Who Will Actually Thrive in the Hybrid A.I.-Human Work Force
A panel of experts explains how job seekers should prepare for the future of work.
This May, a new genre of viral video emerged: clips from several college commencements, at which the new grads heartily booed speakers who talked about A.I. And it’s hard to blame them, given recent headlines about how the technology is poised to disrupt the labor market, particularly for entry-level workers like themselves. Many of these young people grew up hearing that “learning to code” was the ticket to a six-figure salary, only to graduate into a job market in which human computer programmers look like an endangered species. While it’s not at all clear whether A.I. has actually reduced the total labor force yet, individual companies have announced large layoffs linked to plans to use A.I. to automate jobs.
All of which raises a vexing question: If you’re a college student today — or, arguably, any worker confronting this uncertain landscape — how should you prepare for the A.I. future of work? There’s no easy answer, but we assembled a panel of four experts who are closely following the impact of A.I. on work to try to help shed light on this question.
The Participants
Daron Acemoglu, economist at M.I.T. and a Nobel laureate
Dean Ball, formerly an adviser on A.I. and emerging technology for the Trump administration, now a senior fellow at the Foundation for American Innovation
Ethan Mollick, professor at the Wharton School of the University of Pennsylvania and the author of “Co-Intelligence” and the forthcoming “Co-Existence”
Clara Shih, former top A.I. executive at Salesforce and Meta, co-founder of New Work Foundation, a nonprofit that aims to help entry-level workers navigate A.I.
Bill Wasik, moderator, editor of the Science desk at The New York Times and formerly the editorial director of The Times Magazine
I.
BILL WASIK: The conversation about A.I. and the future of work is a frustrating one. It’s not just that everyone’s predictions are all over the map — Elon Musk saying, “There will come a point when no job is needed” because “the A.I. will be able to do everything,” or the Chinese A.I. executive Kai-Fu Lee predicting in 2017 that A.I. would displace 50 percent of the work force in the next 10 years — but that they’re rarely coupled with any insight into how or why A.I. might eliminate jobs. What will a hybrid A.I.-human work force look like? How will our own jobs change as a result? Those, to me, are the much more interesting questions.
ETHAN MOLLICK: I’ve had this fun exercise in which I’ve asked the A.I.s themselves to give me a scenario of future work in an A.I. era. It will usually start with something like: “Marcus Chen” — that’s one of the chatbots’ favorite names for software developers — “goes into the office, where he reads reports from his A.I. agents about what work has been done, and then uses his judgment to assign them new tasks.”
But then I’ll ask the A.I., “Wait — why is he going into the office if the A.I. is doing the work?” So then the chatbot will say, “No, you’re right,” and start over: “Marcus Chen wakes up at his beach house and checks in on his A.I. agents.” And then I’ll ask: “Why is he even checking in if the agents are doing all the work?” At which point the chatbot will apologize again and say: “Marcus Chen sits at the beach. … ”
A lot of this comes down to the question of which kind of world we’re in. Is it one where A.I. is a fairly normal-ish technology and change comes more slowly, or is it a world where A.I. actually becomes super capable?
Computer programming is now offering a really good view of this kind of thing. It used to be that being a coder was writing good code on a regular basis. Now suddenly in the course of a few months, it becomes about managing engineering tasks. So I think you’re going to see other shifts in what people’s jobs are expected to be.
CLARA SHIH: I want to build on the Marcus Chen metaphor. Let’s assume we get to that last scenario: He’s on the beach, he’s having a great life. But what about the other workers? It is now faster, more efficient to have fewer cooks in the kitchen.
DARON ACEMOGLU: Right. How many Marcus Chens can the American economy employ? It’s just not realistic to think that 100 million people will work like Marcus Chen.
DEAN BALL: I’m very bullish about A.I., but I have a somewhat more moderated view on the practical effects. And the fundamental reason for that is that the professions we’re talking about are highly automated professions, right? Sending an email is an incredibly automated process compared to what we used to have to do to communicate written text with one another. Coding is intrinsically highly automated. If you want to talk about mining or agriculture — those are incredibly mechanized already. So the fact that we can take the driver out of the mining truck is a very small portion of the vast majority of the cost of mining.
