A senior economist at one of the world’s largest asset managers has looked at the American jobs data and concluded the AI employment crisis is not there. Torsten Sløk, chief economist at Apollo, points to two numbers: the ratio of job openings to unemployed workers has climbed back above one, and US nonfarm payrolls rose by 172,000 in May. His verdict is blunt — “there are no signs of workers being replaced by ChatGPT.” He is right about the data. The problem is what he takes it to prove. Aggregate payrolls are not a smoke detector for AI disruption; they are the smoke that drifts in long after the fire has changed which rooms are habitable. For UK organisations, the more useful question is not whether the macro data shows a crisis, but why it would be the last place to look.
The business problem behind a reassuring chart
Sløk’s argument is clean, and that is its weakness. The logic runs: if AI were destroying jobs, we would see openings collapse and unemployment climb; instead openings are firm and hiring continues; therefore AI is not displacing labour, and may even be spurring the business formation that creates new openings. Every step is defensible. The conclusion still misleads, because it tests a hypothesis about composition using an instrument that only measures totals.
An economy can add 172,000 jobs in a month whilst quietly closing the front door on an entire cohort. Net payroll growth is the sum of every hire minus every separation across every occupation and age band. A firm that stops recruiting graduates but expands its experienced sales team shows up as growth. A sector that automates its junior analysts whilst poaching senior ones from rivals shows up as churn, not loss. The headline number is doing exactly what it is designed to do — and that design makes it structurally blind to the one pattern that matters here.
Critical Context: Sløk’s claim is not false. It is a claim about the aggregate, presented as a claim about the mechanism. “No net displacement visible in payrolls” and “no AI displacement happening” are different statements, and the gap between them is where this year’s UK hiring story lives.
The UK data makes the gap visible. Where the American aggregate looks calm, the British entry-level market does not.
The numbers that frame the debate
| Indicator | United States (Sløk’s read) | United Kingdom (2026) |
|---|---|---|
| Headline labour signal | Payrolls +172,000 in May; openings-to-unemployed above 1.0 | Unemployment 4.9%–5.2%, a five-year high |
| Competition for roles | More jobs than workers to fill them | 2.5 unemployed per vacancy, up from 1.9 a year earlier |
| Graduate hiring | Not separately flagged | Graduate postings down ~13% year-on-year |
| Entry-level hiring | Not separately flagged | Down ~14% year-on-year |
| Sharpest occupational falls | Not visible in the aggregate | Accountant −29%, graphic designer −28%, software engineer −27% |
| Employer intent | Implied benign | Around half of leaders expect net job losses; roughly one in six plan AI-related cuts |
The two columns are not in conflict. They are the same phenomenon viewed through instruments of different resolution. Sløk is looking at the whole building’s thermostat. The UK figures are a reading from the ground floor, where the temperature is genuinely changing.
What the aggregate cannot see
The reason payroll data lags an AI shift is not a flaw to be corrected. It is inherent to how the disruption propagates. Automation rarely arrives as a redundancy round; it arrives as a hiring decision not taken. A team that would have brought in two junior analysts brings in one and gives the rest to a model. No one is made redundant, no separation is recorded, and the macro series registers nothing — yet the rung at the bottom of the ladder has been quietly sawn off. The displacement is real, but it is denominated in jobs that were never created, and you cannot count an absence.
This is why the signal surfaces first in the data Sløk’s note does not cite: the flow of new openings at entry level, the application-to-vacancy ratio, the occupational mix of what is being advertised. UK graduate postings falling 13% in a year, against an entry-level market down 14%, is the absence becoming measurable — but only because someone is counting openings rather than headcount. Aggregate employment will be the last series to move, because it only moves once the cohort that never got hired ages into the unemployment statistics or drops out of the labour force entirely.
Strategic Reality: The metrics that would vindicate Sløk’s “all clear” — stable total employment — are precisely the metrics that move last. By the time an AI labour shift is unambiguous in nonfarm payrolls, the compositional change that caused it will be three to five years old.
There is a second blind spot. The American picture and the British one diverge partly because the economies are not running the same experiment. The US is absorbing AI into a labour market with stronger business formation and, in Sløk’s framing, AI itself driving new firm creation. The UK is absorbing it into a market already soft — vacancies down 7.1% year-on-year, a hiring freeze that predates the technology, and a graduate pipeline under separate, compounding pressure. The same tool lands differently on the two surfaces. A UK leader who imports Sløk’s conclusion wholesale is borrowing reassurance calibrated to a different economy.
The human cost the totals net out
Behind the netting is a generation. The occupations falling fastest in the UK — accounting, graphic design, software engineering at junior level — are the classic entry points into professional careers. These are the roles where people have historically learned the craft before the craft could be partly automated. Remove the bottom rung and you do not just reduce this year’s graduate intake; you interrupt the mechanism by which an industry produces its next decade of senior talent.
The public has noticed before the economists have agreed. Survey work this year found roughly seven in ten UK workers worried about AI-driven job losses, and a striking minority fearing social unrest as a consequence. That anxiety is not irrational over-reaction to a non-event. It is a population reading the ground-floor temperature directly, whilst being told by aggregate-level commentary that the building is fine.
Reality Check: When public sentiment and macro data diverge this sharply, the instinct is to trust the data and dismiss the sentiment as fear. In a compositional shift, that instinct is exactly backwards. The people closest to the affected roles are sampling a signal the aggregate has not yet integrated.
