From Chatbots to Coordinated Agents: A Construction Management Taxonomy of Agentic AI Design Patterns

Summary

The article translates the general-purpose taxonomy of agency-based AI design patterns into construction management terminology, presenting 20 patterns grouped into four functional categories, and experimentally tests the grounding capability of the Retrieval pattern.

Why it matters

It provides construction firms with a common lexicon to assess the autonomy and reliability of AI tools, addressing the current lack of standards. The experimental case study highlights the challenges of grounding AI in dynamic project documentation, which is crucial for safe and effective digital transformation.

Critical take

The study relies on a synthetic bridge rehabilitation corpus and a limited test set, which may not fully capture the complexity of real-world construction projects. The balance between answer accuracy and distractor reliability raises concerns for practical application.