Our research is on how AI automation is dismantling the entry-level career path and what universities need to do in response. Give us a picture of what you have going on right now that ties to that question.

I continue to be interested in writing about it. I started that Substack column to write about the issues related to job loss and the impact on employment, and that’s mostly what I’ve written about in it..

The problem has gotten much more substantial with the rise of generative AI and now agents, and companies really need to start doing something about it. I’m giving a talk tomorrow to the faculty senate about what Babson should be doing about AI. All universities need to address the issue, and I occasionally find that some are starting to take steps. I read this morning about a new president of Brandeis who is trying to issue two certifications when you graduate. You get a regular degree, and then you also get a certificate saying what kinds of real-world experience you’ve accumulated at Brandeis. I thought it was quite innovative. But in general, not nearly enough is happening in response to this problem. You have islands, pockets trying something.

Let’s say you’re talking to a dean at another university. What would your advice be? What would their first hard step need to be?

One important step is to engage with the naysayers about AI and persuade them that it’s not going away, that it’s going to be a permanent aspect of life. It’s educational malpractice not to train students in the effective use of it and make them more effective in the workplace with it. But you can’t just ignore the people who would like for it to go away. You have to really engage with them and at least get some degree of consensus around the university about what it’s going to do with AI.

Your essay, “AI and the Entry-Level Problem.” I found myself not fully getting an answer to a question that you posed in the article, which is what the middle gets built from. The middle gets built from the bottom, right? How does the middle get built if companies stop hiring at the bottom? Could you give us your thoughts on that?

I’m not satisfied with the answers yet either. There are various hypotheses about how that issue might be addressed. One is to eliminate or shrink the middle. Supposedly we’re not going to need as many middle managers because of AI. I don’t think there’s any evidence for that assertion, but some people would deny the problem in that regard.

I’ve been looking since 2005 for good answers to that problem, and I haven’t yet found any. There is still some doubt about how much of an entry-level problem AI is causing at the moment. Some companies have now said they’re going to revive their entry-level hiring because they think those workers are necessary in the long run. So I don’t think economists have yet fully established that AI is causing an entry-level problem. It’s likely that it will, if it’s not already. We should start addressing the issue now, but I just don’t think we know for sure.

My friend Erik Brynjolfsson at Stanford wrote an article about “Canaries in the Coal Mine.” It looks only at two areas, programmers and customer service people, and finds, using some employment and hiring data, that there is less hiring in those than at higher levels of experience. But there are other jobs, digital creators for example, where entry-level hiring is still quite strong. So I’m not quite sure we have enough of an answer. I’m not an economist, but I know some economists criticize that article for not really having enough evidence for that strong of a claim.

PwC’s AI 2026 predictions argue for an hourglass instead of a diamond, at least for knowledge work. Wide at the top, narrow in the middle, and a wide base of AI-native generalists who can oversee agents. They say the diamond may apply more to frontline task work, where agents directly replace entry-level workers. Where do you stand on which shape applies? Is the diamond the description of the average, or of one type of work?

I was certainly not advocating for the diamond. I was referring to a conversation I had with a technology executive at an insurance company who said they were moving to the diamond, but, like many others, they had no idea how they were going to fill out the middle if they didn’t have very many entry-level workers.

On the PwC idea, we’re in the very early days of agent supervision and orchestration. My guess is that it would be very useful to have some business experience in a particular domain before you’re checking on the work of agents and telling them what they’re doing well and what they’re not. That sounds like more of a middle-level job to me than an entry-level job.

There are certainly some things that entry-level workers could do relative to agents. They could run the output through other agents to see if it makes sense, or check whether there are obvious hallucinations along the way. But some people have described this phenomenon of taste, as something that humans need to develop, to be able to tell whether the output of an AI system is really good or not. And the development of taste requires experience, generally.

We think of taste as something involving food or art. But you could also look at a financial analysis and ask, is it helpful to my organization, and would people take the right actions on the basis of it? That might require a taste in finance, although we don’t usually use those terms in combination.

In the entry-level essay, you say that students need to develop the skills of applying AI fundamentals coupled with deeper domain understanding, so that they can perform like someone with three to five years of experience on day one. In the schools essay, you said that schools aren’t doing much yet other than trying, unsuccessfully, to ban AI. You require AI usage from your undergrads and master’s students. What does three to five years on day one actually look like in your classroom, and what changes about how you grade?

I mostly teach master’s-level students. I have taught undergraduates fairly recently. They tend to be better with technology than the master’s-level students are.

In order to make that happen, we would have to have broad changes in how we educate our students. It would require a much higher level of internships and cooperative kinds of arrangements. Everyone now is quite jealous of the school down the street from Babson, Northeastern, where they really institutionalized the co-op program for almost all of their students.

