I believe that IT workers who are also passionate about gen AI are making a major misjudgment. We wrongly assume that the advances we observe in our field, such as the autonomous or semi-autonomous development of applications, also translate to sectors like medicine or law. This is a false generalization.

The field of IT heavily relies on strict formalism: the raw material consumed by LLMs. In the legal field, for example, this is not the case: it is much more complex. Laws, regulations, and judgments are generally written and presented in standardized forms, but the content is far from being as digestible formalism as lines of code written in a programming language. In my opinion, we should remember that when we share our enthusiasm for gen AI. We must be lucid while also setting the right expectations for decision-makers and lawmakers.

Margaret-Anne Storey introduces “cognitive debt” as a concept that may be more threatening than technical debt in AI-augmented development. Unlike technical debt (which lives in code), cognitive debt is the erosion of shared understanding that resides in developers’ minds. Drawing on Peter Naur’s concept of a program as a “theory” distributed across teams, the article argues that as AI and agentic tools push for development velocity, teams risk losing their collective understanding of why systems work the way they do. Even if AI generates technically clean code, teams can become paralyzed when no one can explain design decisions or anticipate the consequences of changes. The author calls for intentional slowdowns, collaborative practices, and serious research into measuring and mitigating this growing challenge.

“As generative and agentic AI accelerate development, protecting that shared theory of what the software does and how it can change may matter more for long-term software health than any single metric of speed or output.”