if that's helpful at all but…

Kathleen Barrow 00:25:39

No, I think that's fantastic advice and honestly something that we hear a lot at Fuqua, right. Building relationships with people, critical thinking and communicating your ideas to others. So things are changing, but the game is often still the same in a lot of ways. Well, thank you so much for joining me today, professor. I really enjoyed the conversation.

John Graham

It's been fun. Thank you.

Kathleen Barrow 00:26:02

Duke Fuqua Insights is produced by the Fuqua School of Business at Duke University. You can learn more at www.fuqua.duke.edu/podcast