A two-factor model of AI growth pays the gains to whoever knows how the office works

Complementarity is a statement about prices. When two inputs are mostly complements and one of them becomes abundant, the other becomes dear. The post predicts precisely that: the shock to Intelligence raises marginal returns, employment and real wages in the Polanyi knowledge sector, because that sector's inputs are now relatively scarce against a larger quantity of Intelligence ^[7]. Substitution does not get you out of it. The post calls the two factors not quite Leontief complements but mostly complements ^[4], and says a dose of AI-drenched technocratic knowledge usually will not fix an office with norms problems and can make them worse by empowering rent-seekers further ^[5].

The second factor moves slowly, and the post says why. It cannot be boosted quickly or with direct and simple efficacy, and it is "messy by its nature" in a term the author takes from Luis Garicano. The wage and employment gains there arrive late and then keep coming for a long period ^[9]. Absorbing and exploiting the advances already made also takes a long time, with further Intelligence advances on tap ^[10]. So far, the post says, there is no way you can "bring an AI into your office and have it figure out how that office works" ^[3]. The argument runs at the level of an economy. The post puts no numbers on the shock, the lag or the gains ^[17].

The claimed fit with the data is the weakest link. Real wages and employment in the tacit-knowledge sector should rise, slowly, and keep rising ^[9]. Centaur pairings should begin to fade, as they did in chess, and that fading should show up as transitional unemployment in the Intelligence sector ^[8]. The post reports that Centaur models are still holding, and that mathematicians did the prompting for the new math work ^[8].

The model is older than the current capex cycle. The post dates it to a debate the author had at seventeen, prompted by the Soviet preoccupation with cybernetics, central planning and possible supercomputers, and answers that older question with a no ^[15].

I would take the direction of the complementarity claim and treat its degree as the open question. The counter-thesis is entry. A firm built after the shock has no existing office to convert, pays nothing for absorption, and would post the gains early. On that reading, the slow catch-up described in the post is a story about incumbents rebuilding process while the tacit sector's wages climb ^[14]. Either way, the post concedes that the Solow model usually does fine. The Intelligence sector rarely swings up this fast, and the ratios between the two sectors are normally fairly constant in the short run, though not in 2026 or the years to come ^[12].