In this episode of The Parlor Room Presents: Hello AI, host and Harvard Business School Online Creative Director Chris Linnane speaks with HBS Professor Joe Fuller about how AI is transforming the future of work. Fuller explains why organizations must rethink how work gets done—rather than simply adding AI to existing processes—and what it takes to build the skills and structures to use it effectively.
Resources
Learn more from Fuller in the HBS Online course Management Essentials
You can also watch The Parlor Room on YouTube
Transcript
Editor's Note: The following was prepared by a machine algorithm and may not perfectly reflect the interview's audio file.
Joe Fuller: What companies have been doing too much is saying, "How do I append AI to my existing process?" That's the wrong question.
Voiceover: Joe Fuller is a professor at Harvard Business School, and he's spent 15 years studying the future of work.
Joe Fuller: Too many people essentially take out a scissors and cut the right-hand side of the distribution off, and don't spend enough time mining that for value.
Voiceover: The real opportunity in AI, almost everyone is missing it.
Joe Fuller: What are the risks in these processes? You'll notice I haven't mentioned the chief technology officer.
Voiceover: The disruption isn't coming. For millions of workers, it's already here.
Chris Linnane: There's this word. I don't know how you feel about this word, Joe, but there's a word that breaks my heart: re-skilling.
Voiceover: Meanwhile, AI is doing something nobody expected.
Joe Fuller: And then, about 100 moves later, that was the move that had won the game, and they were all like, "Wow."
Voiceover: Welcome to The Parlor Room Presents: Hello, AI.
Chris Linnane: Before we start, I wanted to take a quick moment to thank you for watching this season of The Parlor Room. If you're watching on video, you'll see that I am in the Klarman Auditorium. Some really amazing things happen in this room. Anyway, I wanted to thank you for listening, for watching, and please subscribe, if you can. It helps us out a lot. Thank you very much, and enjoy the episode.
Chris Linnane: Before AI took over so many conversations about work, was there a moment when you personally started seeing jobs just differently?
Joe Fuller: Yes, I think, actually, in the last decade, jobs have been changing at an ever-accelerating rate. Our research shows that about a third of the skills in a job turn over every three years, and that was pre-AI. We're seeing a couple of phenomena.
One is: Generative AI is the next in what have been a growing number of really pretty disruptive technologies. It's the headliner. It's the one that everyone's gotten all excited about. And it is a general-purpose technology, which is the big difference from earlier versions of AI, and the development of sensors, the development of computer vision, and the development of voice recognition.
So, what we've been seeing is that more and more, what we think of as hard or technical skills are being disrupted or displaced, or augmented by AI, and that the consequence is that job descriptions change to feature, first of all, more human skills. What we call social skills become a bigger percentage of the skills that are being sought, because if the ratio of technological skills in the job descriptions is falling, by definition, those human interpersonal skills become a bigger part of what people are looking for. So, change has been a constant. I know it's a terrible platitude, but the acceleration of change in job requirements has been going on for some time.
Chris Linnane: And why does it feel like AI's forcing companies to rethink how work is organized, and not just what tools we're using? So it feels like it's not just like, "Oh, I'll pull this in." It's changing much more.
Joe Fuller: I would actually argue that companies aren't doing that enough. That companies don't have any experience in adopting general-purpose technologies. No one here lived through moving from steam power to electricity, and very few senior management teams, really, were senior leaders during the launch of the internet, which is probably the closest analogy. Although I think the real innovation at that period of time that caused me to sit up and take notice about the internet was when the Mosaic browser, that could move the internet from alphanumeric only to have images and photography, and that was just—my jaw dropped, and I said, "The world's going to change now."
So what companies have been doing too much is saying, "How do I append AI, particularly generative AI, to my existing process?" Almost like it's just another big new SaaS system. And in fact, because it's a general-purpose technology, that's the wrong question. The right question is, "How do I reconfigure my process around this general-purpose technology? Around the AI?"
And yet, you're seeing companies—I would point to companies like Procter & Gamble, and Coca-Cola, and Consumer Goods, JPMorgan Chase—who are being, really, pretty aggressive in the way they're deploying AI. Google, obviously, is a core AI innovator. But what those companies are doing is taking very important processes and reconfiguring them around the AI, and that's what this calls for. And I think a lot of the numbers we see about the failure of AI experiments are, yeah, badly configured experiments. Fail. You'd be surprised.