The transformation that’s coming is going to take place in the world as it is familiar to us today, and every single day will feel familiar. And there’ll be tiny, tiny changes along the margin. There’ll be tiny bits of automation along the margins. And 10, 15, 20 years later, we’ll look back and we’ll say, My god, everything is different. But you’ll never notice it happening. That’s the way it always goes.
ACEMOGLU: There are a lot of things that A.I. models currently and in the near future will not be able to do. The current view is that somehow agents are going to do a lot of the work and we just need to supervise them. I find that very unrealistic. But if it was realistic, it would be a horrible thing.
SHIH: Agree and disagree with you. I’m building a tech start-up today. So this is not theoretical. I started about two months ago. I think back to our first 30 days when we had zero employees. Prior to the A.I.-agent world, there’s no way that we would have been able to incorporate, file all of our I.R.S., State of California designations and figure out our privacy policy in a matter of days with so few people. In the past, we would have needed to hire dozens of people and engage with multiple external law firms, marketing and design agencies, etc., working over many months. So A.I. displacement is realistic and real, and it’s both horrible and wonderful.
MOLLICK: The supercharging of entrepreneurship is a very big deal. You have very early work with GPT-4 showing that it helped successful Kenyan entrepreneurs get better stuff done because they got better advice.
ACEMOGLU: Though one finding in economics is that if a lot of people can enter into an occupation or an economic activity, you get excess entry that’s very, very costly and very distortionary. So I’m worried about exactly that happening in entrepreneurship.
MOLLICK: Meanwhile, if you are, you know, Coca-Cola or Walmart, the nature of change is a little bit different. And there’s no obvious plan for: How do I implement this? How do I reorganize my company around A.I. agents? There’s physical labor and there’s an organizational structure that’s been built up. And everything involves meetings and conversations. You’re also not going to replace Walmart overnight. So I think we will see a boom in start-ups. But I think many things are slowed down by this. And that gives us time to react.
II.
WASIK: I would love to think about the Marcus Chens, so to speak, from industries other than the tech industry — the question of which types of work are most threatened. Clara, you’re starting a nonprofit to help entry-level workers navigate this transition. I’m curious to hear about what industries are front of mind for you when you think about this as a potential crisis moment for entry-level workers across the board.
SHIH: I think if you just look at companies like Amazon over the last 10 years, it’s a great example of how this plays out. There are more than a million workers at Amazon, and a fraction of them are part of the elite group. They’re coders, they’re supervising systems, they are working at HQ. And everyone else works in the warehouse and as delivery drivers. And they are managed to the minute: what their tasks should be, how long they should be spending getting from Stop A to Stop B. And over time, I expect more and more of their tasks, including the delivery and the driving, to be taken over by A.I. I see that playing out in multiple industries.
Beyond that, let’s add to the Marcus Chen portfolio. Let’s talk about Stacey Smith, a hypothetical health insurance claims adjuster. She and millions of people like her are employed in the U.S. in insurance and banking because of regulatory reasons that those jobs could not be off-shored. But now a lot of those approvals, whether it’s mortgage underwriting or claims adjusting, are easily done by A.I. agents — more consistently, more easily able to detect fraud, much more cheaply. What’s going to happen to the Stacey Smiths who are making a decent wage in places like Kentucky and Mississippi?
What about Bob Johnson? Let’s talk about him. He is a long-distance truck driver. He is one of three and a half million truck drivers in the U.S. who again make a decent wage. They live in the South, they live in Texas, Louisiana, Mississippi. They are the breadwinners for their family. They’re the pillars of their community. What happens when the Waymo technology diffuses and we get past regulatory barriers to the Bob Johnsons?
So maybe Marcus has a great outcome. What about Stacey? What about Bob?
BALL: I think if we want to get into specific industries or roles, the obvious ones are consulting, marketing, customer service, entry-level legal work, administrative work. All that is definitely real.
