Does automating some tasks make other human tasks more valuable?

Autor's 2015 essay asks why "technological change hasn't already wiped out employment for the vast majority of workers" given that labor-saving technologies keep multiplying. His answer: "tasks that cannot be substituted by automation are generally complemented by it." He draws on Kremer's 1993 O-ring production function, "an iconic representation of this idea," in which "failure of any one step in the chain of production leads the entire production process to fail" and, conversely, "improvements in the reliability of any given link increase the value of improvements in all of the others." Applied to work: "when automation or computerization makes some steps in a work process more reliable, cheaper, or faster, this increases the value of the remaining human links in the production chain."

As evidence, Autor cites Bessen's (2015) account of automated teller machines: the US ATM count "quadrupled from approximately 100,000 to 400,000 between 1995 and 2010," yet "US bank teller employment actually rose modestly from 500,000 to approximately 550,000" over the 1980-to-2010 period (though tellers fell as a share of total employment given labor-force growth). Bessen attributes this to "two forces work[ing] in opposite directions." Autor sets this against a longer sweep: agriculture's share of the US workforce fell from 41 percent in 1900 to 2 percent in 2000 (Autor 2014), and Nordhaus (2007) estimates the cost of a standardized computation fell "by at least 1.7 trillion-fold since the manual computing era," without employment collapsing economy-wide.

Autor's essay is the historical and conceptual root both Does automating some tasks actually increase the value of remaining work? and Does automation raise or lower the skills that remaining work demands? build on and formalize. Where Gans and Goldfarb give the "focus effect" a Nash-bargaining model with no empirical calibration, Autor supplies the concrete historical case — the ATM and bank teller data — that the later formal model is built to explain; where Autor and Thompson isolate the expertise channel specifically, Autor's 2015 statement is the broader claim that any task multiplying rather than adding to output creates this dynamic, not only tasks bundled by expertise. The essay also anticipates Does automating labor concentrate power more than augmenting it?: both treat "substitute" and "complement" as the two channels automation can run through, though Autor describes complementarity arising mechanically from production structure, not a political-economy trap from choosing automation over augmentation.

Autor's own essay offers no formal model or quantitative estimate of how much complementarity offsets substitution in general — that is left to the later papers above. Its final section, speculating about "recent and future advances in artificial intelligence and robotics," was written in 2015, before transformer-based LLMs or agentic AI systems existed, so its predictions about occupational polarization slowing concern robotics-and-software-era automation, not language-model agents. The implication the essay supports is a general one: any account of AI's labor effects that only counts tasks removed, without asking whether the remaining tasks were quality complements, is likely to overstate displacement — a claim the later, model-based notes above make precise.