Artificial intelligence (AI) is increasingly integrated into workplace operations through general-purpose systems, including large language models, foundation-model platforms, and AI assistants. The Unified Theory of Acceptance and Use of Technology (UTAUT) explains adoption through performance expectancy, effort expectancy, social influence, and facilitating conditions; however, the four UTAUT-related constructs used here do not directly represent whether an AI system’s technical and governance supports fit the accountability demands of its intended use. This study tests perceived architectural fit, defined as the perceived match between those supports and the accountability demands of the intended use, as an added predictor in a UTAUT-based model. A quantitative vignette-based survey with prebalanced 2 × 2 conditions and post-vignette perception measures collected 350 responses from U.S. workers with workplace AI exposure and retained 282 after screening. The four UTAUT-related predictors jointly explained 59.7% of the variance in AI reliance intention (ARI). Adding perceived architectural fit increased explained variance by 10.0 percentage points, and perceived architectural fit remained significant in the full model. Participants receiving the enhanced architectural-support description reported higher perceived architectural fit than participants receiving the limited-support description. Assigned conditions did not produce statistically significant differences in perceived legitimacy or ARI. Secondary analyses found that perceived architectural fit was associated with perceived legitimacy and that perceived legitimacy was associated with ARI. A bootstrap analysis also identified a positive indirect association through perceived legitimacy after controlling for the UTAUT-related predictors. The findings provide quantitative evidence that perceived architectural fit may contribute explanatory value to UTAUT-based research on workplace ARI.
How do entrepreneurs claim the authority to lead an enterprise that was never theirs alone? Although the discursive construction of leadership has attracted growing interest, most existing work is qualitative and case-limited, and leadership authority has rarely been examined as a discursive phenomenon at scale. Drawing on discursive, relational, and narrative leadership theory, this study analyzes 59 entrepreneurial podcast interviews from the Innovation Fuel series, recorded between 2020 and 2026, combining lexicon-based authority detection, sentence-embedding, topic modeling, and clustering. Under the stated lexical rules, soft classifications account for 89.8% of transcripts, and individually voiced and collectively voiced authority language routinely co-occur within the same narratives. Borrowed authority is used as a provisional interpretive label for legitimacy claimed individually yet drawn from teams and networks; it is offered as a testable interpretation rather than a separately validated construct. Recorded guest gender is neither associated with the episode-level collective ratio (p = .205) nor recoverable from the analyzed discourse out-of-sample (cross-validated AUC of about 0.59), and gender labels are interspersed in the embedding space. Because complete transcripts include host speech, these findings cannot be attributed to guest speech alone. Crucially, each apparent result is submitted to a matched null test: the soft-authority prevalence and the gender null survive, whereas apparent cluster and topic authority bands are shown to be aggregation artifacts over poorly separated groups. Distinguishing patterns that survive such tests from those that do not, the study models a transparent, self-auditing approach to computational discourse analysis and contributes a reproducible workflow for leadership communication research on large corpora of naturally occurring talk.
The accelerating diffusion of generative artificial intelligence (AI) into executive decision-making raises a foundational question: what distinctively human capacities must senior leaders cultivate to lead with judgment, accountability, and ethical integrity in an AI-augmented organizational world? This conceptual paper advances the HALE Framework — Humanistic AI-Augmented Leadership for Executives — comprising five interdependent pillars: (1) self-actualization as a developmental foundation; (2) spiritual intelligence as an integrative meta-capacity; (3) complexity navigation as a strategic lens; (4) conscious engagement with the automation-augmentation paradox; and (5) human-centric design as an ethical orientation. Drawing on management theory, humanistic management scholarship, spiritual intelligence literature, complexity science, and recent empirical findings on AI and leadership, the framework positions executive leadership not as competing with AI but as cultivating the developmental, ethical, and integrative capacities that AI cannot replicate. Five formal propositions are advanced to guide future empirical inquiry. Implications are developed for executive development, board governance, and future research design. The paper's marginal contribution relative to extant scholarship — particularly De Cremer's (2023) human-centric AI leadership work — lies in integrating developmental psychology, spiritual intelligence, and complexity theory into a unified, proposition-bearing framework explicitly calibrated to the C-suite context.
This paper presents a novel semi-automated approach for creating high-quality datasets through ontology-guided knowledge extraction for domain-specific large language model fine-tuning. We address the challenge of sparse knowledge graphs (KG) generated from traditional triplet extraction methods by developing a hierarchical ontology construction framework applied to procurement domain data. Our methodology begins with procurement-specific filtering of FineWeb data using keyword-based selection, reducing the dataset size by 80%. We used Llama-3.2-3B for data annotation, achieving 3,000 positive and negative samples from 44,000 processed samples, followed by training a BERT-based classifier with an F1 score of 75%. We introduce a semi-manual ontology development approach that combines structured Resource Description Framework (RDF) with targeted large language models (LLMs) prompting for focused graph node expansion. The process involves clustering of extracted nodes to reduce complexity and enable topic-specific investigation. With procurement expert validation, we generated a dataset of 140 question-answer pairs covering key ontology nodes, while rest 460 samples were generated in automated fashion using ontology prompt. Our ontology achieves a Weighted Composite Score (WCS) of 76.42%, indicating high topic coverage across the procurement domain graph. Fine-tuning experiments on Llama-3.2-1B and Llama-3.2-3B models demonstrate improvements validated through blind A/B testing using the DeepEval framework: the fine-tuned Llama-3.2-1B model was preferred over the base model in 78.15% of comparisons for answer relevancy, 77.87% for faithfulness, and 77.95% for factual consistency rate (FCR). The fine-tuned Llama-3.2-3B model showed moderate gains, winning 68.35% for answer relevancy, 72.29% for faithfulness, and 72.36% for FCR.
Automated unit testing methods in the software development process are crucial for reducing costs, improving product quality, and ensuring system reliability. While current Large Language Models (LLMs) are highly successful in general-purpose code generation, they may fall short in ensuring structural integrity and producing executable code in industrial fields such as C++ and ROS 2, where memory management is critical and external dependencies are frequently used. The proposed method fills this gap by focusing not only on high-level languages, unlike existing studies in the literature, but also on industrial embedded system architectures. The proposed method developed in this study aims to create high-accuracy unit tests by reducing the hallucination rate for systems without existing test scope, and to develop systems with existing test scope using developer logic. Recently distinguished by its success in code generation, the 7-billion-parameter Qwen 2.5 Coder model was selected as the base model. A multilingual dataset consisting of over 13,000 unique code-test pairs was created to reduce the model's computational costs and improve test code generation speed. The model was trained using QLoRA (Quantized Low-Rank Adaptation) and LLM fine-tuning methods. The proposed method has contributed to time savings and increased efficiency by accelerating test code generation speed by approximately 4 times compared to existing cloud-based approaches. Furthermore, unlike functionality-focused black-box testing and raw text-based approaches in the literature, the model's understanding of the project context is ensured by using Abstract Syntax Trees (AST), and the hallucination problem is significantly reduced by employing white-box and structural testing principles that examine the internal structure and dependencies of the source code. The proposed method addresses the limitations of leveraging large language models when generating unit test code and the key points in producing effective unit test code for industrial applications.