Chris Linnane: And when they're putting together through successful experiments and pushing, what part of the org chart is the pressure hitting the most?
Joe Fuller: Yep. Right now it's hitting entry-level and lower managerial jobs, and the reason for that is those jobs often have tasks that are heavily—what I'm going to call—rules-based. A good example is a credit analyst. If you get hired out of an undergraduate business program to be a credit analyst at a bank or for a company that does what's called vendor financing, I will lend my customer the money to buy my goods.
When you start off, you're basically just applying a set of rules. You've been told companies of this size, of this history with us, of this volume, with us, not more than X dollars—here are our terms, here are the interest rates you're going to charge, et cetera, et cetera. And usually for large companies, which are quicker to adopt AI because they can afford it, they've got a lot of data. So, rules-based decisions where you've got lots of longitudinal data, that's almost a perfect environment for AI.
So, a lot of those tasks that have those features tend to reside in lower managerial ranks, and the people they supervise, who tend to be entry-level workers or what a lot of companies call individual contributors. Who are out of that entry-level kind of probationary status, but are but are still doing core work of the company, but they're not decision makers.
Also, a lot of lower-level managers do a lot of routine reporting. They take the sales report and the inventory report, and the production schedule, and write the bimonthly uncertainties we're facing into considerations for middle management or upper management of changing our plan. Though relating two or three databases and integrating them into a coherent written document, particularly if you can upload 10 or 20, or 2,000 previous versions of it, it'll do a really great draft.
Now it won't be exactly what you want, but rather than looking at that accursed blinking cursor, where you have to write the first topic sentence for the first paragraph, it's much easier and quicker to edit than it is to create. And for routine work like that, it's really, really effective.
Chris Linnane: Who owns those experiments? Like, what part of the organization—is that coming straight from the top? I know they have to support it, but do they actually have to make it happen for most successful scenarios?
Joe Fuller: This is a really interesting question.
Chris Linnane: Yeah.
Joe Fuller: I strongly believe, and I think, really, our work across issues concerning the future of work strongly suggests that having a senior champion or responsible business executive driving the change is a sine qua non. So it could be you're trying to get supervisors to be more aware of and sensitive to the caregiving obligations of the people they supervise. If you've got a senior executive who's championing that, walking the halls, providing a narrative for calling out people who are doing a good job, you get a change. Same thing with the adoption of AI.
But the thing about AI that makes it harder to govern is that it's not best thought of as a technology, which may sound strange. Because when we think about technology, we're going to get a faster router, or we're going to get new machine tools, or we're going to get new mobile devices. Those are technologies. Really, what AI is is a way of working.
With some of the companies I'm advising, what I'm suggesting they do is actually create a two-headed leadership model. One is the chief people person, the chief human resources officer, it's often called, because all the fundamentals of work of the employee base are going to be exposed to change because of AI. So you have to understand job descriptions and progressions, and incentives and metrics, and performance management.
But the other person you want involved is a very experienced operating executive, someone who really understands what quality outcomes in the core processes of a business look like. What are the risks in these processes? You'll notice I haven't mentioned the chief technology officer because this is not a set of adopting. This is not a question of the technology. It's a general-purpose technology. It's really easy to use.
So the real question is, "How am I going to deploy it, and how do I need to ensure that, as it gets deployed, we're not exposing ourselves to a lot of risk?" We've got enough human oversight of the AI to help prevent hallucinations from doing damage to us. And also, importantly, to spot when it comes up with something highly original.
This is something that doesn't get talked about much. When you look at writing about what I'm going to call non-standard outcomes from the AI, a standard deviation or too negative, in terms of quality—what we call hallucination—or the literature highly focuses on those risks. But what we also see is that it often comes up with highly original and valid ideas. And so you need someone who can look at that distribution of outcomes, spot the problem areas—spot the hallucinations—but also pick out what I'm going to call the inside areas where it generates something, which causes a knowledgeable person to say, "Wow, that's interesting. I want to know more about that."
Just to give you a quick illustration of that: There's a famous episode where the leading Go champion from Korea played against a generative AI and got swept. But in one of the games, about early on in the game, the AI made a move that all the observers, who were all expert Go players, said, "Oh, that's a huge mistake." That just shows the machine isn't perfect.
And then about 100 moves later, that was the move that had won the game, and they were all like, "Now we understand what it thought. Wow."
Chris Linanne: Yeah.
