Integrating Agentic Artificial Intelligence into Architectural Design Studio Pedagogy

Architectural design studios rely on iterative critique to support students' design development; however, assessing functional performance, regulatory compliance, and life-safety requirements remains time-intensive and dependent on expert judgment. This study develops and evaluates a five-agent artificial intelligence framework integrating regulatory retrieval, plan analysis, compliance assessment, pedagogical feedback, and graphical markup production. The framework supports expert-led preliminary assessment. It was evaluated using 15 anonymized third-year hotel-design submissions representing four plan typologies. The retained classifications comprise 224 true positives, 41 false positives, and 14 false negatives, yielding micro-averaged precision of 84.53%, recall of 94.12%, and F1 of 89.07%. Mean case precision (macro precision) was 84.49%. AI functional scores averaged 56.50, compared with 54.87 for the calculated functional grades, with an exploratory rank association of Spearman ρ = 0.750. Four trials on each of three samples produced pooled precision of 76.24%, alongside recurring findings and variable outputs. Detection estimates remain conditional because the completeness of the expert reference inventory at a consistent finding unit and severity scope has not been established, and two additional correct findings are recorded separately. The findings support further evaluation of multi-agent AI for expert-supervised formative feedback while retaining human responsibility for grading and regulatory compliance decisions.