AI Use in Structural Engineering Practice, and the Preservation of Engineering Judgment

To view the figures and tables associated with this article, please refer to the flipbook above.

Discussions around artificial intelligence use in structural engineering practice have largely centered on how engineering work can be reorganized around automation, workflow optimization, and decomposable tasks—raising deeper questions about the role of professional judgment in that transition. These discussions often present AI integration as efficient, rational, and inevitable. That convergence is worth pausing on because of what it quietly assumes about the nature of engineering work and professional judgment. These discussions often focus on automating or restructuring tasks such as load calculations, connection design checks, and code compliance verification—activities that are already highly structured and therefore readily amenable to AI-assisted workflows.

Together, these developments reflect a broader shift in structural engineering practice—from work grounded in sustained professional judgment toward work organized around decomposable tasks, procedural compliance, and workflow optimization using software, and, more recently, large language models and AI tools. This article considers what these assumptions imply for engineering judgment over time.

Before proceeding, I note that this reflection was developed with the assistance of a large language model. As a practicing structural engineer and educator, I recognize the value of these tools in supporting analysis, structure, and communication. At the same time, their use highlights a central tension: they can extend professional capability while also potentially altering the conditions under which engineering judgment is developed, particularly for early-career engineers within increasingly task-discretized practice. The arguments, judgments, and conclusions presented here are my own.

At the center of the recent discussion about AI adoption in structural engineering practice is an implicit premise: that engineering work can be broken into separable tasks and reassigned or automated. Tasks are separable, workflows decomposable, and professional value is often measured through speed, consistency, and cost. Within this logic, replacing human effort with automation is framed as optimization rather than loss of professional engagement. The routine appearance of this perspective in professional venues suggests that engineering work is increasingly treated as a sequence of procedures that can be specified, checked, and transferred, rather than as a practice that depends on accumulated experience and judgment.
An NCSEA professional development webinar on structural engineering judgment in urgent failure response, “What Should We Do Now? Implementing Structural Engineering Judgment in Urgent Response to Localized Structural Failure” highlights a paradox: structural engineers are expected to make rapid, high-consequence decisions under uncertainty in moments of failure, even as everyday practice becomes increasingly organized around modular tasks, decomposed workflows, and compliance metrics. The profession affirms judgment rhetorically, even as the opportunities for engineers—especially early in their careers—to develop judgment through direct project responsibility are reduced.

Over the past several decades, structural engineering practice has largely organized around compliance, verification, and liability management. Codes, standards, and software have delivered undeniable gains in safety and reliability. At the same time, these tools and requirements have narrowed the domain in which judgment is expected to operate. Engineers are rewarded for demonstrating procedural correctness rather than exercising discretion informed by long experience. Judgment remains essential to practice, but it is often developed and exercised informally and is not consistently supported within formal structures of accountability. Early-career engineers receive few explicit signals—through incentives, evaluation, or mentorship—about when and how judgment should be exercised or developed.
In this context, AI does not arrive as a rupture but as a continuation. If engineering work is defined principally as rule-based application and bounded optimization, automation follows naturally. The convergence of recent discussions around artificial intelligence in structural engineering practice suggests that a broader systemic shift may be underway: the normalization of AI as a tool within the engineering consulting practice. At the same time, less attention is given to how early-career engineers develop grounded judgment—historically through continuity of practice and experience across projects, exposure to consequences, and responsibility carried over time. Whether the development of such judgment can persist among these engineers as practice is increasingly organized around discretized tasks that can be distributed across human and automated assistants remains an open question.

Engineering judgment develops through continuity of practice, responsibility for outcomes, and repeated engagement with uncertainty in real settings—what can be understood as practice-based knowledge. For structural engineers, this has traditionally meant learning by doing: developing fundamental skills, working through calculations and detailing decisions, and seeing how designs perform when field conditions inevitably diverge from assumptions. It is in these moments—when conditions are incomplete, conflicting, or unexpected—that judgment is formed. Engineering judgment is not simply rapid decision-making or pattern recognition. It is produced through prolonged exposure to consequences while operating under constraint: seeing how structures age, how failures unfold, and how systems perform over time in real projects, including conditions that fall outside design assumptions.