Joe Fuller: So it creates a normal distribution of outcomes in terms of quality, classic bell curve, and too many people essentially take out a scissors and cut the right-hand side of the distribution off, and not spend enough time mining that for value and focus on the left-hand side of the distribution where the risk is.
Chris Linnane: Wow. Now, this is interesting because I've never heard it described that way before, but the idea that you're not bringing up—the chief technology officer and not thinking of it as a technology—is kind of wild to me, but it makes complete sense. It's really fascinating to think of it that way. Now, a lot of leaders right now, AI is here, but they're still moving slowly and cautiously.
Joe Fuller: Yes. Yes.
Chris Linnane: I mean, there are some reasons I think that's happening, but maybe you could talk a little bit about that.
Joe Fuller: I think there are four or five.
Chris Linnane: Oh, OK. More than I was imagining.
Joe Fuller: The first is that a lot of companies have been nervous that they'll have what they call data sovereignty issues, where somehow their data escapes into the wild. That their employees will use it in ways that are problematic, like violating other people's copyrights or trademarks.
And so the first barrier has been cautious adoption. That manifests itself in lower investment in training than you'd think. A lot of companies aren't actively training people, and also they're not buying licenses for a lot of people. Our research shows that Americans use it 50 percent more at home than they do at work.
Chris Linnane: Wow, OK.
Joe Fuller: So there's a lot of the—I think of it as a high school science project AI going on out there, for all you helicopter parents in the audience.
A third is that early experiments in a lot of companies weren't very successful, in the eyes of the users. When you double-click on that, you do find something which is very often not some problem with the AI. It's a problem with the data. That companies have found that their internal data isn't what they call clean. It's not accurate. It isn't tagged well. It's hard to access.
If you look at the big enterprise resource planning models like Oracle and SAP, they are not configured to make using AI or overlaying AI easy. It's not some purposeful impediment being created by those companies that impede it. It's just that these systems, architecturally, are all from the 1990s, so they're quite cumbersome. And if you go to those companies' websites, you see an awful lot of, what I call, the coming attractions. Lots of promises about how they're all going to be AI-friendly and have lots of AI built in, but it's always mañana.
Chris Linnane: Yeah. Yeah.
Joe Fuller: So bad data or faulty data, frustrated experiments, lack of investment in human capital. And then a funny one is that this came on so fast that the normal rentable talent, usually in the form of consultants and from big firms, they weren't ready. So they woke up in 2023 and started looking at OpenAI, and they had the exact same reaction as their client, which is, "Wow, what is this and how do I use it?" So historically, if you were buying a big new system, you could call up Deloitte, you could call up Accenture, you could call up Infosys, and they were way down the learning curve. Not this time.
Chris Linnane: Not this time at all. Everyone's getting to the same spot. So those are five good reasons. Is there ever a scenario, or do we know of any scenarios, where the company—I don't know a better way to say it—but more to protect the human as their reason for not going in? Is there like, "No, I have too much value in what humans can do, and I don't want to automate everything," or—
Joe Fuller: I don't think we've gotten there yet, but I think we're approaching that for a couple of reasons. One is there are a lot of champions for AI, and I'm generally bullish that, over time, AI is going to have a favorable impact on work, that it'll create new types of jobs, and it'll be another of a wave of technologies that have taken—we used to talk about dirty, dark, dangerous jobs, and that a lot of automation—take an extreme example, coal mining, super automated now, because it's literally all three: dirty, dark, dangerous.
There's a fourth D, which is dull. And in a post-industrial society, an awful lot of white-collar work is dull, or at least elements are. And so, I think we have an opportunity to reduce the ratio of dull, non-fulfilling work in a lot of jobs. All good.
But there are real concerns that AI is not an unalloyed good. One of the effects it's going to have, as our earlier discussion indicates, is that it's going to crimp the entry points at a bunch of careers. So go back to those credit analysts.
Right now, you wouldn't entrust AI to set your credit policy. You could ask it to opine on your credit policy. You could ask it about what they infer about other people's credit policies, but I don't think any responsible business executive would say set the credit policy to an agentic AI and send out the communiques to all the people that have an influence over the credit we extend. That would be highly imprudent.
Well, usually those people who set credit policy or do complicated credits, or process exceptions, they came up as a credit analyst. You can't have a 10-year-old if you didn't have a one-year-old. So I think there's that concern. I think there are some solutions to that. I think there's also a growing suspicion and anxiety among younger workers around AI. Part of that is fueled by concern that this is going to seriously affect their career prospects. And for many, it will, so it's a legitimate concern.
