Absorption: How AI Reroutes Junior Work
The software-engineering case makes the upward redistribution of work concrete. The focus is the mechanism the researchers call Absorption: how generative AI expands a senior engineer’s capacity so routine work that once went to a beginner is handled directly by the senior and an AI tool. Tracing that sequence explains how a team can report no immediate productivity loss while quietly weakening the path into senior competence.
Absorption is a claim about task allocation, not a claim that an occupation vanishes on its own. An organization decides who receives a task, who bears the cost of explaining it, and who is accountable for the result. AI changes the economics of that decision. It can make routine implementation, initial debugging, and documentation cheap enough for an experienced engineer to perform directly. The consequential question is what happens to the person who would previously have learned by doing that work.
From delegation to senior–AI workflow
Before the widespread use of generative AI, a junior engineer could enter a team through work that looked unimportant from a distance: implement a small change, fix a minor defect, write a test, investigate a failure, or document what the system does. These assignments had a production purpose, but they also placed the beginner inside the team’s real operating context. A senior decided what could safely be delegated, the junior made an imperfect attempt, and review turned that attempt into a shared understanding of the code and the team’s standards.
The process was not free. The senior had to specify the task, answer questions, inspect the result, and sometimes repair mistakes. A beginner might take longer than an experienced engineer, and the first version might not be the best version. Yet the additional time was not only a cost of production. It was also the cost of transferring context and building a person who could later work with less supervision.
Generative AI changes the comparison by giving the senior another way to absorb the task. Instead of translating the assignment for a junior and paying the review cost, the senior can ask an AI system to generate code, suggest a fix, or produce documentation. The senior already knows enough about the surrounding system to judge whether the result is plausible, revise it, and accept responsibility for shipping it. The junior is not necessarily replaced by a machine in a direct one-for-one exchange. The junior’s opportunity is removed because the senior–AI combination becomes the preferred route for completing the work.
This is the sequence at the heart of Absorption:
The mechanism is easy to miss because the visible result can look positive. A senior completes more work. The team ships on time. The generated output may be good enough after review. If the junior headcount is reduced and productivity does not fall, the organization receives a strong signal that the old allocation was economically replaceable. That signal answers a narrow question—whether the current output can be maintained with fewer junior assignments. It does not answer whether the assignment was building capability that the organization will need later.