Where Is the AI Jobs Crisis? The Macro Data Can't See What It Isn't Measuring
Apollo's chief economist uses rebounding job openings and the May payroll print to argue there's 'no sign of workers being replaced by ChatGPT.' But aggregate averages are a natural muffler for localized shocks. The real disagreement isn't about the data. It's about which lens you use to read it.
Summary
On June 9, Apollo chief economist Torsten Slok used his Daily Spark column to ask a blunt question: where is the AI jobs crisis? His answer was equally blunt. If AI were really triggering a jobs crisis, we’d expect job openings to collapse and unemployment to climb. The opposite is happening. Job openings per unemployed worker have climbed back above 1.0, meaning there are still more openings than people to fill them, and May nonfarm payrolls jumped by 172,000. From that he concludes there’s “no sign of workers being replaced by ChatGPT.”
The post drew 261 comments on Hacker News. What’s telling isn’t that people called him wrong. It’s how fast the argument slid from “is the data accurate” to “this data can’t even see the thing we’re worried about.” That’s the crux. Slok isn’t lying; every figure he cites is official Bureau of Labor Statistics data. But the lens he chose, economy-wide hiring totals and all-industry unemployment, is a natural muffler for localized shocks. A labor market with a stable headline can be hiding an entry-level collapse and a health-care boom at the same time, the two canceling out in the aggregate while the macro curve sits dead flat. So the question to ask isn’t whether there’s a crisis. It’s which lens lets you see it at all.
The debate
On the surface, the fight is about whether AI is destroying jobs. Pull it apart and the disagreement lands on three competing but not mutually exclusive explanations, each carrying a different implication for policy and hiring.
The first is the lagged effect. Technology shocks to employment never show up the same quarter: firms freeze hiring first, then stop backfilling, then cut, a sequence that plays out over several quarters or years. People in this camp argue that today’s aggregate numbers reflect yesterday’s decisions, and AI’s real bite is still in transit. One HN commenter named the weakness of this view in a single line: the crisis is “always two to three years away”; 2027 was once billed as the inflection point, and that prediction is already a year or two old. The lagged effect is an honest possibility, but it can’t be falsified, because any “we don’t see it yet” can be deflected with “just wait.” That makes it strong as explanation and weak as prediction.
The second is the overblown narrative. This camp holds that “AI jobs crisis” is largely a rhetorical wrapper for corporate restructuring: the real driver of layoffs is the post-ZIRP unwind after years of over-hiring, plus jobs moving offshore, with AI as the more dignified-sounding excuse. One practitioner on HN said flatly that for US software engineering, offshoring is still the bigger contributor to cuts. The strength of this read is that it explains why CEOs love to talk about AI layoffs: dressing cost-cutting up as technological progress plays better with the stock price and with morale. Its weakness is that it slides easily into conspiracy, writing off every real AI effect as PR.
The third, and the most durable, is the statistical blind spot. It doesn’t deny AI is doing something; it argues the macro average can’t see it. The densest pushback on HN all points here: a job opening isn’t a job, and “ghost listings” inflate the openings count; May’s gains skewed toward health care, a long-running trend driven by the aging of the baby boom generation that has nothing to do with AI yet papers over tech losses; and average wages are up, but you need the median and the percentiles to tell whether the middle is being hollowed out. These three aren’t a multiple-choice question. It’s most likely a lagged effect stacked on top of a statistical blind spot, then amplified or downplayed by whoever finds the narrative convenient.
Who’s right
Weighing the three together: the statistical blind spot has the most explanatory power right now; the lagged effect is a reasonable open question; and the overblown narrative is partly true but badly overused as a convenience.
The reason is the granularity of the evidence. Slok’s case rests entirely on aggregates, and aggregates are exactly where a structural shock is least likely to surface. A persuasive detail came from an HN commenter who runs a job-search site: he sees no crisis in overall openings, not even for software engineers, but a very clear signal that entry-level demand is under pressure. That’s eyewitness testimony for “totals fine, structure broken.” Several engineers seconded it: they’re now hiring essentially only people who are already senior, with junior openings near zero. If AI’s first wave concentrates on junior, directly-replaceable functions, it’s invisible inside that 172,000 figure, because the simultaneous gains in health care and services are enough to bury it.