But I think also, they have well-founded concerns about the universal efficacy of AI, about the role AI is going to have on—the capacity of AI, for example, to come up with ways to entice people to buy more, to be less prudent, things like that.
It's pretty clear that one of the things that AI, the better consumer-oriented companies do with AI is they create what we call synthetic customers, digital representations of either customers you've got, that you can then target based on their historical data you've got about customers. New features, new offers, new pricing, new promotion.
I don't know this, but I have a very strong sense that the online gambling companies are in the vanguard of that. If you look at the innovations that they advertise about different ways to bet—even the messaging—there's one, I don't remember which one it is—their tagline is, "It's good to be right." And I don't bet on sports because I'm regularly astonished by who beats whom, but I think that betting on sports can become a pathological, harmful behavior really quick.
Chris Linnane: Oh, yeah.
Joe Fuller: And that we have a growing epidemic with young men in the United States suffering from that. The same ads will have, "Oh, by the way, here's the help line."
Chris Linnane: Totally, I think about that every time I hear it. It's like, somehow, that just erased all of the negativity, that you're giving people a helpline.
Joe Fuller: Yeah, exactly.
Chris Linnane: Scary.
Joe Fuller: So I think that there's going to be very meaningful and profound discussions and debates about this, and that that will be influenced by this generation that's doing the high school science projects and are broadly concerned about affordability, their prospects of being prosperous, replicating their parents' lifestyle. This will be a hot topic under that broader umbrella.
Chris Linnane: Yeah, I'd like to go into that a little bit more. I have two teenage daughters and one younger, and I'm thinking about that all the time. Are we in a position where the first one or two rungs of the career ladder are just going to be gone?
Joe Fuller: Evaporating?
Chris Linnane: Yeah.
Joe Fuller: I would say that the question I get most often...I do several dozen speeches a year, and this obviously features in it—and I would say the most common question I get after the speech, when I'm mingling, people come up, is, "What should I tell my kids to study? Is this going to be a threat to my kids, et cetera?" Very human. Human and understandable reaction.
Yes and no. I did with my frequent collaborator, Matt Sigelman, a graduate of our school, who's the founder and CEO of the Burning Glass Institute. We looked at how entry-level jobs would get affected, and it showed about 12 percent of entry-level jobs could basically be displaced right now by AI, and AI today was going to be more effective than a new employee.
Chris Linnane: Wow.
Joe Fuller: But it showed about 17 percent of jobs would actually become more accessible to more candidates, because AI was going to take over tasks that require credentials or training that are hard to get. So let's just take a basic knowledge base—a good knowledge of statistics. Which, by the way, just a vent—a pet peeve of mine—the terminal math course for high performing high schoolers, BC calculus. Almost nobody uses BC calculus, other than rocket scientists and engineers, very advanced engineers.
The terminal math class for a high-performing math student should be statistics, because statistics is everywhere in your life. If you're not strong on statistics, I'll tell you who is. Claude or Anthropic, or Gemini, or Gemini or OpenAI. So if you had a job, historically, which really required college-level statistics or even beyond, all of a sudden, that's not going to be required anymore, and maybe I'm going to be hiring more on those human skills.
So I think the solution here, by the way, which will be imperative for employers, but you're not seeing quite yet, is an employer that's thinking about this tactically and not systemically is going to say, "Oh, good, I can get immediate efficiency by curtailing my hires next year, because AI is going to make the fewer hires as productive as the old cohort of human beings." A more thoughtful employer is going to say, "How am I going to change my entry-level jobs to make them more reliant on human being skills, as the AI absorbs some of the tasks that are in those jobs as currently described?"
So I'm working with a large technology company, and as many of your listeners will know, the emergence of Claude Code and Cursor, Copilot has made entry-level software programming and engineering jobs hard to sustain because the technology is so good. It's making senior programmers—senior software engineers much more productive, but they still have to—AI writes kind of flabby code. It's right, but it's not tight and lean.
What I told this company is to stop hiring people to...don't make your lead entry-level position for technically educated talent programmer. Make it a product manager. Bring in the people you were hiring as junior programmers as assistant product managers. They will come in, they'll learn your product line, they'll understand customers. They'll start developing more practical skills in terms of programming.
And that lift responds to the lift of complaints that customers made about entry-level programmers historically. They don't understand customers. They don't really understand product. They know theory, but they can't really knock out code. Well, give them a new entry point that addresses those things you didn't like about those workers to begin with, but also keeps your talent base evergreen.
