Firms built billion-dollar margins on armies of cheap junior analysts. AI now does most of that work, and the business model that turned graduates into partners is starting to buckle.
The short version: More than one in four college graduates aged 23 to 27 now live with their parents, up from about 18% in 2001. The headline reason is a weak entry-level job market. The business reason sits underneath it: the professional-services pyramid, where a wide base of junior analysts funds a narrow tier of partners, is the corporate structure most exposed to AI. Cut the base and you raise near-term margins. You also break the apprenticeship that manufactures your future partners. That is the trade firms are making right now, and most of them are not pricing the second half of it.
A recent graduate with a literature degree moves back into his childhood bedroom in the Bay Area, fires off applications, and dates another writer who moved home too. The New York Times ran that scene this week as a story about shifting family norms and cheaper rent. Read the labor data underneath the anecdote and a different story shows up, one about which company types can no longer justify hiring the people those graduates used to become.
What happened
The unemployment rate for recent college graduates hit 5.6% in the first quarter of 2026, according to the Federal Reserve Bank of New York. National unemployment sat near 4.2%. For the first time in the four decades the government has tracked it, the credential that was supposed to buy a career advantage now correlates with a slightly worse chance of having a job.
The damage concentrates in exactly the fields you would expect AI to touch. Recent-graduate unemployment for computer engineers more than tripled to 7.8% between 2022 and 2024. For chemical engineers it more than doubled to 4.7%. These are not the humanities majors that career-advice columns love to scold. They are the technical degrees that were supposed to be automation-proof.
The buyers of that labor have pulled back hard. Entry-level roles fell to roughly 7% of new hires at big technology firms in 2024, a 25% drop from the year before and more than half below pre-pandemic levels, according to an analysis by the venture firm SignalFire. In the UK, a single graduate consulting role now draws around 140 applications, up from 86 two years earlier, per the Institute for Student Employers.
The professional-services firms show the clearest fingerprints, because they publish their intentions. KPMG cut its UK graduate class by 29%, from 1,399 hires to 942. Deloitte trimmed its UK intake by about 18%, EY by 11%, and PwC by 6%. PwC then cut graduate hiring again in 2025 and abandoned a five-year-old target to add 100,000 employees globally by 2026, blaming generative AI directly. Its global headcount fell by 5,600 last year. Two Big Four executives told the Financial Times that UK graduate recruitment could drop by roughly half in the coming year.
How the pyramid actually prints money
Consulting, accounting, and law all run the same machine. A wide base of junior analysts does the research, builds the models, and assembles the slides. A firm bills those juniors out to clients at a heavy markup over what it pays them. The spread funds the partners at the top, who sell the work and own the profit. Consultants call it leverage: the more juniors a partner can profitably supervise, the more money the partner makes.
The base was never staffed for cheap labor alone. Ask most partners how they learned to structure a problem or read a client and they will describe their analyst years. Our own breakdown of the highest-paying entry-level jobs describes a first consulting job as a compressed MBA, the place where a graduate learns operating models, cost structures, and executive communication by doing them under pressure. The base is both the profit engine and the training academy. Those two functions ran on the same people, which is why nobody had to think about them separately.
Why professional services is ground zero
Generative AI is good at precisely the work the base was hired to do. One industry estimate puts AI at roughly 80% of a junior analyst’s typical research and slide-generation output. McKinsey’s internal tool, Lilli, reached more than 7,000 consultants and reportedly cut about 30% of the time they spend on research and synthesis. BCG built a deck-formatting tool to do what analysts once did by hand at midnight.
Brookings researchers found AI could automate more than half the tasks in entry-level positions, roughly five times the exposure of senior roles. That asymmetry is the whole problem for a leveraged business. The cheapest, most numerous, most billable layer is the one the technology hollows out first.
The near-term math looks like a gift. Fewer analysts on an engagement means fewer salaries against the same fee, so margin per partner climbs. Firms have started freezing what they pay the survivors: the three big strategy houses have held starting salaries flat for three years running, around $135,000 to $140,000 for undergraduates, and the Big Four have not raised entry pay since 2022. Accenture cut about 22,000 roles in 2025 as part of an $865 million restructuring and framed it around AI-driven efficiency. On a spreadsheet, this is a firm getting leaner. You can see why leadership likes the slide. Accenture’s own service-delivery model has always leaned on scale, and scale is the thing AI compresses first.
The part nobody is pricing
Kill the base and the margin goes up this year. The seed corn goes with it.
