Are New Graduates Losing Jobs to AI, or Us?

Rethinking higher ed, industry, and the real causes of early career struggles

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Try with our 30-day free trial and Upgrade to the On EdTech+ newsletter.

Over the past week, higher education and the media generally has been abuzz about the recent study by Stanford researchers Brynjolfsson, Chandar, and Chen on the impact of AI on recent college graduates. The study concludes that some occupations are more affected by AI than others and, in those fields, new graduate employment has taken a hit. For example, even controlling for firm effects, there has been a 13% relative employment decline for young workers in the most exposed occupations.

The decline in graduate employment is certainly a problem. But I think it’s a mistake to attribute this entirely to AI, even within a narrow set of occupations, as the Brynjolfsson et al. paper and much of the discussion it has sparked seem to do.

AI is undoubtedly reshaping the world of work. Yet, the evidence suggests to me that the deeper issue lies in how universities are preparing students for the workforce. The mismatch between the education students receive and the preparation employers expect is particularly troubling, especially given current hiring practices. Both sides, higher education and industry, need to change. But higher education should see this as a wake up call to make the education they provide of higher value and better able to help graduates land jobs.

Brynjolfsson, Chandar & Chen’s argument

I wrote about this briefly on the weekend, in my weekly Interesting Reads post that came out on Saturday. But given the attention generated by the Stanford paper, I wanted to write a longer piece about the need to look beyond the obvious. As I wrote on Saturday.

In an elegant study, the researchers used payroll data from millions of ADP records to examine the impact of generative AI exposure. They found that employment for young workers (ages 22–25) in occupations highly exposed to AI has declined by 13%. In contrast, workers in less-exposed occupations, and older workers in general, have seen stable or even growing employment levels.

Crucially, Brynjolfsson et al. show that AI’s effects are visible in employment levels rather than wages, with the strongest impact in occupations where AI is used to automate, not augment, tasks. The results are striking: for example, young software developers, an occupation highly exposed to automation, have experienced steep employment declines since 2022.

Brynjolfsson et al. look closely at a few occupations, including software developers and customer service representatives. It’s fair to say that some roles are particularly vulnerable to automation by AI. Just as desktop computers once eroded demand for secretarial work, today’s increasingly sophisticated AI systems are likely to replace many customer service representatives in call centers.

For the rest of this post, I’ll focus on software developers, both because they’ve been central to the discussion sparked by the Brynjolfsson et al. paper and because they’re an especially revealing case. In some ways, they’re a perfect example; in others, less so. Software engineering is inherently technology-based, and coding has been one of the earliest and most visible use cases for generative AI. But unlike some of the other occupations analyzed in the study, software developers generally hold college degrees, and computer science is a field where universities have a direct and growing role in preparing graduates. That makes it a particularly useful occupation for examining higher education’s responsibilities and shortcomings in this debate.

By July, 2025, employment for software developers aged 22-25 declined by nearly 20% compared to its peak in late 2022.

Chart showing headcount over time by software developers in age groups

Contrast this with Mid, Senior, and Senior + employees where employment has continued to rise.

The paper has received some criticisms that are worth noting.

  • The age scales are not the same and thus not comparable. There are four years worth of Early Career workers, from age 22-25. Mid career workers are measured across nine years (41-49) and Senior workers over at least 15.

  • Some take exception to the way they determine which occupations are heavily impacted and which are not.

  • The decline in early career hiring goes back to 2012, predating the availability of generative AI by some time. Further, using 2022 as the base year may skew the data due to a post-pandemic readjustment in the tech sector.

But in general, the paper has been received as proof that generative AI is indeed destroying the job market for new grads. In doing so it joins a range of other papers that make a similar argument.

AI is not the problem

In making this argument, I think Brynjolfsson et al., and others making similar claims, are overlooking something obvious. Their own data shows that while hiring of entry-level software developers is down, employment for more senior developers remains strong. Additional evidence from The Pragmatic Engineer confirms that demand for senior talent in software engineering is particularly robust.

