Many predictions about AI and employment rest on a flawed assumption: that tasks and jobs are interchangeable. The simple truth is that they are not.
A job is rarely a single activity. Most combine technical work, communication, judgment, and coordination. Yet many forecasts about AI and employment miss this crucial point. Understanding this distinction between a task and a job is essential for assessing which jobs are genuinely vulnerable and which are likely to remain resilient.
This framing has shaped much of the analysis around AI's labor market impact. Researchers measure the extent to which AI can perform a job's component tasks and then draw conclusions about the job's future viability.
As Luis Garicano, professor of Public Policy at the London School of Economics and co-author of Messy Jobs: The Work That AI Cannot Reach, said, "The task is not the job." In fact, Garicano argues that exposure scores tell us relatively little about whether a role will be replaced because they ignore how tasks interact within a given position.
Consider radiology. Many predicted AI would make radiologists obsolete because it could read medical scans, yet the demand for radiologists continues to increase. While AI can interpret a scan, it cannot handle consultation, clinical liability, patient history, or care coordination. Automating one task does not erase the rest of the job.
Automating a Task Versus Replacing a Role
Automating a task is not the same as replacing a role. When a job combines cognitive, relational, and physical responsibilities, AI may remove or accelerate certain activities without eliminating the need for the person performing the broader role.
This distinction carries important implications. AI may significantly reduce time on the task, but if that task is embedded in a strong bundle of responsibilities, the human role does not disappear. In some cases, it may even expand. Garicano links this to Jevons' Paradox, which suggests that when the cost of a productive input falls, demand for the final output often rises.
Applied to radiology, if AI drives the cost of scan analysis toward zero, the economically rational response may be to scan patients more frequently. The result could be more work for radiologists, not less, because the profession encompasses far more than reading scans.
Why Are Some Jobs More Durable Than Others?
Garicano describes the most resilient roles as "messy jobs," where relational, physical, and cognitive demands are inseparable, and in which context accumulated over time is essential to performance quality.
Sales roles offer a clear example. Success depends on understanding a client’s business, identifying opportunities, and managing relationships. AI may assist with parts of the process, but automating a single element does not constitute a meaningful substitute for the entire role.
Conversely, clean, single-task roles face a greater risk of direct displacement. Where tasks are clearly defined and can be performed independently, AI has increasingly demonstrated the ability to perform at or above human levels.
For those evaluating AI's impact on workforce composition and productivity, the more useful question is not "How much of this job can AI do?" but rather "Can AI replace this role as a whole?" Strong bundles indicate resilience. Weak bundles, in which tasks are modular and independently executable, face a greater risk of substitution.
The Long-Term Analysis
The competitive advantage in the AI transition may not accrue primarily to the firms building large language models. It may accrue to those who can navigate the operational complexity of implementation, reassigning tasks, restructuring roles, and preserving the institutional knowledge embedded in messy, bundled work.
For long-term decision-making, the key challenge is identifying where human judgment, relationships, and experience continue to create value. The organizations most likely to generate durable productivity gains from AI are those that understand which roles to preserve, which to restructure, and which human capabilities sit beyond the frontier of current automation.
For more information on related investments and insights, please listen to our William Blair Thinking podcast, Monthly Macro: The Work AI Cannot Reach, recorded on July 15, 2026, featuring William Blair macro analyst Richard de Chazal, and Spanish economist and former Member of the European Parliament, Luis Garicano, Ph.D.