Because if everyone in every industry takes a maximum efficiency stance, and then they wake up and they need a bunch of 10-year-olds. What choice do they have other than raid each other? So you want to be the company that's sustaining that talent pipeline and cultivating one, starting now, with your entry-level workers for what you're going to need, as opposed to what you used to need. And the average entry-level job description in the United States for a lot of companies doesn't get rewritten very often, because usually people just add a paragraph to it.
Chris Linnane: Yeah, so it almost sounds like it's got to come from both below and above.
Joe Fuller: Yes.
Chris Linnane: You have to think, "What is the thing I can be better at?" Which seems like maybe some more human skills. But the employers also need to think of, "What's the position we actually need that's going to help our company?" Not just axe certain companies, odd jobs.
Joe Fuller: Yeah, I think that, in fact, this is really going to put a new set of obligations and challenges in front of companies. I think it's primarily the companies that have to think that way.
Chris Linnane: Yeah. Yeah. It's fascinating. If I think about, I like this. It's a slightly more positive way to look at this that you're seeing right now, and that makes me feel a little bit better. Because I look at my kids and other people getting out, and I'm thinking it's almost like they have to take any job they're given. Because I remember I had an intern job at a marketing agency years ago. I mean, years ago. And the woman who was running it came up to me and said, "We'd like to offer you a full-time job."
And in that climate—I was in a band—and I was like, right on the spot, I was like, "No thanks. I'm in a band, and there's no way I can do that while touring." But in that period, it's like there's always more jobs coming. It felt like that 25 years ago or whatever.
But now you can't have that mindset, I don't think, because it is still uncertain, and these things will make it better. It sounds like, if everyone's thinking about what the future looks like, it won't be as dire as what we're hearing, maybe, but there's still some fear there that—I have, at least for my kids, of being like, whatever job is offered to you, make sure you're really considering it—if you even get a job offer. It's scary.
Joe Fuller: Well, all my kids are in their thirties, and they're well employed and seem to be happy. They're certainly healthy, but I still worry about them. So you're never going to escape. That's not going away. I will say, though, that one other thing. I think, in fact, companies are going to start understanding that they actually need to adopt a new and more active stance in this dimension, and that one of the things they're going to conclude is they want to do a lot more things like internships, apprenticeships, because they're going to want to see a prospective worker in action.
Historically, companies have relied on credentials that are easy to validate and are inferential. So, what is our history with graduates of this person's alma mater? What was their grade point average? What was their major?
When I was a CEO, we focused on specific schools because we had plenty of people who had been there recently. You could look at a transcript and say, "This is a hard semester here, and boy, that one B they got that semester, that's the toughest professor at the university," that sort of thing.
Those things are going to be less and less relevant. Why? Because technology is changing so fast that the academy, whether it's K through 12, community colleges, four-year universities, business schools aren't going to be able to keep up. We're not built for speed in education. In fact, we've already seen, between 2000 and 2025, a significant acceleration in the number of companies saying, "I can't find the skills I need," just in that quarter century, because that's the digital quarter century. That's the AI now in five years.
And the consequence of that is that to evaluate somebody's fitness for the role, I'm going to want to see him in it. And rather than relying on higher ed to just dropship me someone I can mold into what I need—particularly as these social skills become more integral to the value proposition—I think you'll find progressive companies are doing things that companies currently don't do.
I've been advising a company that's in an industry where, it's kind of a technical skill, the supply of graduates is shrinking, even though demand for the service remains healthy. I'm telling them, "Have your first internships be after sophomore year, not after junior year." If they're very good, bring them back second year and make them that offer as they're walking out the door. If they're successful junior year, pay for their senior year in college if they accept your job offer. And you pay for it in the form of a demand note. So if they leave in less than three or four years, then they owe you some of it back.
I'm trying to imagine someone with an offer from a prestigious firm that goes home to their parents and says, "Gosh, they offered me a full-time job, and they'll even pay for my senior year." One of two things happens: Either the kid accepts the offer or their parents murder them. But either way, right?
Chris Linnane: Yeah. It's got to be one or the other.
Joe Fuller: So you have to start thinking like that. Simultaneously, companies are going to have to get much better at formal training than they are today. Formal training budgets have shrunk in the last quarter-century in most of these companies. Part of it is the endless quest for efficiency. Part of it is a vast over-reliance on vendor-provided training, which is part of buying their technology. Doesn't matter if it's a machine tool or Salesforce.com.