Senior-heavy recruitment

Somewhat unexpectedly, there are almost as many open senior positions as there are mid-level and entry-level ones. Usually, far fewer senior roles are advertised.

Chart showing open positions for software developers by seniority

This is striking: these workers are older and command higher salaries, yet they are still being hired. If AI were simply eliminating jobs across the board, we would expect declines at all levels. The fact that experienced, costlier employees are thriving suggests that the problem is less about AI displacing all software roles and more about how we prepare new graduates to step into the workforce.

What kind of education?

This suggests to me that there’s a deeper problem in how we prepare graduates for the workforce. The challenge has two dimensions: theory and practice. In a post I came across after writing my original reflections, Azeem Azhar makes an interesting argument.

AI substitutes for book learning, not the skill that comes with practice. Experience and judgement are therefore becoming more valuable.

While there’s truth in this, I don’t think it’s as simple as drawing a sharp line between theory and practice. I also haven’t heard the phrase “book learning” since I stopped hanging out with Mark Twain.

But, more seriously, theory remains essential, not only for building a deeper understanding of processes, but also for creativity and innovation, troubleshooting, problem solving, and, most importantly, learning how to learn. Without it, we risk producing shallowly educated graduates who struggle to adapt, innovate, or move on to the next challenge. That said, I wonder whether the topics taught in higher education sufficiently reflect changing industry realities and whether departments are keeping pace with the needs of the workforce. This lag contributes significantly to the disconnect.

The bigger issue, however, lies with experience. Too often, courses fail to provide students with meaningful, hands-on opportunities. Homework is not the same as workplace practice. Students need to learn in ways that mirror the environments they will encounter on the job, not in sterile, artificial settings designed primarily to prevent cheating. In some ways, these environments create their own form of “artificial intelligence.” At one institution I know, computer science faculty even asked the CIO to block access to GitHub, ignoring the fact that professional software developers rely on repositories like it daily. If we want graduates who can thrive, we need to embed real-world applications into their education: projects, internships, co-ops, and other experiential learning opportunities. Only then can we help students climb out of the early-career hole so many fall into after graduation.

Taken together, these two issues point to a larger imperative: higher education must help students move up the value chain of work and knowledge before they graduate. That way, they can enter the workforce already equipped to handle the kinds of tasks more seasoned employees perform. In software development, for example, this would mean preparing students not just for basic coding, but for work in areas such as system architecture, code verification, and validation, roles that demand a deeper blend of theory and practice.

Doesn’t industry bear any responsibility?

In response to my original post, I received feedback that I was unfairly placing too much of the blame on higher education, and that the sector should not bear full responsibility for the problem.

That’s a fair point. There is plenty of higher ed bashing these days, and like most complex issues, this one has two sides. Industry also carries significant responsibility.

  • Declining investment in training - Surveys consistently show that employers are spending less on training their workforce. Rather than building talent internally, many companies expect to hire employees who are already “job-ready.”

  • Shifting expectations of higher ed - Increasingly, industry seems to view the role of colleges and universities not as providing a solid base of knowledge, but as, in Clay Shirky’s words, doing their onboarding for them.

  • Conservative hiring practices - Like higher ed itself, many employers shy away from hiring people with potential and developing them into roles. For instance, in CIO searches at universities, committees often insist on hiring a sitting CIO rather than considering someone who is clearly ready for the next step.

  • Bias in HR systems - Relatedly, with AI-driven hiring tools, it has even been suggested that resume-screening algorithms are being trained on the profiles of people already in the role. This risks creating bias against entry-level applicants and reinforcing a cycle of “experience required” that locks out new graduates.

These factors highlight the actions industry must take if it is to meet its own employment needs. Higher ed certainly has work to do, but the burden cannot fall solely on universities. Industry has to step up as well, and we should continue to shine a light on these shortcomings and press for change.

Parting thoughts

But this does not absolve higher education of the responsibility to address how it prepares graduates. Increasingly, institutions will be judged, and even funded, based on the outcomes of their graduates.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.