Beyond technical capability: how institutional resistance shapes generative AI exposure in civil and environmental engineering occupations

Purpose This article introduces the Integrated Sociotechnical Exposure Framework (ISEF) to assess generative AI exposure in civil and environmental engineering by separating technical AI suitability (TA) from institutional resistance (IR). The contribution is an operational measurement framework, not a claim that institutional constraint is a new concept. Design/methodology/approach The framework is applied to 674 O*NET tasks across 28 occupations. Task-variable pairs are scored by three large language models under a fixed rubric and then aggregated to occupation level. Exposure is modelled with a multiplicative form, and sensitivity checks compare additive, min-rule, and geometric alternatives. A small external practitioner survey is used as a calibration check rather than as task-level validation. Findings A large group of occupations score well on technical AI suitability but stay heavily constrained by liability, compliance, and sign-off requirements. The cleaned practitioner survey points in the same direction: across 15 task vignettes, higher perceived institutional constraint is strongly associated with lower adjusted AI exposure. Research limitations/implications The scoring still leans on LLMs, and the practitioner sample is small. Because no same-task expert panel is available in the present revision, the results should be read as a transparent first-stage index. The nine-response practitioner exercise is preliminary calibration, not validation of the occupation-level index or its constructed adjusted-score analogue. The natural next step is a larger same-task expert panel paired with adoption-linked field evidence. Practical implications For engineering organisations, adoption strategy is better anchored in accountable workflows, auditability, and professional review processes than in model capability on its own. Originality/value The paper pulls governance constraints into the exposure measure itself, and uses that lens to explain why technical-only indices tend to overstate near-term generative AI uptake in regulated engineering work. Its measurement logic is potentially transferable to other regulated professions, but any such application requires domain-specific specification and calibration of the institutional-resistance construct.