For UK organisations, the stakeholder picture is uneven by design.
| Stakeholder | Exposure to the compositional shift | What they actually need from you |
|---|---|---|
| Graduate and entry-level hires | Highest — the roles being quietly not-created | A defined route in that survives automation of the tasks, not just the title |
| Mid-career staff | Moderate — tasks reshaped, roles mostly intact | Reskilling toward judgement and oversight work AI cannot yet own |
| Senior leaders | Low displacement, high decision risk | A workforce strategy that does not wait for the macro all-clear |
| The talent pipeline (5-year view) | Severe if the bottom rung stays cut | Deliberate investment in early-career development as a strategic asset |
What to do before the data agrees with you
The strategic error Sløk’s note invites is patience — the sense that, because the aggregate is calm, there is time to wait and see. There is not, for a specific reason: if the disruption is compositional, the cost of waiting is paid by your future talent base, and that cost is incurred silently, year on year, in cohorts you never recruited. The organisations that handle this well will act on the leading indicators their own hiring data already contains, rather than the lagging one a US payroll print provides.
Take Action: Stop benchmarking your AI-and-jobs position against national employment figures. Benchmark it against your own funnel — applications per entry-level vacancy, graduate offers made versus three years ago, and the proportion of junior tasks now handled by tooling. Those are your ground-floor thermometers, and they will move long before the macro series does.
A practical sequence, scaled to where an organisation sits:
- If you are early in AI adoption: Audit which roles you have quietly stopped recruiting for, not which you have made redundant. The “not-hired” column is where your exposure actually sits, and it rarely appears in any formal review.
- If you are mid-adoption: Redesign the entry-level role rather than deleting it. If a model now does 60% of a junior analyst’s former tasks, the role becomes 40% craft and 60% oversight — and that is a development pathway worth keeping, not a redundancy to book.
- If you are mature in adoption: Treat early-career development as a deliberate investment with a multi-year payback, accepting that the market is currently making it cheap to stop doing so. The firms that keep their pipeline funded through this period will own the senior talent shortage everyone else is busy creating.
SME Advantage: Smaller UK organisations can turn the slack market to their benefit. With graduate roles drawing an average of around 140 applications each, a firm that keeps a credible early-career route open has access to a stronger applicant pool than it has seen in years — at exactly the moment larger competitors are pulling back.
Four challenges the headline hides
The lag is a trap, not a reassurance. The longer the aggregate stays calm, the more confident the “no crisis” reading becomes, and the larger the compositional gap grows beneath it. Confidence and exposure rise together. Mitigate by tracking openings and occupational mix, never relying on total employment as your early-warning system.
Automation of tasks is mistaken for safety of roles. Because few people are being made redundant, leaders conclude the roles are secure. The roles are being hollowed from within — same title, fewer of them created, different skill mix required. Treat “we haven’t cut anyone” as a statement about the past, not a forecast.
Cross-border data borrowing. US labour data is more abundant and more frequently cited than the UK equivalent, so British decisions quietly inherit American conclusions. The two markets are running the technology through different conditions. Anchor on UK-specific series — ONS vacancy and entry-level data — before importing any transatlantic “all clear.”
The pipeline cost is invisible until it is structural. Cutting graduate intake saves money this year and shows up nowhere problematic for several. The bill arrives as a senior-talent shortage half a decade later, by which point it cannot be quickly fixed. Mitigate by costing the pipeline as an asset with depreciation, not a discretionary line to trim when hiring is soft.
The takeaway for UK leaders
Sløk has done something useful: he has shown that the AI jobs crisis is not yet a macro event, and that the apocalyptic framing of total employment collapse is not supported by the data. That is worth knowing, and worth saying. But “not a macro event” is not “not happening,” and the distance between those two is the entire UK entry-level story of 2026 — graduate postings down, junior roles thinning, the bottom rung quietly removed whilst the totals hold.
The leaders who navigate this well will hold three things at once. First, that the absence of an aggregate signal is evidence about measurement, not about reality. Second, that the disruption is compositional and concentrated, so it must be tracked with compositional instruments — your own hiring funnel, not the national headline. Third, that the cost of waiting for the macro data to agree is paid in a talent pipeline you cannot rebuild on demand.
Strategic Insight: The most expensive way to manage an AI workforce transition is to wait until it is visible in the figures everyone watches. By then the cheap, reversible decisions are behind you and only the expensive, structural ones remain.
A short checklist to leave with:
- Replace national employment benchmarks with your own funnel metrics as your AI-and-jobs dashboard.
- Audit the roles you have stopped creating, separately from the roles you have cut.
- Redesign entry-level positions around oversight and judgement rather than deleting them.
- Fund the early-career pipeline through the soft market as a deliberate, costed investment.
- Anchor every transatlantic data point against UK-specific ONS series before acting on it.
The AI jobs crisis is not where Sløk looked, and he was honest enough to say so. It is one floor down, in the data about who is no longer being hired — and that floor is where UK strategy has to be set.
Source and attribution
This analysis responds to Torsten Sløk’s note “Where Is the AI Jobs Crisis?” published via Apollo’s The Daily Spark (Apollo Global Management, June 2026), which argues that US labour-market data shows no evidence of AI-driven displacement. UK labour-market figures referenced here draw on Office for National Statistics labour-market reporting and contemporaneous 2026 survey and hiring-tracker coverage of graduate and entry-level employment.
Resultsense provides independent analysis of AI developments for UK professionals and businesses. We translate technical and economic signals into strategic guidance, helping organisations act on leading indicators rather than lagging ones. Get in touch to discuss your AI workforce strategy.