Jon Krohn: 00:00 This is episode number 994 on whether AI is putting new computer science grads out of work. Welcome back to the SuperDataScience Podcast. I’m your host, Jon Krohn. Today’s topic is whether AI is already putting recent graduates out of work. Earlier this month, a White House advisor named Kevin Hassett told reporters there’s no sign in the data that AI is costing anyone their job right now. That’s a confident claim, but if you asked any member of America’s class of 2026 walking off the graduation stage around now, you’d probably get a different answer. At one commencement ceremony in Florida recently, a speaker was actually booed for mentioning artificial intelligence. And the underlying numbers help explain the booing. So let me walk you through what the data show, what economists disagree about, and then we’ll spend the back half of this episode on what graduates and even senior professionals can do about it.
00:59 First, the headline figures. According to the New York Federal Reserve, the unemployment rate for recent computer science grads age 22 to 27 currently sits at 7%. For computer engineering grads, it’s nearly 8%. To put that in perspective, those rates are roughly comparable to graduates in anthropology, fine arts, and the performing arts, historically the punching bags of the is your degree worth it debates. The overall unemployment rate for recent grads is around 6% and the figure for the entire US workforce is hovering around 4%. So computer science grads with that seven to 8% unemployment rate who were sold a golden ticket degree, well, supposedly a golden ticket degree for the better part of two decades are now noticeably worse off than the average new alum. You can see the effect on the supply side of education as well. Undergraduate enrollment in computer science in the US fell 11% in 2025.
01:58 Enrollment in computer programming, the more applied coding focus track fell a stunning 26%. Whoa, year over year. That’s crazy. 18 year olds evidently are voting with their feet and the feet are walking away from the keyboard. On the demand side, the picture is just as stark. Job postings on Handshake a platform American university students use to find work are down roughly 50% from their 2022 peak. Data from Revellio labs suggest entry level postings in software development and data analytics have fallen by as much as 67% and less than a fifth of the class of 2026 thinks this is a good time to find a good job, the lowest reading in over a decade. So is AI the culprit? Here’s where it gets interesting because the economists are divided. In the YES camp, you have a 2025 study from Stanford’s Eric Brindolfson and colleagues who found that employment among 22 to 25 year olds in AI exposed jobs has dropped 13% since 2022 when ChatGPT came out while employment for older workers and less exposed roles has held steady or even ticked up in that time.
03:10 The Dallas Fed independently replicated this pattern earlier this year and crucially, they found the decline isn’t being driven by layoffs. It’s being driven by young people simply not being hired in the first place. That is, they never even make it onto the bottom rung of the career ladder. So that’s the yes camp. In the not so fast camp, two economists at Google, I’m probably butchering their names here, but Zana Ischenko and Fabian Kurto Millet. They published a paper earlier this year showing that job posting declines in AI exposed occupations have been just as steep for senior workers as for juniors and that the trend predates the launch of ChatGPT in 2022.
03:50 Pretty much the opposite findings of what I was saying just a minute ago. And in addition to that, a Federal Reserve study last month analyzed data from over a million firms and found what the authors called precisely estimated null effects that is no statistically detectable link between AI adoption and reduced hiring. The skeptic’s explanation is that we’ve simply lived through a brutal stretch of high interest rates, the end of the post-pandemic tech boom, and a hangover of overhiring from 2021. AI in this telling is just a convenient scapegoat. Indeed, Stanford’s Eric Roberts has pointed out the striking parallels to the dotcom bust when fears about offshoring kept students out of computer science right before the industry roared back to pre-crash hiring levels by 2004. My own view now, having watched this space for years, is that the truth is somewhere in the middle. Macroeconomic forces explain a lot of the slump, I agree, but the granular evidence, particularly that young AI exposed workers are being uniquely shut out at the entry point, it’s hard to wave that away.