As technology's half-life shrinks, the capacity of human beings to learn a new technology doesn't match that progression. So companies are going to have to get good at really quick, bursty, effective training. Not just about how this technology works, but how does it work in our context, so that those people can get productive much faster than they do on average today.
Chris Linanne: That's perfect. And I have a little bit more I want to talk about that, because that's a great topic. But before we do that, I want to try to make this—some of the things we talked about, I want to play a quick game, if that's OK with you.
Joe Fuller: Should I fire up AI to give me some quick answers to these?
Chris Linnane: Yeah, yeah. This is good. You don't need it. It'll be real easy. Alright, so let's pretend we're in a big corporation that's doing well. We still have a career ladder in our organization. I'm going to give you a few scenarios, and in those two, you have to trade off between two things. But you can only pick one trade-off, if that makes sense.
Alright, so let's do the first one. And this is all tying into what we've talked about: entry-level roles. Do you design entry-level jobs for learning or for immediate productivity?
Joe Fuller: Going forward, for learning. It's because the things that are knowable to an entry-level person are much more likely to be exposed to automation of different types, and specifically agentic AI, but other technologies as well, than historically. So if I'm thinking about cultivating talent for the future, what I'm most— productivity is going to be less in the future about: I can immediately come and do a fairly straightforward task that an inexperienced person can do well. That's not automatically something that can be automated, but many of those tasks will be automatable. So I want to get someone up to speed as fast as I can about a word I used earlier, which is context.
It's context that is the difference between the one-year-old and the 10-year-old. It's context that allows someone to catch the insight in the hallucination. It's context that causes someone to be able to understand that even though a technology is generating an answer that makes sense, relative to the way it's trained, it's not what the company stands for, or it's not something that you're comfortable doing morally, or it's not something that would be consistent with the regulations or mores of the business.
Chris Linnane: So yeah, you come in learning because you have to be able to understand in that context and provide that context. You're not just going to show up and start performing.
Joe Fuller: Yes, exactly. And employers, by the way, currently complain already about how much they have to invest, even with a highly credentialed person, to make them productive. And I think they've got it backwards. Or at least we're going to an upside-down world now, right? So, in an AI environment, you have to embrace that and launch someone into your company where they're getting that familiarity and contextual intelligence much faster, because then they're going to be able to take these very extraordinary technologies that are emerging and really put them to use. And that's where you get your productivity pop.
Chris Linnane: Perfect. Alright. Scenario two: Hiring signals. Do you hire for AI fluency or for raw judgment? And listeners, at home, you play along while we think about that.
Joe Fuller: Given that dyad, I'd say for judgment, but I'd probably expand that to include a broad set of social skills: capacity to deal with different types of people, spontaneous written and oral communication, negotiating skills, not negotiating with baseball players' contracts. But how do I, when I have a point of contention, not necessarily argumentative contention, with a fellow colleague or with a customer or supplier, how adept am I at getting to a mutually acceptable resolution of that?
Those are things that are going to be integral to anybody's success to an increasing degree, and they're hard to evaluate using those inferential credentials we talked about earlier. Now, if someone was a captain of their sports team or president of their sorority or something, that's probably a leading indicator that they've got some of those skills.
But nonetheless, for most jobs, certainly for middle skills jobs, which are defined as a job that requires something more than a high school diploma but less than a four-year degree, the principal reason for both voluntary and involuntary turnover—quitting or being fired—are social skills deficits. Because it's pretty easy to validate whether someone's got hard skills, technical skills.
You're getting hired as an IT help desk person, you walk into a room, there are four laptops, each of which has a different type of problem. The IT department has set it up that way. This one can't print. This one won't boot up. This one has lost a bunch of data. I'll be back in an hour. Have them all up and running. Either they're up and running, or they're not.
Chris Linnane: Sure. Yeah.
Joe Fuller: But is that person, can he get them up and running, but does he also have really poor interaction skills, and gets frustrated and angry if someone's being impatient with him, or someone didn't describe something in a way that made their work efficient?
Chris Linnane: The design of this organization is our final scenario. Do you centralize AI expertise or embed it everywhere?
