By Lee Flanagan
Morgan Housel told The Tetr Podcast’s Pratham Mittal that he would “move mountains” to send his own children to a good college. In the same conversation, he flagged something no tuition check fixes: artificial intelligence is eliminating the junior jobs where graduates used to learn how to work. That framing lands hard with parents. It hits them where they live: the anxiety over what a degree actually buys. TA leaders should care about a narrower, harder problem tucked inside his comment. The evidence a resume used to carry, the track record of work already done, is running dry, and almost nobody is talking about what replaces it inside the interview room.
What the Harvard paper actually found
A working paper from Seyed Mahdi Hosseini Maasoum and Guy Lichtinger tracked 62 million workers across 285,000 U.S. firms.
It found that companies adopting generative AI cut junior employment by about 9% within six quarters. This is based on the updated May 2026 version of the paper, while senior employment at the same companies held steady. The authors call the pattern “seniority-biased technological change.” The name fits: the cut landed on the bottom rung, not across the workforce.
The freeze shows up everywhere recruiters look
SignalFire’s 2025 State of Tech Talent Report tracked hiring at the fifteen largest tech companies in the industry.
Entry-level hiring there dropped more than 50% from pre-pandemic levels through 2024.
New graduates fell from about 15% of all new hires to roughly 7% over that stretch.
A June 2026 SignalFire update pushed the decline further.
Across SignalFire’s 12 Tech Majors, entry-level hiring is now down close to 65%.
That decline is measured against 2019 levels.
Early-stage startups fared worse, with entry-level hiring falling around 75% over the same period.
TheStreet ties this to a Harvard-scale dataset and an industry tracker built specifically to watch tech hiring, and both point the same direction.
NACE raised its hiring forecast to 5.6%, but landing the job got harder
The National Association of Colleges and Employers initially projected a 1.6% increase in hiring for the Class of 2026.
NACE’s April 2026 Spring Update revised that number up to 5.6%.
More than a third of employers now report plans to add hires. On paper, that reads like relief for this year’s graduates.
Bari Williams, a startup advisor and former senior legal counsel at Facebook, describes what is actually happening inside that number: “Employers are increasingly unwilling to gamble on unproven candidates. Instead, they favor applicants who can point to concrete work they’ve already done and explain how it translates to the role.” The lesson for TA is straightforward: more openings does not mean hiring gets easier for a graduate with no finished work to point to.
The judgment AI cannot inherit, and the job IBM rebuilt around it
Jossie Haines, an executive coach and former engineering leader at Apple, told Forbes that AI cannot automate human judgment: “AI could potentially figure out how to process copyright tickets, but it cannot figure out why the product team keeps building features that raise copyright concerns, or how to address that from a process perspective.”
That kind of systems-level thinking used to develop through proximity to real decisions: catching an error before it spread, absorbing how accountability actually moved through an organization. Debugging, testing, routine coding, data entry and basic financial analysis were the training ground for exactly that. When AI handles that work, the foundation shrinks along with the roles that used to sit on top of it.
IBM’s own move shows the roles are not disappearing everywhere, only the version that assumed a graduate could learn on the job before proving anything. Nickle LaMoreaux, IBM’s Chief Human Resources Officer, told the Charter AI Summit that IBM plans to triple its U.S. entry-level hiring in 2026 after having “rewritten every job” to shift junior roles toward analysis, problem-solving and AI oversight.
Natasha Pillay-Bemath, IBM’s VP of Global Talent Acquisition and Executive Search, describes the same shift in plainer terms: junior roles are moving away from task completion and toward analysis, problem-solving and responsible AI use. IBM has not eliminated the junior job. In our read, IBM redesigned the role around the judgment AI still cannot fake, and is hiring more people into it, not fewer.
The interview now has to prove what a resume used to prove
Here is the shift TA teams building 2026 and 2027 graduate programs are actually up against. The old bargain assumed college taught a student to think and a junior job taught them to work. By the time that graduate interviewed for the next role, they carried real evidence: finished projects and judgment calls that had held up under scrutiny. The Harvard paper and the SignalFire numbers describe that evidence supply shrinking, because the jobs that used to generate it are frozen rather than growing.
A resume used to be proof of work already done. Now it risks becoming proof of work AI already did.
In our work with hiring teams, interview processes that still lean on prior output as the main signal for entry-level candidates are struggling hardest to fill 2026 graduate cohorts. There is simply far less prior output left to lean on.
If your process still treats a portfolio of finished work as the deciding factor for a junior hire, ask what happens once that portfolio is gone. The processes holding up test reasoning directly: how a candidate works through an unfamiliar problem in the room, how they respond when a first answer turns out wrong, and how they explain a decision rather than simply state one. That is the one thing a candidate cannot show up having already generated.
Housel is right that no acceptance letter fixes this. He is describing a parent’s problem. The version TA leaders inherit is sharper: the graduate sitting across the table in 2026 carries less finished work behind them than the one who sat there in 2019, through no fault of their own. The interview is now the only place left to find out what they can actually do.
Original reporting: thestreet.com.
Frequently asked questions
What does “seniority-biased technological change” mean in the Harvard working paper?
It is the term Seyed Mahdi Hosseini Maasoum and Guy Lichtinger use for their finding that companies adopting generative AI cut junior employment by about 9% within six quarters while senior employment at those firms held steady. The phrase describes a pattern where AI adoption hits entry-level jobs specifically, not headcount broadly.
Does NACE’s revised 5.6% hiring projection mean entry-level hiring is recovering?
Not on its own. NACE raised its Class of 2026 hiring projection from 1.6% to 5.6%, but Bari Williams, a startup advisor and former Facebook senior legal counsel, says employers remain unwilling to gamble on unproven candidates and favor those who can show finished work. More openings does not make hiring easier for a graduate without that evidence.
Does IBM’s expanded entry-level hiring contradict the broader junior-job freeze?
No. IBM plans to triple its U.S. entry-level hiring in 2026 after having “rewritten every job” to shift junior roles toward analysis, problem-solving and AI oversight, according to CHRO Nickle LaMoreaux and VP Natasha Pillay-Bemath. It shows one company redesigning the junior role rather than eliminating it. That sits alongside the freeze described by the Harvard paper and SignalFire data, not against it.
Why can’t an AI-assisted portfolio replace the missing work history?
Because the interview is meant to test judgment a candidate could not have generated with a tool, and a portfolio built with AI assistance does not prove that on its own. A resume built this way risks showing what a model produced rather than what the candidate can do, which is why reasoning tested live in the room now carries more weight than finished output.