Research: AI 'Unbundling' Jobs into Narrower, Lower-Paid Tasks

A research paper from Luis Garicano (London School of Economics), Jin Li, and Yanhui Wu (both University of Hong Kong) challenges the assumption that AI exposure directly leads to job loss. Instead, they argue AI is 'unbundling' jobs by automating specific tasks within roles.

Weak vs. Strong Bundles

The paper distinguishes between 'weak-bundle' and 'strong-bundle' occupations. In weak-bundle jobs, AI automates some tasks and narrows the job's boundaries. Examples include support ticket processing or predictable coding tasks. The human is left doing whatever the machine can't handle, often resulting in a narrower slice of the original role.

In strong-bundle occupations, AI improves performance within the job but doesn't remove the human from the bundle. The authors use radiologists as an example: they don't just read scans but also interpret edge cases, consult with clinicians, and sign off on decisions.

Economic Impact

When AI takes over part of the work, humans focus exclusively on remaining tasks, increasing output per worker. This leads to falling prices and reduced demand for workers. The employment impact comes not from AI doing the entire job, but from humans becoming more efficient at the leftover tasks.

The research suggests this explains why employment and hours haven't shifted dramatically despite AI adoption. In many cases, the job bundle remains intact.

Implications for Developers

For those in weak-bundle coding roles involving predictable tasks, AI may gradually hollow out the role. For developers in strong-bundle positions requiring judgment, context, or responsibility, AI is more likely to enhance performance and potentially increase compensation.

The paper contrasts with forecasts predicting 10.4 million US job losses by 2030 (roughly 6% of the workforce), suggesting the reality is more nuanced than simple replacement.