Development and Student Perceptions of a Custom AI Agent for Asynchronous Construction Education

Generative AI agents are increasingly used in education; however, most applications position them as supplementary tools rather than central instructional systems. This study develops and evaluates students’ perceptions of a custom AI agent designed to serve as an instructional interface for an asynchronous learning activity focused on BIM-based quantity takeoff (QTO). The agent was developed using a structured dialogue tree and a curated knowledge base to provide guidance through technical workflows. A classroom implementation was conducted with 114 senior-level undergraduate construction management students across eight course sections, with no direct instructor intervention during the activity. Because most participants had limited prior exposure to Autodesk Assemble and BIM-based QTO, the study examined whether students could use the AI-guided environment to support task completion and confidence in learning. Results from a mixed-methods post-activity survey indicated that students perceived the agent as useful, accurate, and effective in supporting model navigation, data extraction, and QTO tasks. Students also reported low reliance on instructor assistance, moderate confidence after the activity, and willingness to use similar tools in future learning. Qualitative responses emphasized the value of self-paced learning, immediate support, and structured, step-by-step guidance, while also identifying limitations in specificity of responses and conversational flexibility.