When the Frontier Moves, the Bottleneck Moves
The OECD exposure measure becomes more informative when its AI frontier is treated as a variable rather than a permanent fact. This chapter examines the report’s illustrative scenario in which cognitive AI capabilities rise by one level, then tracks which occupational gaps narrow and which remain. The purpose is not to predict the next model release. It is to show how selective progress changes the exposure map—and why even an updated map cannot, by itself, tell us how a workplace will redesign tasks.
A static ranking invites a static conclusion: occupations near the frontier seem exposed, and occupations farther away seem protected. But the frontier does not advance as one block. Language, problem solving, social interaction, manipulation, and robotic intelligence can improve at different rates. If only some capabilities move, the exposure of an occupation depends on which of its requirements were binding constraints in the first place.
What the one-level scenario changes
The OECD demonstrates this by changing one input to the calculation. Imagine that AI’s cognitive capabilities increase by one level while the occupational profiles and the other capability levels remain as they are. The index is recalculated under that assumption. This is a controlled scenario: it asks, “What would the gap look like if this specified capability improvement occurred?” It does not assign a probability to the improvement, specify a product, or assume that employers will adopt it.
The result is a movement in the map rather than a universal shift. Office and Administrative Support, already the group with the smallest current gap, falls to 0.2 in the scenario. Computer and Mathematical, Sales and Related, Business and Financial Operations, and Architecture and Engineering fall into a range from 0.8 to 1.1. Management and Life, Physical and Social Science also move closer to the frontier, with remaining gaps of 2.2 and 1.9 respectively.
The mechanism is visible in the capability profiles. These occupations rely heavily on language, codified knowledge, analysis, problem solving, and other cognitive activities that the scenario changes directly. When those capabilities improve, the distance between the occupation’s requirements and AI’s assessed performance shrinks. The resulting index says that current AI would be closer to the measured profile under the assumed trajectory.
The size of the movement should be read as a conditional change, not as a forecast of equal change in every workplace. An office-support role may include a large volume of structured records, routine correspondence, or repeatable classification, so a cognitive improvement can bring much of the measured profile close to the frontier. Yet the practical result could be automation of a narrow process, augmentation of an existing worker, a shift toward exception handling, or no deployment at all. The index does not observe the organisation’s data quality, procurement choices, error tolerance, or division of responsibility.
Computer, engineering, and business roles illustrate a different point. Their scenario gaps become relatively small, but their work may still involve selecting goals, reconciling conflicting constraints, checking outputs, and carrying accountability. A smaller gap means that more of the measured capability profile is technically within reach; it does not mean that the occupation’s decisions can be delegated without review.