AI is not killing jobs wholesale – it is quietly chipping away at them, one task at a time. The concept of job unbundling pushes back against the narrative that AI exposure automatically means fewer jobs, arguing instead that the real impact is the decomposition of roles into component tasks, with AI handling some while reshuffling who does the rest. ¹⁾

A March 2026 research paper by Luis Garicano (London School of Economics), Jin Li and Yanhui Wu (University of Hong Kong) argues that the real question is not how many tasks a model can do, but whether those tasks can actually be split out without breaking the role. ²⁾

Jobs are not neat lists of tasks – they are bundles. Radiologists, for example, do not just read scans. They interpret edge cases, communicate with clinicians, and sign off on decisions people act on. Replace the image-reading task and you have not necessarily replaced the job.

The researchers draw a critical distinction: ³⁾

• Weak bundles: Roles where tasks can be split apart without much friction. These jobs are vulnerable to unbundling because their component tasks do not depend heavily on each other.
• Strong bundles: Roles where tasks are tightly interdependent. Splitting them apart degrades the quality or safety of the work. These jobs are more resistant to AI disruption.

The implication: AI does not replace jobs uniformly. It selectively automates the separable parts, leaving behind a transformed (and often diminished) version of the original role.

In most organizations, AI adoption follows a predictable pattern. It gets deployed where work already looks like a queue: documents to draft, tickets to triage, emails to respond to, code to refactor, spreadsheets to reconcile, calls to summarize. ⁴⁾

When that happens, the job stays on the org chart, but its internal composition changes rapidly. Research on large language models estimated that around 80% of the U.S. workforce could have at least 10% of their tasks affected by AI. ⁵⁾

A practical example: a marketing Content Manager role in 2025 fragments into prompt engineering, AI output fact-checking, image curation, SEO polishing, and compliance review – each task potentially outsourceable at a fraction of the full-time salary. ⁶⁾

The concern is not job extinction but job hollowing: roles get narrowed to their least automatable components, and pay adjusts accordingly. The paper argues that the real risk is narrowing human work and compressing wages for the remaining tasks. ⁷⁾

This creates a paradox: the job title persists but the accountability stays with the human while the value-generating tasks migrate to AI systems.

• Hiring shifts from roles to tasks: Organizations increasingly define work at the task level rather than the job level
• The gig economy expands: Unbundled tasks are well-suited to freelance and contract arrangements
• Skill premiums change: Tasks that are hard to unbundle – judgment, stakeholder management, physical dexterity – command higher value
• Career development evolves: Workers must manage their portfolio of tasks and skills rather than climbing a single job ladder

• How the Automation Cliff Is Driving a Blue-Collar Renaissance
• AI Software Factory
• Core Limitations of Standard Off-the-Shelf AI Chatbots