AI's first labour-market mark is a pay gap, and it sits below the top wage quartile

The gradient is the finding, not the average. Apply one blended figure across a workforce and you under-provision the bottom quartile by four percentage points of wage growth while over-provisioning the top by the entire gap, because Apollo reports no significant effect there at all ^[16][1][12]. The bottom quartile's shortfall runs roughly 2.7 times the third quartile's ^[13]. That is a pay band problem with a known address, not a hiring forecast.

Diane Gherson's accounting point explains the shape better than any strategy document. Severance can be booked as a one-off restructuring charge that investors discount, while retraining lands in operating expenses every quarter ^[7]. A finance function managing reported earnings will prefer the cut, and where cutting is awkward it will prefer not replacing. Neither route produces an announcement, and neither reads as a wage decision, but the residue collects in pay.

Worth stating what the study cannot carry. Exposure is measured from observed usage inside one company's product, via the Anthropic Economic Index, rather than from a theoretical exposure score ^[2]. Only 321 of roughly 800 occupations were matched, about 40% of the map ^[8][14]. The authors' largest number, a 24.3% gap for service workers, they flag themselves as a small subsample to treat with caution ^[9]. Direction travels further than magnitude here.

The counter-evidence is not really a rebuttal. US statisticians counted a 0.2% job fall across 18 exposed occupations against 0.8% payroll growth overall, a one-point spread ^[4][15], and Goldman Sachs reported openings declining faster in fields exposed to substitution, in a market already squeezing new entrants ^[5]. Openings data reads the door; wage data reads the people already inside. Both can hold, and for anyone doing retention planning they point at the same cohort.

The one worked alternative in the material is European: Ikea retrained call centre staff as remote interior design advisers after automating much of their previous work, and the resulting service has been widely reported as a business worth around 1.3bn euros ^[17]. That is Gherson's chosen counterexample, not a controlled comparison. Against it sits a concession from Apollo's chief economist Torsten Slok ^[3], who points to record business formation ^[10] while acknowledging that margins outside the largest technology companies have not yet risen ^[11]. On Apollo's own numbers the pay gap is measurable and the productivity gain is not. Europe has no equivalent study, and this channel would not show up in most of its labour data ^[18].