There aren’t many things I agree with in the current US federal administration, but they are trying to create greater emphasis on internships. The US has always been quite weak in that regard compared to a number of other countries. Germany, for example, has been great at it relative to the US. In Germany it was really more for manufacturing and skilled trades, but internships should be for almost all kinds of jobs. That would give you a fair amount of experience early on.

Someone suggested to me, and I haven’t actually seen this in practice, that you could develop simulations that would give students greater experience and expertise at solving particular business issues, in the same way we have flight simulators for airplanes. But I haven’t heard of anybody actually doing that for business purposes.

It seems like AI is already ahead of us in some senses. We’re not sure what to do next, with so many institutional issues. Do you see the risk of AI developing at a speed where we’re basically trying to understand a million-year-old civilization, or a ten-million-year-old one, while we’re still on horses from a hundred years ago, and just completely losing any understanding and touch?

I don’t know the exact shape it will take, but AI is advancing at a rate that is already beyond the capabilities of most humans to keep track of. Now that AI companies are using AI to generate code and AI models, that’s drastically accelerating the pace. Anthropic has said that something like three quarters of their code is now generated by AI, and they’re producing new versions faster and faster. And some of them, like the latest one, they realize, have real potential for wreaking havoc, at least from a cybersecurity perspective.

So I do think we really need to slow it down. How that happens, I’m not sure. In the US, it’s unlikely there will be any regulation that slows it down any time soon. But there are a couple of other possibilities. One is consumer reactions to AI, which are increasingly negative and might yield a pretty strong backlash.

Economically, two interesting things could happen that would slow things down a bit. One would be if all of these AI companies stop getting so much money, because people realize the value is not really being achieved. Then they wouldn’t have as much money and energy to pour into developing new models.

The other possibility: I heard a podcast recently from a couple of MIT economists, one of whom I know fairly well. The other was a Nobel Prize winner. They were pointing out that if AI succeeds in putting a lot of people out of work, that’s going to create one of the worst economic climates we’ve ever seen. People have to be paid in order to buy everything, including AI, so that could really slow things down dramatically.

So it’s possible that we could have greater prosperity from AI, that people would have more fulfilling work and find new jobs. But it seems like the chances of that all working out well are increasingly small to me.

I read a study from Cambridge just a few days ago where they tested whether people’s work might be replaced by AI automation. For the people who did run that risk, they told them about it and then asked, what are you going to do about it? They said they were going to unionize, ask the government for redistribution, and retrain and learn new skills. Then the researchers came back a few years later and asked, what have you done? None of them had retrained at all. Everyone keeps telling us that if AI takes a job, people will just retrain. That does not seem to be the case right now.

The economists pointed out that in the Industrial Revolution, you didn’t have much of that. Retraining is not something humans are very good at. You can have generational change. People say there are no jobs in programming anymore, so they stop studying programming and move on to other things. But if you’re a programmer and you’ve never done anything else, it’s unlikely you’re going to become a welder or a data scientist or a plumber, or anything that requires very different skills and career orientations.

So I think that’s correct. What it means is that you have a generation or two that really suffers greatly, as they did in the UK, for example. I always say that if you want to see the impact of technology on people, read some Dickens novels. It’s a horrible life for many people for a while. Over long periods it’ll work out well, but in the short run a lot of people will suffer.

Since you’ve spent years in the heart of what large companies are actually doing, I was thinking about the hiring side of things in pattern terms. If a company decides AI is going to do an entry-level job now, do you have any view into how that actually looks? How do they make that decision? How do they see that kind of choice?

I think they’re trying to be very careful about it. A few years ago, there was a senior KPMG person in Europe who said that they will need far fewer entry-level auditors, and the rest of the profession was shocked that he would say something like that. These companies succeed by selling the hours of people, and historically, partners make money by selling the hours of junior, entry-level people for the most part. Accounting is a very reliable place for people with accounting backgrounds to get their first jobs. So they’re very careful about announcing, “we’re not going to need nearly as many people in the future.” But I’m pretty sure they’re all thinking about it. And entry-level accounting jobs in particular, I suspect, will be one of the earlier ones to be reduced.

Final question from my side. If you had to bet on what a credible business degree certifies in 2035, or what replaces it entirely, what would your bet be?

I suspect it will be disaggregated a bit into the particular types of skills a person has been certified in. The more generic degrees, like a Bachelor of Science in Business, or even MBAs, are already starting to give way to more specialized, skill-based degrees. Some universities have been quite progressive in this regard, like Southern New Hampshire and Western Governors. I think we’ll see more of that going forward.

Certification of AI capabilities will be important too. Some schools are already doing it, and we need to. Certification of real-world experience from internships and co-ops will be quite important. So I think it will be a disaggregation of a lot of the degree programs we’ve seen in the past.

*This interview has been lightly edited for clarity and readability. The interviewee reviewed and approved the transcript before publication.