The Architecture, Engineering, and Construction (AEC) sector continues to struggle with productivity stagnation and limited digitalization, even though daily coordination already occurs through digital channels such as email and messaging apps. This paper presents a field deployment of artificial intelligence (AI) in a large construction–mining project (~2,000 workers; ~400 machines) operating continuously, 24 hours a day and 7 days a week. The objective was to convert unstructured text, photos, and videos into structured maintenance records (tickets), aligned with a predefined schema designed to support the administration, documentation, and analysis of the interventions performed. In steady-state operation, the AI-driven system processed an average of ~77.5 maintenance tickets per day (~542.5 per week; ~3.23 per hour). After forms-based trials (both external tools and forms embedded within the messaging app) resulted in partial adoption and fragmented visibility, the in-group agent achieved near-complete capture without requiring new apps, logins, or software training. It provided in-thread transparency for acknowledgements and status updates, while cutting ~90% of manual classification and data-entry effort through automated extraction and structuring, complemented by light human-in-the-loop review. This generated significant savings by mitigating equipment downtime and its associated costs (including both rental/use costs and lost productivity). A semantics-first design (LLM + dictionaries + schema validation) produced analysis-ready data and enabled broad adoption with minimal friction. The case demonstrates a successful real-world AI deployment in AEC and highlights a transferable principle: adapt technology to the systems people already use, enabling broad adoption and frictionless data capture, and allowing intelligence to operate automatically and transparently in the background of the system.
Cisterna et al. (Wed,) studied this question.