Workplaces are undergoing a rapid transformation toward the integration of generative artificial intelligence tools into daily tasks and leadership processes. However, this integration faces a fundamental knowledge gap known as AI illiteracy, which does not refer to ignorance of technology itself, but rather to the inability to critically evaluate its outputs, calibrate the appropriate level of trust in them, and integrate them responsibly into decision-making contexts. This article reviews recent scientific literature to document the impact of AI illiteracy on two central variables: individual and organizational productivity, and the quality of managerial decision-making. Empirical evidence indicates that generative AI can narrow education-related productivity gaps in the short term; however, these gains may be fragile if they are not accompanied by critical-use skills. Survey-based studies also demonstrate a significant negative association between AI illiteracy and the quality of strategic decision-making, mediated by psychological and cognitive variables such as automation bias and cognitive offloading. Finally, the article proposes a framework for developing AI fluency as a leadership competency, emphasizing the importance of managing technostress to ensure that investments in skills translate into sustainable performance. Keywords: AI illiteracy; productivity; managerial decision-making; automation bias; cognitive offloading Introduction Over the past few years, generative artificial intelligence tools have evolved from experimental side projects into structural components of organizational workflows. Data from McKinsey and Deloitte indicate that organizations are already beginning to redesign their processes around these technologies, accompanied by a sharp increase in investment. However, the return on these investments remains highly uneven: only approximately one-fifth of organizations are classified as genuine “ROI leaders.” This gap is not primarily a problem of tools, but rather a problem of leadership and knowledge. In this context, the phenomenon of AI illiteracy emerges as a concept that extends beyond the basic technical definition of familiarity with algorithms. Bertoncini argues that genuine illiteracy in this field is not about the ability to operate generative tools, but about the ability to question the logic underlying such systems and recognize that algorithms are “not neutral”; rather, they represent opinions embedded in software code that may perpetuate historical inequalities. This article aims to provide an analytical review of scientific evidence concerning how AI illiteracy manifests itself in workplace environments and what documented effects it has on productivity, on the one hand, and the quality of managerial decision-making, on the other. It is based on the premise that AI illiteracy is not merely a lack of technical skills, but rather a cognitive and ethical condition that generates measurable operational and organizational risks. 1. Defining the Concept: Beyond Technical Skills A precise distinction should be made between Tool Fluency and AI Fluency. The former refers to knowing how to formulate prompts, summarize documents, and generate ideas—skills whose competitive value diminishes with every update that makes these tools easier to use. The latter refers to understanding how artificial intelligence changes the nature of business decision-making: What problem are we actually solving? Do we possess the appropriate data? What should be automated, and what should remain human? How will value be measured, and when should we stop? This distinction is reflected in contemporary research approaches to measuring AI competence in professional contexts. A task-oriented approach proposes that standard assessment tools should measure the ability to interact practically with AI systems in workplace contexts rather than merely assessing abstract conceptual understanding. These practical competencies include evaluating AI outputs, matching tools to tasks, collaborating with intelligent systems, and trust calibration based on risk assessment. In an applied study involving a global pharmaceutical company (PharmaCo), gaps in these competencies were found to vary systematically according to organizational level. Executives required strategic literacy to understand the implications of AI for the business model, middle managers required operational fluency to integrate AI into workflows, while non-technical employees required foundational knowledge to interact effectively with AI tools. This diversity of needs means that AI illiteracy is not a single condition, but rather a spectrum of overlapping gaps. 2. Impact on Productivity: Short-Term Gains and Long-Term Risks 2.1 Empirical Evidence on Narrowing Productivity Gaps The economic literature provides strong evidence that generative AI increases productivity across different skill levels, with larger gains among less-educated groups. In a randomized experiment involving 1,174 adults outside corporate settings, researchers found that the performance gap between participants with higher and lower levels of education on work-like tasks was 0.548 standard deviations in the absence of AI. When an AI assistant was available, this gap declined to 0.139 standard deviations, meaning that AI closed approximately three-quarters of the initial gap. This partly explains why 66% of leaders across 31 countries prefer not to hire individuals who lack AI skills, and why 66% of executives in Latin America choose candidates who are “AI-literate” over candidates with longer experience but without these skills. 2.2 The “Augmentation Trap”: When Gains Become Losses However, the picture is more complex than it initially appears. Follow-up results from the aforementioned experiment indicate that the gains achieved were not simply the result of delegating tasks to the machine. Participants who used AI did not perform worse than the control group when the tool was removed; however, the substantial educational performance gap re-emerged. In other words, the gains partially declined in the absence of technological support, suggesting that some improvements were temporary and dependent on the availability of the tool rather than being the result of permanent learning. These concerns take a more specific form in the concept of The Augmentation Trap, proposed in a recent theoretical analysis. When workplaces are designed around “production pressure” and the ease of cognitive delegation, workers may fall into a cycle of excessive dependence on AI, gradually eroding their core expertise. The fundamental problem is that the boundaries of what can be delegated to a machine are determined by the user rather than by the technology itself. This creates considerable room for forms of cognitive offloading that may appear beneficial in the short term but become costly in the long term. This phenomenon is evident in observations reported in the U.S. labor market, where millions of employees use AI tools to conceal weaknesses in fundamental reading, writing, and numeracy skills. These tools can produce reports that appear professional “on the surface,” while employees may lack a genuine understanding of their content, creating a fragile workforce that is “unable to make decisions or solve complex problems in the absence of technological support.” This phenomenon has been referred to as Cognitive Surrender, in which employees blindly follow technological outputs without adequate scrutiny or review. 