AI didn't kill the entry-level job. The business model killed it.

September 13, 2026 · AiHiringProfessional servicesEntry level

The “AI killed the entry-level job” narrative is running well ahead of the evidence for it.

Tracy Layney, who has run human resources for a decade, argues the collapse is a leadership choice, not a technological one. New York’s WARN Act filings back her up: in the year since the state started requiring an AI-disclosure checkbox on mass-layoff notices, zero of more than 160 filings checked it. And Benedict Evans’s “decoupled business” frame explains why the narrative sticks even when the data doesn’t: a job can be AI-immune while the business model that employs the person isn’t.

The real story isn’t whether AI can do the work juniors do. It’s whether AI breaks the business model that employs them. And that’s different in every sector — sometimes in the same sector, at the same time, depending on which business model a firm runs.

The layoff data that doesn’t support the narrative

Tracy Layney put it plainly in a recent Charter piece: “AI did not kill the entry-level job. Leaders did.” Young people are entering the job market in an era of “impervious, bot-driven online applications” and headlines warning of an AI-driven apocalypse — but the reason they can’t find jobs, she argues, is that leaders made a decision. IBM’s CHRO tripling entry-level hiring is her evidence that this is a choice, not physics.

The layoff filings back that up more directly than she does. New York requires employers to check a box on their WARN Act mass-layoff notice if “technological innovation or automation” contributed to the cuts. In the year since the disclosure requirement took effect, more than 160 companies filed WARN notices — and not one checked the box. Separately, a Harvard Business Review survey of 1,006 executives found that 39% of organizations made low-to-moderate headcount reductions in anticipation of AI’s future impact, and a further 21% made large reductions for the same anticipatory reason — but only 2% said they’d made large cuts tied to AI actually delivering.

Both readings have holes. The WARN checkbox is new, undefined in statute, and companies may simply not know how to answer it, or choose not to — its absence of AI-attributed cuts could mean AI isn’t the cause, or it could mean the reporting system doesn’t work. The HBR gap is self-reported executive sentiment, not verified fact. But together they point the same direction: when companies are asked directly whether AI caused the cuts, the honest answer, in the mechanisms that exist to capture it, is mostly no.

This is the first point most readers will push back on. The narrative feels true because everyone in it has a reason to believe it. New graduates do. HR directors do. And so do consultants — because the narrative makes AI the villain, and the villain is easier to manage than your own leadership decisions.

The decoupled business risk

Benedict Evans has the sharper frame for why the narrative persists anyway. A job can be AI-immune while the business model that employs the person isn’t.

Journalism as a craft was untouched by the internet. The classified-ads business that paid for it was gutted. The skill survived; the revenue source didn’t.

The analogous question for AI is this: which low-exposure jobs depend on companies whose economics get wrecked by AI cutting the cost of some other input, or whose “moat” was literally “buildings full of people doing boring things”?

For professional services, this is where the thesis gets useful. The big three — accounting, consulting, legal — all face the same AI capability pressure. Their junior workers can all be assisted or partly replaced by AI in theory. But their business models differ, and the hiring outcomes differ with them.

Accounting: the pyramid base is thinning

The Financial Times analyzed more than 50,000 job postings from Deloitte, EY, KPMG, and PwC across six English-speaking countries between January 2020 and January 2026. Roles requiring AI skills made up nearly 7% of postings in 2025, up from under 2% in 2022 — more than tripling. Audit postings, in decline for longer, accounted for just under 3%.

Graduate intake has fallen sharply, and by verified, if UK-specific, amounts: KPMG cut its 2023 graduate class from 1,399 to 942 — a 29% drop. Deloitte cut 18%, EY 11%, PwC 6%. Separately, UK accountancy graduate job adverts fell 44% year-over-year as of mid-2025, per Adzuna posting data — a distinct metric from the firms’ own reported intake figures above, but pointing the same direction.

A May 2026 BambooHR survey — vendor research, from an HR software company — of 1,248 US small-to-midsize business respondents, paired with six years of workforce data from 480,000+ employee records, found a 3:1 senior-to-entry-level hiring ratio in finance and accounting, with entry-level job-posting share down from 12% in 2020 to 10% in recent years and analyst postings down 21%.

The traditional accounting pyramid — lots of juniors doing data entry and basic compliance checks, seniors reviewing, partners selling — is visibly thinning at the base by these numbers. The resulting shape looks less like a pyramid than a diamond: a thinner base, a wider middle of technical and managerial roles, and AI handling volume work underneath.

There is no regulatory barrier here. The question is skill transition, not legality: can accountants learn to audit AI output faster than AI automates their current tasks? The postings and graduate-intake data say, at the UK Big Four at least, the answer so far is “not fast enough” — though the graduate-intake figures are UK-specific; the equivalent US figures aren’t public.

Consulting: contradictory data

McKinsey’s own numbers contradict the narrative directly. Eric Kutcher, senior partner and chair of McKinsey North America, told reporters in September 2025 that the firm plans to hire 12% more people in North America in 2026 than in 2025, and expects non-partner headcount in the region (currently 5,000–7,000) to grow 15–20% over five years.

At the same time, McKinsey CEO Bob Sternfels told Harvard Business Review’s IdeaCast that the firm’s total headcount is now 60,000 — 40,000 humans and roughly 20,000 AI agents, up from about 3,000 agents eighteen months earlier — and that he expects every employee to be “enabled by at least one or more agents” within 18 months. Separate reporting puts McKinsey’s cuts at 3,000–4,000 positions in 2025–2026, roughly 10% of its global workforce and its largest reduction since 2008, concentrated in back-office functions and junior research roles.