04:50 And part of that is that anecdotally at my consulting firm, Y Carrot, we can now so easily use tools like Cloud Code and OpenAI’s Codex to generate such high quality code so fast for such negligible financial cost that in the foreseeable future, I’m not sure why we would hire any entry level software developer. All right. Well, whichever camp turns out to be right, the practical question from you listeners is the same. What do you actually do about it? Well, let me give you five concrete pieces of advice for those folks who are still hiring entry level folks. So the first one is to stop competing on raw coding output. Large language models are extraordinarily good at producing functional code from a clear specification. The differentiated human skills are now system design, architectural decision making, integration with messy real world data and figuring out what should be built in the first place.
05:47 Someone named Lana Yarosh, who heads up undergraduate computer science studies at the University of Minnesota frames it nicely. They say that graduates today spend less time writing code and more time designing and organizing software systems at a higher level. That’s actually exactly the kind of stuff that anyone and my businesses are doing today. All right. Tip two, pick a domain. The graduates getting hired right now are the ones who pair AI fluency with deep knowledge of healthcare, finance, manufacturing, agricultural, law, biology, whatever. Pick your industry. Generic AI engineer candidates are competing with thousands of relatively identical resumes. In contrast, someone who can say that they’re an AI engineer who has spent two summers working alongside a hospital revenue cycle team is by contrast a very short list. So those are the first two tips. Stop competing on raw coding output and pick a domain. The third is to build a public portfolio.
06:43 Recruiters in 2026 are skeptical of credentials and frankly skeptical of resumes. Indeed, automated applicant tracking systems are filtering most candidates out before any human sees them. And I was using Indeed with a lowercase either, but probably literally indeed with a capitalized applicant tracking system automated stuff, filtering candidates out as well. Now, by building a public portfolio like a handful of substantial GitHub repositories, a keggle competition or two, a clear write-up of a project you built end-to-end, those kinds of things will dramatically outperform a stack of CVs sent into the void. All right. So now onto tip number four, get fluent with agentic tooling. Come on. I don’t just mean prompting ChatGPT. I mean knowing how to wire up rag systems, retrieve for logment generation, how to evaluate, how to evaluate model output, how to orchestrate multiple agents and how to integrate models into real workflows.
07:41 PWC, for example, in their most recent AI jobs barometer found that workers with AI skills earned a 56% wage premium over their peers. That’s the kind of differentiation worth investing in. And there are outstanding resources for this online that are freely available or extremely affordable. I highly recommend checking out the Udemy courses from my longtime friend and colleague, Ed Donner, for example. All right and my fifth and final tip for you is to lean on your network, not the job board. Multiple recent surveys suggest referrals and warm introductions are outperforming mass applications by enormous margins in this market. Going to industry meetups, contributing to open source projects and reaching out directly to people whose work you admire will move the needle far more than another LinkedIn Easy Apply. All right, those are the five tips. Stop competing on raw coding output. Pick a domain, build a public portfolio, get fluent with agentic tooling and lean on your network, not the job board.
08:39 Now all of those tips were primarily intended for recent graduates, but all of the advice I just provided is useful for any technical listener looking to make a career move regardles of your career stage. More senior folks will, of course, also be able to leverage their existing experience and professional networks so that’s a boon for you too. These fast moving times can be anxiety inducing for sure, but especially for anyone trying to enter or navigate a technical career. But this is also perhaps the most leveraged moment in history to learn quickly and build things. So keep learning, keep building and keep showing your work publicly. The market for people who can do that is, if anything, getting bigger, not smaller. All right, that’s it for today’s episode. If you enjoyed it or know someone who might, perhaps someone looking for that first job in AI or software, consider sharing this episode with them.
09:31 Leave a review of the show on your favorite podcasting platform or YouTube. Tag me in a LinkedIn post with your thoughts. Maybe some of the things that I said in this episode are controversial. I don’t know. You tell me. And of course, if you aren’t already, be sure to subscribe to the show. Most importantly, however, we hope you’ll just keep on listening. Until next time, keep on rocking it out there and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.