And then there’s this fuzzy layer of things in the physical world. I’m kind of doubtful that a humanoid robot’s going to be making you a cocktail at a bar, even if it could, right? People don’t want that. And that’s a very important thing for thinking about the future of work. What are people’s preferences going to be? Similarly, a lot of knowledge work, in the end, especially as you move up the ranks, comes down to persuading people of things. I find myself skeptical that the process of internal politicking within a firm or another organization is just going to be automated away by A.I.
MOLLICK: The story inside big firms will be complicated, I think. At Procter & Gamble, we did an experiment with 776 of their employees. They either were technical or business people, and they either worked individually or on teams of two. The finding at the time was individuals using A.I. performed as well as teams not using A.I. To me, the really interesting part was that it also blurred lines between roles. Business people used to have business ideas, tech people came up with technical ideas. But add A.I., and everyone comes up with ideas in between each other. And that’s happening everywhere. When I talk to people in coding, especially in industries that have some quasi-creative element, like the gaming industry, suddenly the people who are designers can code, suddenly the coders can do design work, the artists can start writing.
ACEMOGLU: Let me turn the question around and talk about three industries that I think are absolutely critical for future productivity: manufacturing, health care and education. All of them show the potential for A.I., but also the huge bottlenecks. In manufacturing, the U.S. is behind China in applying A.I. to manufacturing, and every small step of additional introduction of automation or A.I. into the manufacturing process requires a huge amount of engineering services. It’s very laborious. Then there’s education, where the effects of A.I. so far have been a huge disaster: There’s a nice large-scale study coming out from China, for example, which shows horrible, horrible outcomes from students who use A.I.
With health care, the promise has been in digitization, for example, with electronic health records and more addition of software into the process, but that so far seems to have led to a worsening of productivity. So the potential for A.I. to do cost saving is there, but I don’t think anybody has a good road map of how to do that. Chatbots are potentially useful in health care, but my fear is that we’re going to repeat exactly what we did in education, unleashing chatbots in a way that really leads to horrible outcomes.
BALL: But Daron, you mentioned chatbots — if all we had were chatbots, I would probably share your pessimism. But the remarkable thing about the new coding agents is that they’re essentially computer programs that can use computers, right?
MOLLICK: Yes, I don’t think we’re talking about chatbots anymore. The world has changed in the last five years. And we already know from data that A.I. is better at medical diagnosis than doctors under many circumstances. And controlled experiments show patients prefer talking to A.I. over doctors because it has higher empathy — perceived empathy, I should say.
Similarly with A.I. and education — research has found that unstructured A.I. use hurts learning, but A.I. tutors can have large-scale positive impacts. So I actually am incredibly bullish about A.I. and education once we figure out how to actually integrate A.I. tutors in the classrooms.
III.
WASIK: I’m curious about new opportunities that will exist in this A.I. world. What do you tell a young person about interesting potential skills or careers to think about for the future?
BALL: In many ways, I think curious generalists will do quite well in the future. That’s not to say you don’t want to cultivate narrow niches of expertise, but that also being able to step back and think broadly will be rewarded.
I also think there are obvious high-demand areas in the physical world — in particular things like plumbing and electricians and HVAC technicians.
MOLLICK: Though I also don’t think we can predict well enough to say, Go be a plumber, because then the Plumbot10000 could come out in a couple of weeks. It’s very hard to anticipate the future.
ACEMOGLU: The job of electricians, HVAC technicians, all of these are going to become more complex. We already have a huge shortage of these trades. Again, A.I. can be massively helpful. A novice electrician with the right A.I. tool could be 10 times as productive as they are at the moment. Using A.I. for training electricians with new equipment could be hugely useful. But that’s not where the center of gravity of the investments in A.I. are going.
WASIK: What advice would you give a smart young person in terms of how to think about where there might be opportunities we don’t even know about yet?
