ISFB Insight
AI and Deskilling: The Downside of Productivity Gains
October 30, 2026
In the age of generative AI, an organization’s strength lies in its ability to determine what it refuses to unlearn in order to make room for learning on the job.
By Dr. Mathias Baitan, Executive Director of the ISFB
Artificial intelligence is already improving the speed of execution and, in many tasks, the quality of the work produced. However, a more subtle risk is emerging: becoming more efficient with the tool while gradually losing the ability to work without it.
This is what’s known as deskilling: the erosion of a skill when it is no longer practiced enough.
A working paper published in September 2026 by David Autor and his co-authors presents an interesting finding. Over a three-month period, 133 patent attorneys were randomly assigned to either a group equipped with an AI assistant or a control group. The tool improves the quality of work, particularly among junior employees. However, when all these attorneys were then required to exercise their judgment without the tool, only the most experienced professionals continued to benefit. Among junior attorneys, performance varied widely rather than improving. The authors therefore formulated a hypothesis: solid prior expertise may be necessary to truly learn from AI, rather than simply producing better results thanks to it (Autor et al., 2026).
This question builds on several ideas already explored in the pages of ISFB Insight. Last March, my article“Decision-Making: A Complex Skill at the Heart of the Managerial Role” noted that makingdecisions involves interpreting a situation, accepting a degree of uncertainty, and exercising judgment. In April, in another article titled “In the Age of AI, the Finance Industry Needs Well-Trained Specialists More Than Ever, ” I had already highlighted a risk: that of confusing the speed of access to an answer with mastery of the subject. Then in June, in my post“When Machines Enter the Realm of Language,” I drew another line: as information becomes easier to obtain, the ability to understand it, verify it, and assign value to it becomes all the more important.
The ISFB Observatory’s initial findings, published this summer, point in the same direction: banking jobs appear to be evolving rather than disappearing, while the expected skill level is rising.
Deskilling presents yet another challenge for senior management.
A bank may increase its productivity today and, without realizing it right away, weaken its pool of experts for the future. An analyst builds expertise by analyzing. A risk specialist builds it by testing hypotheses. An advisor builds it by preparing recommendations, presenting arguments, and receiving feedback. If these tasks are gradually phased out during the early years of a career, part of the process through which experience is built disappears along with them.
The issue therefore goes far beyond the use of AI. It touches on succession planning, operational risk, decision-making quality, and accountability. Management teams will need to gradually distinguish between tasks that are worth automating and those that an organization must continue to perform on its own.
Because not all friction is pointless. Some of it is precisely what builds expertise. Performance will therefore certainly not be measured solely by the number of tasks entrusted to AI—quite the contrary. A strong organization will be one that knows what it refuses to unlearn in order to make room for on-the-job learning.
References
Autor, D. H., Rodchenko, T., Martin, J., Iscenko, Z., Strand, S., Pearl, D., & Ferere, M. (2026). Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting. NBER Working Paper No. 35720, doi: https://doi.org/10.3386/w35720
© Institut Supérieur de Formation Bancaire (ISFB). All rights reserved.
The analyses and content published by the ISFB may be quoted or reproduced in part, provided that the source is clearly mentioned. Any full or substantial reproduction of this article in another medium or format is subject to the prior written authorization of the ISFB. In order to facilitate reading and without any intention of discrimination, the masculine gender is generally used, in accordance with the grammatical rule that allows it to be used as a neutral value to refer to a group of people comprising both men and women. This publication is intended for ISFB members and their employees in Switzerland, as well as anyone interested in finance in Switzerland. It is not intended to be read or distributed in any jurisdiction where its distribution would be prohibited.
