Belief in AI displacement is outrunning healthy employment data, Noahpinion argues

A headcount plan needs a premise about which tasks go away and when. What this record supplies instead is a premise about expectations: Ipsos finding that most Americans expect AI to compete with human workers more than complement them, and Pew finding that expectation has strengthened in recent years ^[2][3]. Sentiment series tell you what people anticipate, not how many roles went away, the number a hiring freeze is implicitly forecasting.

On the count itself, the evidence in the post runs the other way. The prime-age employment rate, which Noahpinion calls the single best indicator of how many Americans have jobs, continues to hover near all-time highs, and the aggregate labor market is described as about as healthy as it has ever been ^[14][1]. The post also notes that most other forces acting on employment right now, tariffs and the Iran war among them, push in the negative direction, which makes an offsetting positive shock hard to name ^[8]. Its candidate for that offset is AI itself ^[9].

Its mechanism comes from Acemoglu and Restrepo (2019), which separates a displacement effect from a productivity effect and a reinstatement effect, the last being new tasks in which labor has a comparative advantage ^[6]. Two of those three channels add to labor demand and one subtracts from it ^[7]. Software engineering is the concrete version offered: engineers write less code themselves, spend more time directing AI, and then take on the work of deciding what code to ask for, checking that it is the code they wanted, and integrating it into products ^[10]. Power looms creating technician and engineering work, and the internet automating travel agency while creating web design, are the historical rhymes cited ^[12].

The aggregate is too coarse and too slow to be worth much to a functional plan, because the prime-age rate can sit at a record while the entry-level funnel in one occupation quietly closes. This record cannot answer that: the evidence here is aggregate and attitudinal, with no occupation-level or company-level hiring series ^[13]. What the aggregate does do is set the burden of proof. Economy-wide displacement is not yet visible in the best available headline measure, so a plan that assumes it is already under way inside one function is relying on the narrative rather than on a measurement.

The tradeoff is specific, and it has a sequence. Cutting on the displacement premise books a saving this quarter from the same population that would be doing the specification, checking and integration work the post identifies as the new tasks ^[10]. Rebuilding that population, if the reinstatement channel is the one that dominates, takes several quarters of hiring and ramp, at whatever the market price is then rather than now. The counter-case is in the record too: Acemoglu himself assumes AI's new tasks will be bad ones, misinformation and cybercrime, which hurt the economy rather than help it ^[11]. So whether "new tasks appear" argues for holding headcount depends on whether those tasks turn out to be good ones, and that is precisely what is in dispute.

That leaves the defensible position narrower than the board deck on either side. AI vendors have spent years telling the market their products will render large swathes of humanity economically obsolete, and that pitch has become the assumption their customers plan against ^[5]. Neither that pitch nor a near-record employment rate can settle what happens to a specific function, which is why the plan that survives scrutiny this quarter is the one justified by the firm's own hiring and output numbers.