Why Executives Are Dangerously Wrong About AI's Impact

The latest CFO Survey, directed by The Fuqua School of Business in partnership with the Federal Reserve Banks of Richmond and Atlanta, suggests AI won't meaningfully shrink payrolls. In this episode, Professor Harvey argues that CEOs and CFOs are dramatically underestimating AI's reach. 

The real disruption isn't replacing clerks; it's AI rewriting code for security and speed, redesigning websites and apps, drafting and reviewing contracts, prepping earnings call questions, and serving as an always-on strategy consultant.

Harvey details how "agents" — not mere tools — are already reshaping legal, finance, and sales workflows. He also flags additional disruptions on the horizon: quantum computing, decentralized technologies, and "multiomics” (an integrated approach to DNA, RNA, and protein structure). 

The future is closer than most managers think.

Transcript

SPEAKER_00 0:00

I'm very glad to be back with our next episode here with Cam Harvey's Through the Noise. Cam, there's a lot of change going on. And we've talked about AI in the past, and I was um I was on the internet and I found something that your colleagues, I believe, and perhaps played a role in it at some point in time. The Duke CFO survey just came out, and I want to read something to you. We find little evidence of substantial AI-driven employment decline, although some companies, especially larger firms, anticipate a shift away from routine clerical work toward more skilled technical jobs and tasks. We've had several episodes already on AI, and from our own home turf, there seems to be findings that what we talk about is not true.

SPEAKER_01 0:52

Yeah. Well, uh that that quote is is really powerful in a very negative way. Um in my opinion, uh the CFOs and CEOs are sitting on a beach sipping a pina colada, not noticing that there's a tsunami coming at them and quickly.

unknown 1:16

Okay.

SPEAKER_01 1:16

So it's so it's a big disconnect.

SPEAKER_00 1:19

Well, so where's the error? Is the error within how we're running the survey or is the error in the environment? It's not a problem with the survey.

SPEAKER_01 1:27

So the Duke CFO survey has been running for almost 30 years. Indeed, I'm uh one of the founding uh directors uh of the survey. So the survey is good. Indeed, uh recently uh we partnered with the Federal Reserve because the Fed noticed that we were delivering information in terms of the future of the economy that was leading their information. So uh so we've got a very high-quality partner on this. This has nothing to do with the survey. So what does it mean? It has to do with the respondents to the survey. So this idea uh that they've got that these mechanisms, these AIs are just tools, that they can replace a few clerical staff. Uh this just says to me that the old guard doesn't understand um how to fully utilize this new technology. So we've got a technology that has got vast implications for productivity uh increases, but it's kind of brushed aside. And oh, well, we can replace a few clerical staff. And what are they thinking? They're thinking transcriptions, editing, stuff like that is old stuff. We've known that for at least five years. There's a next level that's much more important. Okay, go on. Okay, so um, and again, let me go through a few applications, and this is just scratching at the surface. So the number one thing that is so easy to do uh is any software that has been built at a company should be put into Claude code to figure out number one, if there's any vulnerabilities, and number two, if there are improvements that could be made uh in the code. So this is just like incredibly low-hanging fruit. So the cyber uh checks are very important, and uh the AI is excellent at detecting problems and offering improvements. You know, simple things. It's like companies pay tens of thousands of dollars for their websites, or they've got employees that they're paying uh much more. So no, you just give the AI what you want. It will redesign your website.

SPEAKER_00 4:19

This is low-hanging uh fruit. So it sounds like this has tremendous impact on the structure of jobs.

