Expertise-leveling is one of five categories of technological change identified by Daron Acemoglu, David Autor and Simon Johnson in “Building Pro-Worker Artificial Intelligence” (Brookings, 2026), alongside labor-augmenting, capital-augmenting, automating, and new task-creating technology. It describes a specific pattern: the task itself doesn’t disappear, and the specialist doesn’t lose their job, but the technology lets someone with far less training perform a passable version of the same work. Output goes up. The gap between a novice’s result and an expert’s narrows.
That’s what makes it harder to spot than automation. A layoff notice is legible; a flattening wage premium isn’t. An experienced professional can keep their job, keep being useful, and still watch the market’s willingness to pay a growing wage for their specific judgment stop rising, because the thing that made that judgment scarce is being deliberately engineered into something more broadly available. The authors call it ambiguous rather than negative precisely because it can raise a firm’s total output while simultaneously eroding the individual specialist’s bargaining position.
Of the five categories, the authors name only one, new task-creating, as unambiguously pro-worker, because it’s the only one that expands what a person can be paid to do rather than commoditising what they already do. Expertise-leveling sits in between: good for the buyer of expertise, ambiguous at best for the seller of it.