A firm with no analysts has no thirty-year-olds who spent three years learning to run engagements. In a decade it has no partners who came up that way, because the ladder’s bottom rungs are gone. The judgment that clients pay a premium for, the part of the job AI cannot do, gets built by grinding through the part of the job AI now does. Automate the training ground and you stop producing the seniors whose scarcity is the entire pricing power of the model.
Firms know this, which is why leadership keeps promising a “diamond” instead of a pyramid: a thinner base of juniors, a thick middle of experienced experts, a top of advisors. The honest question is where the thick middle comes from once you stop hiring and training the bottom. A diamond with no intake is a countdown. The billable-hour logic is already cracking under the same pressure. McKinsey now ties roughly a third of its work to performance-based fees, because when AI compresses a six-week analysis into three days, clients stop paying for the six weeks.
The counterargument
The apocalypse case has holes, and a good skeptic should press on them. Broad white-collar employment has not collapsed. The US economy added roughly 3 million white-collar jobs in the three years after ChatGPT launched, and several occupations pegged as AI roadkill, including software development, grew rather than shrank. Stanford and Atlanta Fed researchers find the aggregate labor data calm, with strain showing up narrowly at the graduate and entry level rather than economy-wide.
There is also a benign reading of the same numbers. Indeed’s lead economist calls the pattern “experience creep”: employers still want graduates, they just want the judgment that entry-level jobs used to build, and they are demanding it up front. If that holds, the market is repricing what a junior is worth, not eliminating the role. Firms have also survived predictions of their death before, from offshoring to the last three recessions, by adding services faster than technology removed them. BCG says AI and tech work already drives 40% of its revenue, which is a firm feeding on the disruption rather than drowning in it.
The counterargument’s weak point is timing. Experience creep and a broken training pipeline are the same event described by an optimist and a pessimist. If you need three years of analyst work to build judgment, and firms stop offering three years of analyst work, “employers want more experience” and “employers stopped manufacturing experience” describe one problem, not two.
The risk
The exposure is not evenly spread, and the fragile firms are the ones whose only real product is leverage. A boutique that sells a named partner’s expertise barely runs a pyramid, so it has little base to lose. A mid-tier firm that sells a thousand interchangeable analysts at a markup is selling the exact thing a client can now rent from a model. The Big Four straddle both, which is why their graduate cuts read as a hedge rather than a strategy.
The demand side carries its own risk. When clients see AI compress the work, they stop accepting the old bill. Outcome-based pricing spreads, the billable hour erodes, and the firm loses the metric its whole economic model was denominated in. A business can survive automating its costs. Repricing its revenue at the same time is the harder trick, and it is the one the market is now forcing.
Quick questions
Is AI actually causing the graduate job slump, or is it just a weak economy? Both, and separating them cleanly is hard. Overall unemployment is low, which argues against a pure recession story. The damage clusters in AI-exposed, entry-level, technical roles while senior hiring holds, which is the specific signature you would expect from automation hitting the base first.
Are the big firms firing consultants? Mostly not the client-facing seniors, yet. The cuts so far concentrate in graduate intake, back-office and support functions, and headcount targets quietly abandoned. The base is being starved through reduced hiring more than emptied through layoffs.
If juniors get automated, why do salaries stay frozen instead of rising for the survivors? Because supply is flooding in while demand shrinks. Around 140 applicants now chase a single UK graduate consulting seat. When far more people want the role than there are roles, firms have no reason to raise the price.
Does this mean a degree is worthless now? No. The degree still helps, and the return on an MBA still clears for the strongest programs. What changed is the guarantee. A credential used to function like insurance against unemployment. That policy has lapsed, and graduates now compete on demonstrated skill and a foot already in the door.
The Business Model Analyst Take
The graduates moving back home are a leading indicator, and the thing they are indicating is not a soft patch in hiring. It is that a specific, lucrative business structure is being cannibalized from the bottom by its own biggest customers, the firms selling the AI. Consulting’s leverage model was an arbitrage on the gap between what a smart 23-year-old costs and what a client will pay for their output. AI closes that gap, so the arbitrage thins.
Watch what the firms do, not what they say about diamonds. If graduate intake keeps falling for another two or three cycles, the pyramid stops being a staffing debate and becomes a succession crisis, because the partners of 2035 are the analysts nobody is hiring in 2026. The firms treating the automated base as a pure cost line are booking a margin gain today against a talent bill that comes due after the current leadership has cashed out. The ones that survive with pricing power will be the ones that figure out how to build senior judgment without a cheap junior tier to build it in, and right now no major firm has publicly solved that. Until one does, “we’re moving to a diamond” is a press release, and the countdown is running.
