When Anthropic CEO Dario Amodei said in May 2026 that "if you automate 90 percent of the job, then everyone does the 10 percent of the job," he captured a central tension in the future of work, according to a new guest post published by the Economic Innovation Group as part of its American Worker Project. The analysis, written by Joshua Gans, an economics professor at the University of Toronto and chief economist of the Creative Destruction Lab, examines what happens to human work when AI automates most—but not all—of a job's tasks. The report argues that the answer depends less on how many tasks disappear than on which tasks remain and how they relate to each other.

The report doesn't present new survey data but instead synthesizes existing economic frameworks to show how different types of task automation produce opposite outcomes. According to Gans, when AI removes routine bookkeeping from an accounting clerk's day, the remaining activities—exceptions, problem-solving, client communication—become more demanding, potentially raising wages while reducing the number of workers who qualify for the role. Conversely, when technology removes the most demanding diagnostic or administrative activity from an occupation, the remaining role becomes accessible to a broader pool of workers, which can increase employment while lowering wages. The report examines several mechanisms through which automation reshapes jobs: a "displacement effect" when capital takes over tasks previously done by people, a "reinstatement effect" when innovation creates new tasks where humans have an advantage, and a "focus mechanism" when automating one activity frees attention for others that determine service quality.

"The answer will reflect displacement, reinstatement, and demand. It will also depend on the job's internal structure," the report states. Gans emphasizes that tasks can act as substitutes, complements, bottlenecks, filters for expertise, or objects of worker motivation, and how a job changes after automation depends largely on how tasks relate to each other. The report warns that "two occupations with the same average exposure may look quite different once we see whether the exposure is concentrated in a few tasks or spread across the whole bundle." Drawing on economist Michael Kremer's O-ring model, Gans explains that when production depends on several essential activities, a weak link reduces the value of the whole effort—meaning that automating auxiliary tasks can make the remaining human bottlenecks more valuable, not less.

The report identifies a blind spot in standard analysis: payroll records can miss major changes in job quality because workers sometimes expand effort on tasks they find rewarding without additional pay. A software engineer who uses AI to eliminate documentation and debugging might devote the freed time to refining a tool they enjoy, leaving wages unchanged while the job's character transforms entirely. Gans argues that turning exposure measures into forecasts requires asking six questions about any occupation: which tasks are exposed, what expertise remains, whether freed attention improves quality, how firms will redraw job boundaries, whether demand will expand, and what the data miss. For policymakers and business leaders, he writes, "a useful discipline is to keep asking questions after an exposure estimate is released"—because an exposure measure is a starting point, not an answer. The final ten percent may become more valuable, more demanding, easier to enter, or newly divided among specialists, depending on whether automation removes bottlenecks or routine work.