SHIH: I get to spend all day every day with 25-year-olds now, and what I’ve found is a lot of the advice that’s out there and a lot of the A.I. courses out there are too general. We just launched a content platform that goes role by role across the 50 most common entry-level white-collar jobs: marketing, software engineering, accounting, finance, etc. And we interview hiring managers to understand how they’re deploying A.I. in their organizations and how that’s changed the specific skills that they’re looking to hire.
The recurring theme is that it’s a tale of two cities for job seekers. Those who know how A.I. works, specifically A.I. agents, can get their dream job, whether it’s in marketing or software, accounting, finance, you name it. Those who don’t have those skills, those entry-level jobs are disappearing.
WASIK: But how helpful will that A.I. training really be in the long term? As the models get better, it seems like different instincts and different skills are required to get what you want out of them, while the hacks you needed with earlier models just completely go away. Almost their whole magic trick is that they make everything so seamless as they get better.
SHIH: A.I. skilling is a moving target for sure: You’re on the train, but you know that there’s no destination. You know what you’re signing up for is constantly learning and evolving your technique as the model is changing. It means signing up for continually testing different models as they come out and starting to recognize, OK, I use Gemini for this task, but Claude is much better for this. That’s part of the evolving skill set that I think is going to become necessary.
MOLLICK: What I really worry about with junior workers is that field experience is often crucial to evaluating work you didn’t create yourself, whether it comes from humans or A.I. Bill, you can probably glance at an article for The Times and tell whether it’s good or bad. Daron or I can look at an academic paper instantly and say whether it’s going to be worth an hour of our time to read or not. Dean can scan a piece of A.I. legislation and spot the loophole. Clara can, I’m sure, look at any bit of code and judge whether it was written by an idiot or an expert. If you have no experience, you can’t do those things.
And that’s what it takes to manage A.I. agents, too. This isn’t just an issue with junior workers: There was a survey a while ago that only a third of people want leadership positions. Most people are very happy with what their job is and they just want to do it. And I think they’re going to be having some trouble adjusting.
SHIH: Being a manager has already gotten better thanks to A.I.
ACEMOGLU: I actually think most managers feel overburdened, behind a curve, don’t know what is expected of them. I think their stress level is much higher. Perhaps in 10 years’ time, it will be better, but we need to standardize those tools and standardize the expectations, retrain the managers, perhaps. So no, I don’t think at the moment being a manager is a better occupation at all, actually.
SHIH: I believe every young person needs to work on a side project while they’re in college, if not high school, that they own end to end so that they’re evolving with these models. And if they do that, I believe that they’ll have a lot of the practical experience they need to get hired.
MOLLICK: But as somebody who teaches entrepreneurship, not everyone does this, right? I’ve given people these side projects — like, that’s what my job has been for over a decade, giving young people side projects. And it’s not for everyone. There are all kinds of differences, including in background or socioeconomic status, in how people take advantage of those opportunities. It’s a hard thing to just say, Do the side project and make it work.
SHIH: It is very hard and we’re not going to have a 100 percent hit rate, but it is very difficult to then really understand and stay in control of A.I. if you don’t learn something end to end. Really taking it all the way through to understand the outer limits of what A.I. can do and understand how models evolve — that is the core skill set.
ACEMOGLU: I think what Clara described is the reality. Right now, you have to spend a lot of time learning different models, their capabilities, their shortcomings, and then three months later you have to experiment with lots of different models again in order to just stay where you are. That is absolutely not productive, that’s very dystopian. Clara, perhaps you were describing that as an OK future. To me, that’s a horrible future.
SHIH: I said earlier that it’s wonderful and horrible. How do we make it less horrible and more wonderful?
IV.
WASIK: So this brings us around to the last question: What do we do about all of this? Dean, I know you were involved in helping to draft the White House’s A.I. action plan. I’m just curious to know what you think the governmental and societal response should be.
BALL: The action plan conveniently happens to state my beliefs quite well, which is that the first thing we need to do is measure this problem much better than we currently do. We need better empirical economic data. You can’t create policy remedies for a problem you don’t understand.
