Monthly Archives: April 2026

Designing for Designers: A Multi-Agent Multi-Representational AI System to Enhance Automotive Design

Zhang, Xiaoke, Angela Kwon, Mi Zhou, Gene Moo Lee “Designing for Designers: A Multi-Agent Multi-Representational AI System to Enhance Automotive Design,” Work-in-progress.

Organizations increasingly seek to use large language models (LLMs) to support knowledge-intensive work. However, effective deployment requires systems that ground LLM reasoning in heterogeneous, domain-specific knowledge. In collaboration with the vehicle design team of a major automotive manufacturer, we develop Design Insight Atlas, a multi-agent, multi-representational retrieval-augmented generation (RAG) AI system for automotive design intelligence. The system grounds LLM responses in three complementary knowledge representations: structured vehicle specifications for factual analysis, automotive news for temporal and market intelligence, and a knowledge graph for relational reasoning. A central designer assistant orchestrates a news retrieval tool, a data analysis agent, and a knowledge graph analysis agent, integrating their outputs into unified, evidence-grounded responses. We evaluate the system using 160 designer-oriented questions across eight task categories and four backbone LLMs. Design Insight Atlas achieves an average overall win rate of 89.2% against vanilla LLM baselines and consistently improves comprehensiveness, accuracy, and actionability across models and question types. Our study demonstrates how multi-representational knowledge grounding and multi-agent retrieval can enhance LLM support for domain-specific organizational knowledge work.

Client AI Adoption and Auditing: Evidence from Process- and Product-Oriented AI

Park, Jaecheol, Pauline Wu, Rajesh Vijayaraghavan, Gene Moo Lee. “Client AI Adoption and Auditing: Evidence from Process- and Product-Oriented AI”, Under Review.

  • Presentations: TBD.

Artificial Intelligence (AI) is transforming firms’ information production, operations, and business models, with important implications for financial reporting and external auditing. We examine how auditors respond to client AI adoption, focusing on audit pricing and audit outcomes. Using textual disclosures in Form 10-K filings, we construct a novel firm-year measure of client AI adoption and further decompose it into AI embedded in internal processes and AI embedded in products and services. Using U.S. public firm data from 2010 to 2022 and a long-difference research design, we find that client AI adoption improves reporting discipline but does not lead to systematic changes in audit fees, consistent with offsetting efficiency and risk effects. When we distinguish between types of AI adoption, however, we find opposing audit responses. Process-oriented AI adoption leads to lower audit fees and improved reporting discipline, consistent with audit efficiency gains. In contrast, product-oriented AI adoption increases reporting complexity and risk, leading auditors to increase monitoring and scrutiny. Consistent with increased monitoring and error detection, product-oriented AI adoption increases the likelihood of subsequent financial restatements but not material misstatements, suggesting improved detection rather than deterioration in reporting quality. Cross-sectional analyses show that these effects vary with client complexity, operating performance, governance, and auditor industry expertise. Overall, our findings indicate that client AI adoption reshapes how auditors allocate effort, assess risk, and deploy monitoring, highlighting how technological change alters the audit production process and the financial reporting environment.