The case shouldn’t be overstated, though; two points belong to the other side. First, wage data genuinely doesn’t support the hypothesis that masses of people are being squeezed into lower-paying roles: HN commenters citing BLS figures noted Q1 2026 median weekly earnings of $1,235 for full-time workers, up 3.4% year over year, and a rising median means there’s no obvious downward squeeze. Second, junior roles being hard to land isn’t new to the AI era; commenters recalled that hiring juniors was already tough in 2018 and 2023, so pinning the whole current difficulty on AI probably overstates its share. So the more accurate read: AI is nibbling on a small but specific slice of the employment pie, the junior coding, junior legal, and junior accounting work, and that bite is real but not yet big enough to leave a mark on the economy-wide curve.
Why it matters
This matters not because it predicts next year’s unemployment rate, but because it exposes a perception trap: we’re using a statistical toolkit built for the last technology shock to observe a change that may run along an entirely new fault line.
Concretely: if AI’s impact is highly function-specific, hitting juniors and certain standardizable white-collar tasks, then any economy-wide, all-industry average will systematically lag or miss it. By the time an aggregate like the unemployment rate finally moves, it usually means the structural problem has accumulated enough to spill out of its local pocket and into the whole system, and reacting then is too late. For policymakers, that means the thing to watch isn’t the monthly headline payroll number but the breakdowns by seniority, by function, and by age cohort, new graduates especially. For everyone treating “the macro looks fine” as a sedative, it’s a reminder: the calm you see may only be a function of how far back you’re standing.
The flip side is that this also matters for cooling the cry-wolf panic. Whatever its lens problems, Slok’s data proves one thing. As of mid-2026, AI has not triggered an aggregate, industry-sweeping employment tsunami. Claims that “traditional software jobs are already gone, no more need for developers or testers” don’t hold up against the numbers. The real picture sits between the two extremes: not the apocalypse, and not nothing, but a localized adjustment advancing precisely along specific functions, temporarily muffled by aggregate data.
What to ignore
The first thing to ignore is any framing that collapses the whole argument into “did unemployment go up.” The unemployment rate is an aggregate, lagging, coarse-grained measure: it doesn’t distinguish seniority, doesn’t capture underemployment, and can’t see someone sliding from one full-time job into two or three part-time ones. Using it to judge whether AI is acting is like taking blood pressure with a thermometer: the tool is fine, but it’s measuring the wrong thing.
The second thing to ignore is the false precision of the “job openings” number itself. HN commenters repeatedly noted that AI has made posting jobs nearly free, so ghost listings (roles companies post to look like they’re growing) and duplicate postings badly inflate the count. To be fair, the BLS JOLTS survey doesn’t count job ads; it asks firms directly, which is far sturdier than counting ads. But even so, decimal-point moves in a ratio like “1.0 openings per unemployed worker” are buried in statistical noise, and that little uptick arrow on the right edge of the chart isn’t a trend reversal worth reading.
The third thing to ignore is the headline-grabbing conclusion from either side, whether “no sign of workers being replaced by ChatGPT” or “the AI jobs crisis is already here.” Both are half right: the former holds in the aggregate and is blind to structure; the latter holds at the junior-role coalface and is overblown at the economy-wide scale. Pull either one out on its own as a verdict and you’ll be wrong.
Builder impact
If you’re hiring, this debate carries one directly actionable implication: don’t use macro data to underwrite your hiring decisions, in either direction. “The market’s flooded, easy to hire” and “the market’s tight, can’t find anyone” can both be true at once, and the difference is seniority. From the HN coalface, senior engineers are still in demand, with some founders complaining they can’t find qualified seniors; junior demand, meanwhile, is collapsing. That means if your team is still structured around “hire a batch of juniors and grow them into seniors,” you’re standing on the exact fault line AI is rewriting: what gets displaced by tools is precisely the junior layer’s output.
The harder reminder: shrinking junior hiring saves money short term and severs the industry’s talent pipeline long term, because no junior roles today means no promotable seniors in three to five years. One HN commenter put the bind coldly: not hiring juniors now “is going to be a huge problem eventually, but not my problem to deal with.” For founders who actually take the long view, that’s the contrarian opening. When everyone is using AI to displace junior output and giving up on developing people, the team that pairs senior judgment with tools to fill the junior gap can build a depth of talent its competitors can’t, at lower cost. The judgment isn’t about how strong AI is. It’s about whether you treat it as an excuse to cut, or a lever to amplify judgment.
Sources
No official primary source available; this analysis is based on reliable secondary reporting (named outlets, cross-confirmed).