SPEAKER_01 4:27

So I'm only talking right now about the things that are related to software. So you can create new apps, for example, uh uh enhance customer service. All of this stuff is possible uh today. Uh it's also the communications of the firm. So think of social media, think of graphical design. Uh this is this is easy to do right now with uh AI. So anything in terms of the software that is being used by the firm, the communication strategies, anything legal. So AI should be reviewing all of the standard contracts that are used. The AI can draft new contracts. Those contracts, after the AI drafts them, need to be checked by a lawyer. But the expense for outside counsel or internal counsel is significantly decreased. So on outreach, uh the AI can do customer analysis. So for current customers and potentially future customers to really fine-tune a package to maximize the chance of delivering on a sale. Um, there's also kind of the review of financial documents. So companies prepare financial documents for the public, uh, and AI can review those to see if there's any issues, any potential red flags. And it goes beyond just the reporting. We can actually unleash the AI to look at the tax strategy of the company and whether there could be improvements to reduce the amount of taxes that are uh paid. So the company, if it's a public company, has got a quarterly uh earnings call. Again, the AI is really good at coming up with potential questions that the analyst will ask the CEO or the CFO. Uh so the AI is like a free consultant, and we know consultants are very expensive. So a free consultant can give some ideas about the structure of the company and uh suggest improvements, point out weaknesses. Uh, it can analyze investment plans, uh, it can again make suggestions, provide analysis that might take weeks or months of work by a team. And I'm not saying the team disappears, the team is downsized, but be cut much more productive. Indeed, I've also got this uh this idea of um, and it this is just an idea, of having the equivalent of an AI board of directors. So you train the AI on either your current directors or your dream directors, people you would like to have as a director. And all of those AIs consider them uh agents. And then when you're thinking of making a decision, you bounce it off your AI board of directors to see what they think. And and again, these are designed to kind of mimic uh the decision processes of certain people. And I it it is something with the board, it's often meeting quarterly. So you can get some input in real time. And that's what we need. In today's age, speed is really important, and you need to make decisions quickly. You cannot afford to wait uh for a quarter. And of course, really critical decisions you might need to call a board of directors meeting uh before the end of the quarter or the end of the month. Um, but I'm talking about the everyday decision making that is so crucial for companies.

SPEAKER_00 8:54

So it sounds to me what you're what you've been talking about is the the CFO survey captures people looking at the status quo and kind of immediate first order effects, but they're not looking at second and third order effects of what happens once you have a team that's capable of seeing these potentials. So and it sounds like to get there, you must have been using a lot of these tools. So tell me, what what AI do you use? Yeah, so just a step back.

SPEAKER_01 9:22

We've got these tools, and the problem right now is that many companies don't know how to utilize the tools, nor have they thought about creative application of these tools. They see the low-hanging fruit, but in my opinion, this is a much bigger uh set of problems that uh AI can address. So you say, But what tools uh do I use? Well, number one, these are not tools. So a calculator is a tool, uh, a laptop computer is a tool. So what we're talking about is something different. These are agents. So agents are possible um to do things that that we could not do with the previous tools. So that means organizing a plan, delegating to sub-agents, drawing all the information together, and making decisions. So this is this is much different. And in terms of what I personally do, um uh it it's also a mistake, in my opinion, to use just one AI tool. So I use four. So uh Anthropics Claude, OpenAI, Gemini, and Supergroc, all four of them, and they're communicating with each other. And by kind of iterating around with the different AIs, I get better decisions uh or recommendations. And uh again, I want to be careful here. The human element is still very important, but the human can be so much more productive with these tools. You see, I said tools? Yeah, that's that's you know my mistake with these agents. And one last thing AI is only one of the disruptions that's happening right now. There are four. So it's AI, it is quantum computing, it is decentralized technologies, and number four is something called multi-omics, uh, which is kind of the blending of DNA, RNA, and and the protein structure. The four things going on, they're all connected uh together. And uh maybe this is something we could talk about.

SPEAKER_00 12:08

Well, it it's a lot, Cam. Um I think we have to update our mental model today just on tools versus agents. Uh I trust that we're going to return to some of these larger themes, uh, both individually and together. So thanks so much for the time today. Um, it's a really interesting time to be having these dialogues, and I look forward to the next one we have. Okay.