As engineering labor becomes fragmented, outsourced, or automated, the conditions that support development of this form of judgment will weaken. Tasks and associated knowledge are decomposed and redistributed, often limiting the opportunity for early-career engineers to follow a problem from initial design through construction and performance. This approach can reset or shorten professional engagement with a problem after each transaction, resulting in the firm retaining its symbolic outputs—drawings, calculations, reports—while individuals lose the capacity for cumulative learning. Firm-level performance in structural engineering may appear adequate in the near term even as the depth of understanding within individual engineers quietly erodes over time. This divergence is not easily visible within individual projects or tasks; it becomes apparent only when engineering practice is examined across time and across many projects.

Judgment does not accumulate unless engineers remain in contact with the long consequences of their decisions—across years and repeated cycles of use and adaptation. When that contact weakens, learning does not accumulate even if performance metrics are met. Over time, the profession becomes highly efficient at satisfying immediate requirements while becoming less capable of recognizing slow, compounding forms of failure that fall outside contractual or regulatory horizons.

This dynamic is most visible in domains where performance unfolds over decades. Residential infrastructure exposed to natural hazards provides a clear example. Following the 1970 Lubbock, Texas, tornado, research led by Professor Kishor Mehta of Texas Tech University identified practical design approaches for improving the performance of low-rise residential construction under extreme winds. Subsequent decades of research have reinforced these findings and demonstrated their effectiveness in reducing damage. Yet these approaches have not been systematically incorporated into model residential building codes or standard construction practice. As a result, neighborhoods continue to be built with very similar or comparable vulnerabilities in tornado-prone regions, resulting in the catastrophic damage patterns that recur across events. This persistence of vulnerability reflects that loss-mitigating knowledge does not accumulate in our residential infrastructure through practice over time.
This raises a practical question for firms and for the profession: who will develop the experience required to recognize and respond when conditions fall outside standardized workflows—particularly when those conditions are incomplete, conflicting, or diverge from typical design assumptions or if those workflows increasingly define early-career practice?

The divergence between academic research and professional practice makes this pattern easier to see. Research engineers at our universities analyze repetitive damage across hazard events and decades, observing how loss accumulates systemically. Practicing engineers typically operate at the scale of individual new design and construction projects within defined codes and liability structures. Neither perspective is deficient, but the gap can obscure long-horizon concerns and weaken the connection between judgment and the profession’s broader value to society.

Informal discussions with senior engineers in both academia and practice point to a consistent concern. AI is understood as another tool—much like the slide rule, calculator, or finite element software—but one that must be used without displacing independent judgment. At the same time, a more fundamental question emerges: if engineering is reduced to procedural compliance and AI-mediated output, what distinguishes the value of the engineer from the tools themselves? This concern extends to how engineers are trained. When education and early practice reward procedural correctness over judgment under uncertainty, the conditions required to develop that judgment weaken. The challenge, then, is not only how judgment is applied, but how it is cultivated and sustained across generations. This places renewed importance on how firms structure early-career training and mentorship, responsibility and continuity of project engagement.

Engineering practice is increasingly organized around efficiency and compliance, while research develops tools that are accelerating this transformation. Both respond rationally to prevailing incentives. The combined effect, however, is a profession that risks weakening the conditions under which engineering judgment develops—even as it continues to bear responsibility for safety and public trust. The use of AI in developing this reflection underscores a central distinction: AI can extend analytic reach, but the formation and preservation of professional engineering judgment remain human responsibilities. ■

About the Author

David O. Prevatt, Ph.D, PE, F.SEI, F.ASCE is Professor of Civil & Coastal Engineering and Kisinger Campo & Associates Term Professor at the University of Florida. His research focuses on wind engineering, residential infrastructure and structural performance under natural hazards with emphasis on resilience, and long-horizon continuity.