OECD maps AI’s biggest job risks but LPL’s chief economist sees potential upside

A new analysis from the OECD has identified which workers face the most immediate displacement risk from artificial intelligence… and for a broad swath of the clerical and administrative workforce, the answer is: now.

The report introduces what researchers call an AI Capability Gap Index, mapping the distance between what roughly 900 occupations require across nine capabilities and what current AI systems can actually do. The smaller the gap, the higher the exposure. For office and administrative support workers, that gap has effectively closed.

Billing clerks, word processors, file clerks, bookkeeping and auditing clerks, and data entry keyers all score at or near zero on the index, meaning current AI capabilities already match or nearly match what those jobs demand.

The occupational group as a whole records an overall gap index of just 0.8, the lowest of any major category in the analysis. Production occupations follow at 2.0, with food preparation and serving at 2.5 and sales at 2.6.

At the most protected end sit chief executives, ophthalmologists, firefighters, psychiatrists, lawyers, judges, and anesthesiologists — occupations whose full capability profiles combine contextual judgment, social understanding, physical dexterity in unpredictable environments, and complex decision-making under accountability in ways current AI cannot replicate.

Community and social service occupations record the highest overall gap index at 6.4, followed by legal occupations and educational instruction, both at 5.8.

The OECD is careful to frame its index as a measure of potential exposure rather than a direct forecast of job losses. Actual displacement depends on adoption costs, organizational capacity, regulation, and social choice, the authors note.

A zero capability gap does not automatically mean a job disappears. But if AI's cognitive capabilities advance by even one level, exposure extends well beyond clerical work into computer and mathematical occupations, business and financial operations, and parts of management and the life sciences.

Not all negative

Not everyone reads that trajectory as straightforwardly grim. LPL Financial chief economist Dr. Jeffrey Roach, writing in a commentary published late last week, argues that AI's labor-market effects may ultimately prove more expansive than destructive.

Roach draws on a 19th-century economic principle to make the case. The Jevons paradox holds that when technology makes a resource more efficient to use, total demand for that resource tends to rise rather than fall, because lower costs unlock new uses and draw in more customers.

Roach points to the April payroll report, released May 8, as a live illustration: diagnostic imaging centers, widely expected to shed workers as AI handles more of the analytical load, are instead hiring to meet a surge in demand driven by lower service costs. Bookkeeping, by contrast, has seen declining demand — a split that maps neatly onto the OECD's own findings.

Roach's broader argument is that making tasks cheaper and faster doesn't necessarily eliminate the workers performing them. It may instead expand the total volume of work organizations take on, generating new roles and business models in the process. The premium, he argues, will rise for workers who can deploy AI effectively, refine workflows, and apply human judgment where automation falls short.

He also situates the technology within a longer demographic argument: working-age populations across developed economies are projected to shrink as a share of the total through 2070, and AI may be one of the few mechanisms available to offset the resulting drag on output and tax revenue without relying on a workforce that isn't growing. For investors, that framing connects AI adoption directly to long-run profitability in an era of scarcer, more expensive labor.

Uneven impact

Where the two analyses converge is on the point that AI exposure is not uniform.

The OECD report finds that some occupations face pressure primarily from language and reasoning systems, while others are threatened by robotics and machine vision and the retraining required to move workers out of each category is correspondingly distinct. A data entry keyer and a warehouse operative may both register as highly exposed, but what threatens them looks entirely different.

What neither report resolves is the pace of the transition. History suggests that technological reallocation eventually produces net employment gains. It also suggests the workers most directly displaced rarely capture those gains themselves.

The OECD report is The OECD AI exposure measure: Mapping the OECD AI Capability Indicators to occupations.