AI and Entry-Level Jobs: Talent Pipeline Risk

A few months ago, AI and white-collar work was a forecast. This spring it became a contest over payroll data. For the leaders setting 2026 hiring plans now, the question has changed, and the answer will shape what their organization looks like in the fourth quarter and which leadership bench they hold in five years.

The state of play

The argument about AI and white-collar work has moved from prediction to evidence. It is now built on actual payroll data rather than on a forecast.
The empirical center is a Stanford Digital Economy Lab study by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, which finds early-career workers in the most AI-exposed occupations losing ground while older workers hold steady.
The data has sharpened the question without settling it. Leading economists and management researchers read the same evidence and disagree about cause, durability and the right response.
The decision in front of a chief human resources officer or chief learning officer is no longer “will AI displace entry-level workers,” it is “what does cutting the bottom rung this year do to the leadership pipeline five years out.”
Workforce decisions made this summer are what an organization will be living with in October. Leaders who intend to speak credibly about this in the fourth quarter are forming their positions now.

From forecast to payroll evidence

The prediction that AI would erase half of entry-level white-collar jobs traveled on its vividness alone; no one could yet prove it right or wrong. That phase has ended. On May 26, 2026, MIT Technology Review published “A reality check on the AI jobs hysteria,” built on payroll evidence rather than projection, days after ProMarket examined the same Stanford findings on May 21. What changed is not the volume of the conversation but its basis: the debate now rests on a dataset. And data forecloses the option a forecast leaves open, which is to wait for more before deciding anything.

What the Stanford payroll study found

The empirical center of the debate is one study, and it is worth being precise. Erik Brynjolfsson, who directs the Stanford Digital Economy Lab, with Bharat Chandar and Ruyu Chen, used ADP payroll microdata covering millions of workers to test what is happening in occupations most exposed to generative AI. Their paper, “Canaries in the Coal Mine” (Stanford Digital Economy Lab, August 2025, updated November 2025), documents a roughly 13% relative decline in employment for workers ages 22 to 25 in the most AI-exposed roles, while employment for older workers in those same occupations held steady or grew. The title names the argument: the youngest workers are the early warning.

The study sits at the center because both sides reach for it. Fortune used it on April 29 to frame the finding that AI may not take your job so much as the path to your first one. The Observer profiled Brynjolfsson the same month as a measured economist standing between the doom and boom camps. And on May 11, Bloomberg built a feature on his 2020 wager with Northwestern’s Robert Gordon that AI would push U.S. labor-productivity growth above 1.8% a year through 2030. A researcher becomes the center of gravity when the bear case and the bull case cite the same data. That is where Brynjolfsson now sits.

His long-running co-author, MIT’s Andrew McAfee, holds the adjacent ground, and his warning is the one that lands hardest with operators. In Fortune on May 1, McAfee argued that automating Gen Z entry-level roles could backfire and cost companies their future workforce. That is not a claim about the labor market in aggregate. It is a claim about a single firm’s leadership pipeline, and it is what turns an economics paper into a board agenda item.

Three readings of the same evidence

Evidence narrows a debate, but it rarely ends one. The most useful thing a leadership team can grasp right now is that serious people read the same payroll data and reach different conclusions, and that the disagreement is itself the strategic terrain.

Three readings are worth considering.

The first is the case for holding the line. In MIT Sloan Management Review on May 19, 2026, Andrew Winston argued in “Companies Don’t Have to Slash Jobs Because of AI” that firms which preserve their entry-level pipeline gain a durable advantage over those that cut it, because the entry-level rung is where senior talent is grown. McAfee’s Fortune warning runs in parallel. The logic is plain: a company that removes the bottom rung to capture a near-term saving is borrowing against its own future.

The second is the skeptic’s read. In SHRM on May 18, 2026, Wharton’s Peter Cappelli argued that the financial case for AI-driven layoffs is often overstated. “The companies that are laying off are not struggling,” he said, pointing to investor pressure and a changed attitude toward labor rather than to measured AI productivity; many announcements, he noted, say only that firms expect AI to cover the work, not that it has. The point is bracing in a boardroom: some share of what is framed as AI displacement is ordinary cost-cutting in fashionable language, and a board that cannot tell the difference cannot govern the decision.

The third is the causation challenge, and it is the one most likely to surprise a leadership team. In Fortune on May 15, Stanford’s Nicholas Bloom, who has done the defining work on remote work, made the case that the post-2020 U.S. productivity boom predates widespread AI adoption and is driven substantially by working from home. If he is right, a finance chief who attributes recent productivity gains to AI, and plans headcount and real estate accordingly, is solving the wrong equation. The two explanations point to opposite operating decisions, which is why the question cannot be waved away.

Around these readings sits the research frontier. Matt Beane has a working paper, “Hedging With Talent: How Selection Under Uncertainty Reshapes Automation’s Workforce Effects” (SSRN), on how firms actually decide whom to automate and whom to keep. Daniel Rock, the Wharton economist, has pressed the more practical question of which uses of AI generate measurable value, in AEI commentary and at length on James Pethokoukis’s Faster, Please! on April 28. And Amy Edmondson, whose work on psychological safety has shaped a generation of management practice, supplies the inside-the-team view; her April 22 Fast Company essay with Tomas Chamorro-Premuzic asks what happens to candor and learning when leadership tightens and teams are asked to do more with tools they do not yet trust.