Joe Fuller: You embed it everywhere, and this speaks to some of the points I was making earlier about telling your workers you've got to be careful. You've got to be cautious. You can't use it for this. You can't do for that. We're not going to buy licenses for you. We're not going to train you. What does that do? They learn how to use it at home. Most people basically use it like a glorified search engine. They're not good at prompting. They don't understand the power of the tools. They don't know lots of street-smart tricks.
I have four models that open automatically in Chrome, and I crosswalk questions and findings between them. And if I want something that's going to be more visual, I'm going to start with Gemini. If I want to start something that's more quantitative, I'm going to start with Anthropic. If I'm going to do something that's going to start more in narrative and writing, I'm going to start with OpenAI. And I don't start with Grok with anything. That's the one I use the least. And if I want to search for something, I start with Perplexity, not Google.
You want to get—what we're seeing, by the way, and companies are a little more progressive—is lots of user innovation. So if you've got a user that's relying on a free AI that they're using off the books—if people think that their employees aren't using AI right now, even though you've put a lot of restrictions in, I got a bridge across the Charles River. Be in touch with me. I've got to sell it to raise some cash. I'd be happy to sell you one. But they're learning the wrong things, and then they think they know what they're doing, but they're not getting the most productivity out of it.
Now, embedding it, if you're running a big company, tens of thousands of employees, it's a big budget disconnect to suddenly buy everyone a license. So it's OK to say, "Here are the five processes I'm going to start, or the three processes or three operations I'm going to start with, and embed it in that from top to bottom." You don't have to start; you don't have to do it across the board at once.
And I mentioned the two-headed monster of the CHR when a senior operator is really running this transformation. What a CHR at a big company is going to tell you is I can't rewrite every job description, every basis for performance management and evaluation, every career path in the company at once.
I mean, a state government in the United States has about 25,000 discrete positions. So if you went to a governor, and their head of administration, and the head of personnel, and said, "Oh, we're going to make the state all AI"— which, by the way, should be being done on an accelerated basis. A ton of government work is highly amenable to this—and all you have to do is, in the next three months, rewrite 25,000 different job descriptions about multiple different cabinet posts, boards, commissions, it would be literally undoable.
Chris Linnane: Yeah, that's a little overwhelming to think of it from the numbers perspective. There's this word—I don't know how you feel about this word, Joe, but there's a word that breaks my heart: re-skilling. I hear it all the time, and it deflates me because it feels like such a way to take the human out of the human. And the idea that someone, sure, they can learn a new skill, but maybe that first skill, they acquired it over years and years, and they acquired it from family, or they acquired it from heartbreak, but they acquired it from something. And then we're just like, "Yeah, re-skill them to do blank."
How do we look at that—and maybe I'm the only one, but I don't think so. How do we look at that as, how do you make it feel more like an opportunity and less like, well, that thing you did doesn't matter? That special thing that you've developed doesn't matter.
Joe Fuller: Well, we've got a big project run by Professor Raffaella Sadun and working with Professor Jorge Tamayo at the School on that, and I am spending time with them on that question. And we've done, actually, a survey on re-skilling of how workers experience it and how companies are thinking about it. And there are several things that jump out of that. This is another manifestation of something we talked about earlier, where there's a certain level of discomfort and anxiety around this that employers have got to embrace and accept as legitimate.
Also, companies really aren't very good at it. They really don't know what to do. Is it on the clock? Off the clock? Is it a requirement? Is it a preference? Do you make it part of the evaluation process?
We found an interesting thing, by the way. Professor Tamayo really pioneered this. The number one variable that explains whether an employee engages in and successfully completes a re-skilling program is the enthusiasm of their supervisor—that of their supervisors. That's great. Don't worry. We'll cover for you. That supervisor will help advance a learning organization. The supervisor says, "Hey, wait a second. We've got to finish this project you're working on. I'm really busy. I can't get to knock this out at the same time, or you can't take that program right now because I need you."
I think, once again, there's a lot of solutions here, which comes in the form of job design. What did we talk about earlier? The importance of context. What does that older worker have? Context. They've been with the company for 10, 15, or 20 years. They know the way we do things around here. They know what the don't do's and the must do's are. They've got credibility with peers, with customers, with suppliers, with regulators, with whomever, but they don't have some of these technical skills.
A large company I'm currently working with—and I've been talking about this—and my recommendation to them, which they are now experimenting with, and the early results are very promising, is you have to think very differently about team formation in this world. Why? The younger the worker, the more likely they are to be tactically nimble, and the less context they have. The older the workers are the inverse.