But I really don’t want to create inflexibility in the labor market. My favorite line in the White House action plan describes A.I. as a combination of the Industrial Revolution, the Information Revolution and the Renaissance. It has the potential to transform the way we make and produce things; it will change the way information flows through organizations and the world; and it will enable artistic and mathematical and scientific breakthroughs.
Now the pope seems to want us to have a committee of all the A.I. stakeholders, all the impacted communities. But we didn’t create a stakeholder committee for the Industrial Revolution. It didn’t unfold by show of hands. It’s unbelievable to me that there are so many people in polite society who think that’s what we need to apply to A.I.
ACEMOGLU: Well, I guess that summarizes the attitude of some people in Silicon Valley. Yeah, there was no committee for the direction of the Industrial Revolution. And what did we get? In the early stages, we got children working to death in coal mines. Conditions in factories became horrible. Wages for many workers fell. Now A.I. is changing things much faster and affecting many, many sectors at the same time.
SHIH: Right, the lesson isn’t that coordination is unnecessary; it’s that skipping it costs us decades of human suffering. Also, A.I. policies that emerge without broad input will not be trusted, as they will be perceived to represent the agenda of big tech, capital or one political faction.
BALL: I don’t know what the unemployment rate will be in 2028, but I guarantee you that 100 percent of it is going to be blamed on A.I. by the American public and by lots of opportunistic politicians. And unfortunately, I worry that the policy remedies they reach for are going to be things that lock in very vast kinds of labor protections. Things that create the kinds of problems Europe has, which is that their firms can’t take risks because taking risks involves doing things that might require you to fire people in five years if the thing you tried didn’t work out.
I worry that we’re going to do similar things here. And a lot of that’s going to happen at the state level. We’ll blab about it in Washington, but the actual action will be in 50 different state houses. And we’ll all pretend like we know what’s going on there, but no one really will. And we will sort of sleepwalk into a very bad political economy.
SHIH: I agree with that. I mean, you look at the reaction to Eric Schmidt’s commencement speech at the University of Arizona, the recent Gallup poll showing that a third of Gen Z Americans describe their feelings toward A.I. as anger. And these are people we want embracing A.I. so that they can help build this economy and they can find work. But they’re rejecting it on a moral basis. That’s why I believe that an intervention is needed now. That’s why we’re building A.I. agents to help connect job seekers with opportunities, and help them learn A.I. skills so that they don’t have to fear it and that they can shape it. But we have to do something.
ACEMOGLU: I am convinced that artificial intelligence is quite different from human intelligence. Humans are not very good at absorbing massive volumes of information or sifting through unstructured data for relevant patterns. A.I. models have, as yet, no genuine creativity and no capacity for trial-and-error learning grounded in interaction with the physical world. When two things are different, the last thing you want is to try to mimic one with the other. It is a fool’s errand to try to get one to do everything that the other one does. They should work together. And those aren’t the kinds of models we’re trying to develop.
MOLLICK: We are going to obviously need to think about how we train new workers. But how do we even assess young workers when they come on board? Because we used to assess them. We had this great technique, which was apprenticeship. It’s worked for 4,000 years. I hire a white-collar worker, and they do grunt work for me and they work really hard, and I get to assess how good they are at the grunt work.
And as a middle manager, I get the advantage of having someone do the work I don’t want to do, and they learn, and I assess them, and everyone gets paid. And that all collapsed, right? You can’t just say, Oh, we should hire junior workers, if we haven’t thought about how to train junior workers. But guess what? Universities are actually quite good at training and assessing. So colleges could start to think about extending or changing professional education to fill in the gaps where A.I. is creating them.
SHIH: I think the biggest takeaway for me is where we started, which is: The future isn’t automatic. I’m neither a pessimist nor an optimist. I’m a conditional optimist, and there are interventions that we still can try, that we should try, because we’re going to have to test many things before we know what will work. And there’s still time, but the window is closing.
I do think that the key thing about A.I. agents is that they all have a goal. And it depends on who deploys it, because whoever deploys it gets to set the goal. Maybe the goals of A.I. so far haven’t been aligned with the goals of regular people, with everyday workers. But that’s a choice we can make.