Task Exposure, Not Job Loss

The most confident claims about technology and employment are made about the future. The claims that can be checked are about the past, and they are less dramatic.

An exposure index answers a hypothetical question: if a technology could perform a task, how much of an occupation would it cover? This is useful analytical scaffolding. It is not a prediction, and it has repeatedly been reported as one. Whether exposure becomes substitution depends on reliability requirements, liability structures, regulatory constraint, capital availability, organisational capacity for change, and relative prices — none of which an exposure index measures.

What we looked at

We assembled labour force survey microdata for seven economies from 2019 to the first quarter of 2026, harmonised occupational wage series, and a corpus of 41 million job postings. The postings corpus is the most informative source, because it captures changes in what employers ask for at a granularity that survey data cannot reach and with a lag of weeks rather than quarters.

We asked three questions of this data. Has employment in high-exposure occupations fallen relative to low-exposure occupations? Has the task composition within occupations changed? And have wages moved in a pattern consistent with substitution?

Employment: no detectable aggregate effect

The first answer is the clearest. Across all seven economies, employment in high-exposure occupations has not fallen relative to low-exposure occupations by any margin that survives controls for sector, firm size, and macroeconomic conditions. Some high-exposure occupations grew. Some shrank. The pattern does not sort by exposure.

We want to state the limits of this finding carefully. Three years is a short window for a diffusion process. Absence of a detectable effect is not evidence that no effect exists, particularly given the known weakness of employment statistics at occupational granularity. But it does constrain the space of claims that can currently be supported, and several widely repeated claims fall outside that space.

Composition: real and concentrated

The second question yields a more interesting answer. Within three occupational families — software development, customer operations, and structured document work — the composition of advertised tasks has shifted measurably. Postings ask less often for the tasks these systems perform well and more often for review, judgement, integration and exception handling.

The shift is sharpest at entry level. Junior postings in the affected families changed more than senior ones, both in the tasks listed and in the experience requirements attached. This is the finding we hold with the most confidence and consider the most consequential, because entry-level task composition is the mechanism by which occupations reproduce their skilled workforce.

Wages: dispersion, not decline

Median wages in high-exposure occupations did not fall. Dispersion within those occupations widened: the upper part of the distribution moved up relative to the lower part. This is the pattern one expects from task reallocation — where the tasks that remain are those with higher returns to skill — rather than from displacement, which typically shows as a downward shift in the whole distribution.

What we would need to change this assessment

• A sustained divergence in employment growth by exposure quintile persisting for four or more consecutive quarters.
• Evidence of falling entry-level hiring in the affected families that is not explained by macroeconomic conditions.
• A downward shift in the full wage distribution rather than a widening of dispersion.
• Firm-level evidence connecting deployment intensity to headcount decisions, which we currently lack at scale.

The full paper includes the harmonisation methodology, the exposure index comparison in detail, robustness checks under alternative occupational classifications, and country-level results. Pooled results are reported here for brevity and conceal meaningful cross-country variation.