So rather than say we're going to have a task force on X and such, so we're going to take our 10 best individual contributor-level people. They're all around 30. They've been in the company five years, and they're going to work on it. No more teams that are cut horizontally. Cut them all vertically. Put the 60-year-old in with the 24-year-old.
Chris Linnane: Yeah, better output.
Joe Fuller: And the 60-year-olds saying, "That's not the way it actually works." And the 24-year-old says, "Oh, well, how does it work?" And you're not asking the 60-year-old to be nimble with data analysis and a good prompt engineer. Let the 24-year-old do that.
Now we're seeing some data, and there was a very widely noted and sad example from Accenture where Julie Sweet, the CEO there, said that they were letting go of 10,000 people who had said, I'm not interested in learning this stuff. That's a really interesting debate.
I was a CEO for the better part of 30 years. There were a bunch of tasks that just weren't voluntary, and a lot of CEOs in the last three years have been saying, "Actually, coming to work isn't voluntary," and things like that. So I'm very sympathetic with her. And that was a mammoth company, and obviously, a technological leader. So the notion that there would be a significant number of people there who are very resistant to learning a leading-edge technology might be a different discussion if we're talking about the Gap.
But also, I mean, those are people who are employed that, probably, particularly with that sort of attitude, are going to find it hard to replicate the same wages that they're making now. They were constructively employed before these decisions were taken, so they must have been adding value the way they were doing things. So good case discussion about how do we feel about those trade-offs? What's the obligation of both parties?
And I think a lot of the discussion you see in the press misses the fact that it's a negotiation between two parties. It's not that the employer has to do X or the employee must do Y. It's what's the combination? Now, five years from now, what do you hope companies have changed most about how they build careers?
I hope that they have crafted career job descriptions and career paths that allow people who are closest to the task the highest level of control possible of the task. In the DNA of all organizations is the logic of scientific management, that the worker should be instructed not only on what to do, but exactly how to do it, and essentially be controlled from deviating from the way the job has been defined and described to them.
It's not fit for purpose anymore. By the way, Harvard Business School was a huge pillar of scientific management principles for 100 years' worth of managers, so we're certainly an indictable co-conspirator on that one. But scientific management is so far past its sell date.
And actually, there were some wonderful articles written about it, including one by our former Dean, Kim Clark, in the early 2000s, saying that it's dead. I got to tell you, it's alive and kicking in companies, because no one's come up with that alternative logic. So we're heading to a world where the skills that people have and their ability to articulate those skills in productive ways are going to be the big source of competitive advantage for companies in the future. Not the tech they own, because everyone's going to have the tech.
So how you unlock that becomes the design imperative. And that means rethinking the way you incent people; the way you hire people—more experiential learning, for example—the way you give people feedback. Moving away, as a number of companies now have done, from, we're going to have the annual or the semi-annual review, which is just riddled with bias, mostly frequency bias. I remember what we've done over the last eight weeks. I don't remember what we did in December. Bad preparation, whatnot. So lots of real-time feedback.
Interestingly, we talked about that GDP analysis that OpenAI does, evaluating tasks. AI coaching, in many instances, is close to 50.
Chris Linnane: Really?
Joe Fuller: Yeah.
Chris Linnane: Wow.
Joe Fuller: And it turns out that, guess what? You've got patterns of failure which are recurring and probably captured in your performance reviews, and may be known to you. So that when it pertains, either coaching you before you go do something, it kind of knows you're a sucker for the curveball or that you chase high heat.
Chris Linnane: Baseball terminology, for anyone—
Joe Fuller: Yes. I'm sorry. Or you hurry your first free throw, or you have this unfortunate habit of having something you think is funny, but you're not sure, but you say it anyhow. Always a mistake, in my experience, if you're not sure. So we don't say, "Hey, Joe, remember, don't ad lib." Or "Joe, have your first paragraph in your head, if you're doing a presentation, not a speech."
Or, "Joe, you're running this meeting. It's your subordinates. Be the last person to talk. Joe, don't forget to ask someone who's being quiet what they think. Remember to frame your questions about who's got a different point of view? As opposed to, does everyone agree?"
And you can asymptotically approach a human coach because people make the same mistakes and have the same strengths, and they manifest themselves. So caught reminding them of where they tend to make the mistake, encouraging them to bring the best of themselves, and making that visible, is incredibly helpful. Just like your folks told you before final exams, "Read the questions, watch your time. Make sure you check the back of the page if there's a question on the back."