No single voice resolves the question. Together they describe its full shape, which is what a leadership team actually needs.

The compounding cost of waiting

Faced with a contested question, the instinct is to wait for the evidence to resolve. The structure of the decision argues against it. Workforce choices compound: an entry-level cohort hired or forgone in a given year becomes the management bench three years on and the senior-leadership bench in seven or eight. The savings captured by thinning that cohort shows up in the current quarter; the gap it opens does not surface until the people who would have filled it are missing, by which point a later correction can recover only part of what was lost.

This is the rare point on which the optimists and the skeptics converge. Winston’s case for protecting the pipeline and Cappelli’s warning about cuts dressed in the language of AI disagree about almost everything, yet both land on the same caution. The entry-level decision is the one most likely to be made for near-term reasons and regretted on a long horizon, and it is the decision furthest removed in time from its own consequences. That distance is precisely what makes it easy to get wrong.

For any organization planning the back half of the year, that lag is the operative fact. The view a leadership team forms about AI and its workforce takes hold in hiring plans, budgets and board conversations well before the results register anywhere measurable. The evidence assembled this spring is not, on that reading, a story to file for later. It is an input to decisions already underway.

One reckoning, several vantage points

The entry-level question does not stand alone. It is one face of a larger reckoning, and the leaders who see the connections will plan better than those who treat each story as separate.

The most direct adjacency is the one the finance chief is already living. As Nvidia’s results and the broader capital-expenditure debate moved to the center of the business-news cycle this spring, a credible group of voices began asking whether the spending behind the AI boom is priced correctly. In the Atlantic on May 1, Azeem Azhar published “So, About That AI Bubble,” the boardroom-credible version of the skepticism. Ed Zitron has pressed the financial case hardest in his newsletter’s running series on AI unit economics, including “OpenAI Had a Negative 122% Operating Margin in Q1 2026.” Songyee Yoon, the venture investor and former NCSOFT president, framed the cycle for an investor audience on NYSE TV. The link to the jobs question is not incidental: if the capital case for aggressive automation is shakier than it looks, the hiring freezes justified by that case deserve the same scrutiny. The finance chief’s bubble question and the people chief’s pipeline question are two sides of one coin.

A second adjacency sits closer to the daily work of running a team. Even where AI delivers, the gains tend to stall in the middle of the organization, a pattern Tom Davenport has documented in his work on why enterprise AI redesign keeps getting stuck. That stall raises a leadership-design question. Stephan Meier has written in Forbes on what executive judgment looks like when expertise is being commoditized. Linda Hill‘s research on collective genius, captured in a recent McKinsey conversation, and Raffaella Sadun‘s work on organizational transformation both point to the same conclusion: the constraint on AI value is rarely the model and usually the organization.

A third adjacency runs underneath all of it. The promise of AI rests on data most companies have not put in order. In MIT Sloan Management Review on May 21, Barbara Wixom framed data transformation as the chief executive’s business rather than a delegated technical task, with the Caterpillar case as her evidence. Faisal Hoque‘s work on AI-ready enterprise architecture occupies the same ground. The entry-level question, the capital question and the organizational question all rest, finally, on whether a company’s data foundation can support the decisions it is being asked to make.

These are not five separate conversations. They are one conversation about how organizations absorb a general-purpose technology, seen from the seats of the people who have to decide.

From evidence to decision

The pattern across the spring is clear enough to act on. The AI-and-jobs question has matured from forecast into evidence, the evidence is unresolved in instructive ways, and the decisions it forces are being made on a timeline that runs well ahead of the moment they become visible.

What a leadership team needs in that situation is not reassurance and not a single confident answer. It is a clear account of what the data shows, an honest map of where serious researchers disagree, and a way to turn both into the decision on the desk. That is the difference between hearing a compelling talk and changing how an organization hires, develops and deploys its people.

It is the difference Stern is built around. We represent the researchers whose published work is named above: Brynjolfsson on the data, Winston and McAfee on the case for protecting the pipeline, Cappelli on reading layoff announcements honestly, Bloom on what is actually driving productivity, with Beane, Rock and Edmondson on the frontier and inside the team. The keynote is where a leadership team learns the framework. The advisory engagement is where the framework becomes a 2026 workforce design a CHRO can defend to the board and live with for a decade. The gap between those two things is where most engagements end. It is where this work begins.

These researchers and advisors are available for keynotes, board sessions and confidential advisory work through the fall planning cycle. For a custom proposal, please reach out to schedule a call with our speaker team.

Stern Strategy Group represents the researchers and advisors whose work is shaping how organizations absorb artificial intelligence, from the economists who own the data to the advisors who turn it into operating decisions. Available for keynotes, board sessions and confidential advisory work.