JUDGMENTLast week Lengshui pointed to "two crossover points" in AI layoffs — the first (the month compute spend overtakes the total pay of cut staff) some firms have already hit; the second (the quarter when rehire rates pass pre-layoff pay) he places in Q2 2027. Seen on a two-century scale, that second point is actually the third "automation rehiring" cycle since the Industrial Revolution. But this round is structurally different from the first two. It is not a simple replay.

FACTThree long-cycle datasets sit side by side. One: the WEF Future of Jobs Report 2023 notes technological change usually "creates more than it destroys" within 5–10 years, but the new roles' skill mix is badly mismatched with the eliminated ones — a curve not unique to AI; the 1980s PC adoption in enterprises traced the same shape. Two: U.S. BLS data show that across 1980–2000, computerized America net-added roughly 18 million "computer jobs" — programmers, systems analysts — far outpacing the clerical and operative roles lost in the same span. Three: across 2024–2025, Anthropic's and OpenAI's careers pages and official blogs keep surfacing roles that did not exist in any 2022 job catalog: AI output evaluator, prompt-structure engineer, alignment researcher, AI behavior red-teamer. All sources are publicly accessible reports and official pages.

SPECULATIONI name this round the "third wave of rehiring." Against the 1900s industrial line and the 1980s computerization, it differs in three structural ways. First, the earlier rounds substituted muscle and routine paperwork; most new roles fit inside the old labor market. AI substitutes cognition, and its new roles — evaluation, alignment, prompt engineering, behavior testing — mostly sit outside traditional function boundaries; they are job types that grew from inside the AI system. Second, the earlier rounds had geographic return (blue-collar back to blue-collar towns); AI rehiring clusters within 50 km of San Francisco, Seattle, Beijing's Haidian, London's King's Cross — a sharper geographic mismatch. Third, and most decisive: the earlier cost-return curves were roughly linear, while the AI curve is already distorted by both the compute bill and capital-market expectation. The junior devs and testers who were cut now cost, per converted hour when rehired, more than their pre-cut annual salary — pricing detached from labor economics. Lengshui's Q2 2027 second point is, at its core, the moment that curve slams into the hard ceiling of the hiring budget.

OPINIONYuanwang ignores corporate press releases and watches the one rule across two centuries: every mass technological substitution triggers a rehiring — but the cost structure and shape of that rehiring are set by the intrinsic nature of the substituting technology. AI's nature is "compute-dense + scarce training data + fuzzy behavior boundaries," so the winners of this round are not "people who can write prompts" but those who can impose constraints across three lines at once — compute cost, organizational budget, behavioral accountability. That role has no department in any university and no fixed shape on any job board, yet it is already debated in earnings calls as of August 2026. The second crossover point is not an ending. It is the opening price of a new labor market.

---

Further Reading