Sternfels also told HBR the firm is deliberately shifting hiring toward liberal arts majors, because AI is strong at “linear problem solving” but not at “discontinuous leaps, truly novel thinking” — skills he says a liberal arts education selects for more than a traditional consulting-track degree does. And separately, McKinsey is piloting an AI interview stage using its internal tool Lilli in select US final rounds. On business model, Sternfels told HBR the firm is moving from fee-for-service advisory work toward outcomes-based engagements — about a third of revenue today, with a stated goal of a majority before he leaves the role — where McKinsey and the client jointly define a business case and McKinsey’s fees are tied to the outcome delivered.

Deloitte is restructuring its job-title architecture for all 181,500 US employees, effective June 1, 2026, replacing the analyst-consultant-manager ladder with function-specific titles and alphanumeric levels. Deloitte itself says day-to-day work and compensation philosophy are unchanged — this is a title and leveling change, not a confirmed headcount signal either way. Separately, Deloitte’s own research — vendor material, self-published by the firm — argues AI is producing a “broken skills ladder”: analysis of roughly 19,000 workplace tasks found AI automates disproportionately more of the low- and mid-level tasks that historically taught early-career workers their trade, and Deloitte’s analysis of 2022–2025 US job postings found employers hiring fewer entry-level data scientists and software developers while senior-level demand held. Worker access to AI tools rose 50% in 2025.

Accenture cut roughly 11,000 roles between May and August 2025 — headcount fell from about 791,000 to 779,000 — as part of an $865 million restructuring charge. CEO Julie Sweet told investors the firm is “exiting people” on a compressed timeline “where reskilling simply isn’t a viable path,” while separately reporting it had trained 550,000+ employees in generative AI fundamentals and grown its AI and data specialist headcount from 40,000 in FY23 to 77,000 in FY25. Accenture also said it expects to grow overall headcount in FY26.

The differentiator that best explains the divergence is business model, not firm size or reputation: McKinsey is moving toward outcomes-based fees, by Sternfels’s own account, and is expanding junior headcount even as it automates the linear-problem-solving work juniors used to do. Firms still running pure fee-for-service billing — where junior hours are the product sold — look more exposed when AI compresses those hours. That’s an inference from the McKinsey contrast, not a documented finding.

Law: where regulation creates a ceiling

Legal services have a different constraint. Unauthorized practice of law (UPL) is a crime in most US jurisdictions, enforced at the state level. An AI can draft a contract or research a case, but it cannot independently provide legal advice or sign off on filings — a licensed attorney must review and certify.

In March 2026, Nippon Life Insurance sued OpenAI, alleging that ChatGPT helped a former claimant draft dozens of post-settlement legal filings — including a fabricated case citation — effectively practicing law without a license, and separately interfered with a signed settlement agreement.

A suit over an AI tool producing fabricated legal authority isn’t evidence that UPL rules are under pressure to relax. If anything, it points the other way: the fabrication surfaced precisely because AI-drafted work still needs a licensed professional to catch it, which is the argument for keeping the sign-off requirement, not loosening it. Read it as a live liability dispute, not a policy signal yet.

The bottleneck in law is regulatory, not technical. The pipeline is intact so far because a licensed hand still has to turn the final page. Whether that continues depends on state-level UPL enforcement and litigation outcomes like Nippon Life’s, not on AI capability.

What a reader should do

If you’re in accounting, the FT’s postings data and the UK graduate-intake cuts are the clearest sourced signal in this piece. Ask which of your junior roles are being eliminated for AI-efficiency reasons versus leadership-structure decisions — and note the clearest numbers here are UK-specific; ask for your own region’s data before generalizing.

If you’re in consulting, read the signals by business model, not by firm. McKinsey’s hiring increase and its shift toward outcomes-based fees are both real and both sourced directly to its own leadership. Firms still running pure fee-for-service work may be more exposed, but that’s inference, not documented fact yet.

If you’re in law, the regulatory ceiling is holding, and the Nippon Life case is more likely to reinforce it than erode it. Watch UPL enforcement and litigation, not AI capability announcements.

If you’re hiring — or deciding whether to hire — the sourced version of the question is: does your business model bundle junior labor with billable hours in a way AI compresses, the way accounting’s pyramid does? Or does it look more like McKinsey’s outcomes-based shift, where junior headcount can grow even as the nature of junior work changes?

The counter-case

Evans’s own Jevons-paradox argument is the strongest reading against all of this: cheaper task execution can increase total demand for that task category rather than cut headcount. A DCF that takes 30 seconds instead of a week might just mean more DCFs get requested, more analysts needed to interpret them. This applies best to elastic-demand analytical work — exactly the kind of work entry-level analysts do.

It’s a real counter-argument, and it’s a theory, not yet data specific to this claim. The Indeed Hiring Lab’s May 2026 data shows the opposite pattern in aggregate: senior-level postings up 14.7% year-over-year, entry-level postings down 6.3%. But Indeed measures postings, not hires, and doesn’t isolate professional services — it’s a labor-market-wide signal, not a direct test of Jevons within accounting or consulting. If Jevons is working specifically in the sectors this piece covers, the postings data so far doesn’t show it, but that’s a different and weaker claim than “the data contradicts Jevons.”

What to watch

Sources

Cut from this post: a widely circulated “$9 billion” Big Four AI infrastructure figure. No current, dated primary source supports a live combined figure at that size — the number traces to 2020-era reporting on three firms’ initial commitments, not current spend.

← All Articles