2.3 The Critical Condition: Managing Technostress A field study involving 403 employee–supervisor dyads in South Africa reveals a critical organizational condition. Collaborative AI literacy is associated with improved performance, but only when technostress is low. Under conditions of high technostress, increased AI competence no longer improves task–technology fit and may instead undermine it, reducing performance benefits or even reversing them. The managerial implication is clear: investment in skills produces positive returns only when the psychological conditions of the workplace are managed simultaneously. 3. Impact on Managerial Decision-Making: From Automation Bias to the Erosion of Judgment 3.1 AI Illiteracy as a Predictor of Decision Quality A recent study employing an explainable framework that integrates semantic language analysis with structural equation modeling found that higher levels of AI illiteracy are negatively associated with strategic decision quality and positively associated with three potentially harmful mediating variables: automation bias, uncritical trust in AI, and cognitive offloading. Automation bias is not merely a theoretical concept. In a synthetic simulation of educational and research contexts, Bertoncini and colleagues found that 78% of human agents accepted AI decisions without review in the absence of strong critical competencies. Field evidence from the Netherlands adds another dimension: blind trust in algorithms did not decline significantly until after the consequences of a major government scandal. The scandal served as an “ethical wake-up call,” but the associated social costs were substantial. 3.2 Cognitive Offloading as an Organizational Risk The problem with cognitive offloading is not merely ethical; it is also operational. When a manager delegates a sensitive decision—such as determining who is eligible for benefits or who should be audited—to a “black-box” system, they relinquish part of their ethical agency. Bertoncini, citing Serafim and colleagues, argues that assigning responsibility to machines that lack genuine ethical expertise represents a fundamental error. AI ethical competence lies, paradoxically, in knowing where AI should not be used. The literature on business leaders indicates that leaders need more than mastery of AI tools. As one practitioner at Boston University’s business school puts it, “AI can make a weak strategy execute faster. It can produce more outputs without producing more value.” At the leadership level, AI fluency means the ability to ask the right questions about what should be automated, what should remain human, how value should be measured, and what ethical and legal risks are being introduced. 3.3 Toward a Framework for Developing Leadership AI Fluency The development of tools such as the AI Literacy Development Canvas demonstrates that organizations are beginning to approach this gap as a systematic challenge rather than as a one-time training issue. The canvas enables organizations to map targeted competencies across three levels: 1. Conceptual literacy — understanding the role of AI within the value chain. 2. Ethical literacy — governance, bias, and compliance. 3. Practical literacy — operationalizing AI-generated insights within strategic planning. These competencies can then be tailored to different organizational segments. In the PharmaCo case study, assessments revealed specific gaps at each level. Executives lacked ethical literacy related to governing AI-generated insights in drug-approval decisions; middle managers lacked an understanding of how models operate within workflows; and non-technical employees lacked foundational knowledge of how AI tools generate their outputs. These precise assessments allow organizations to implement targeted training interventions rather than relying on broad and general “awareness-raising” initiatives. 4. Conclusion and Recommendations The review indicates that AI illiteracy in workplace environments is not merely a technical skills gap, but rather a cognitive-organizational condition that produces dual effects. In terms of productivity, initial gains may appear rapidly but can be fragile and reversible if they develop into persistent cognitive delegation. In terms of managerial decision-making, AI illiteracy is negatively associated with strategic decision quality through pathways involving automation bias and uncritical trust. The central paradox revealed by the literature is that AI can narrow performance gaps on individual tasks while simultaneously widening the gap between those who use it with critical awareness and those who become overly dependent on it. The critical factor is not simply access to the tool, but the presence of AI fluency that enables users to distinguish between productive delegation and harmful cognitive offloading. Operational Recommendations First, development programs should distinguish between tool fluency and AI fluency, with greater emphasis on the latter because it is less vulnerable to rapid imitation and technological change. Second, training should be designed according to organizational level: strategic competencies for leadership, operational competencies for middle managers, and foundational competencies for non-technical employees. Third, technostress management should be incorporated as a core component of any skills-development initiative, because evidence indicates that benefits may disappear or even reverse under conditions of high technostress. Fourth, organizations should establish institutional mechanisms for measuring decision quality, rather than focusing exclusively on speed of execution, because short-term productivity indicators may conceal the erosion of critical decision-making capabilities. Fifth, recognizing where AI should not operate should be considered part of AI competence rather than a limitation of that competence. References [1] Cruces, G., Fernández Meijide, D., Galiani, S., Gálvez, R. H., & Lombardi, M. (2026). Does Generative AI Narrow Education-Based Productivity Gaps? Evidence from a Randomized Experiment. NBER Working Paper 34851. [2] Bertoncini, A. L. C. (2025). AI Literacy as a Mechanism for Democratic Defense: From the Technical to the Socio-technical. Admethics. [3] Albannai, N. A. A., & Raziq, M. M. (2026). Navigating Ethical, Human-Centric Leadership in AI-Driven Organizations: A Thematic Literature Review. The Service Industries Journal, 46(5–6), 555–582. [4] Benlian, A., & Pinski, M. (2025). The AI Literacy Development Canvas: Assessing and Building AI Literacy in Organizations. Business Horizons. [5] Axios Report (2026). AI Is Hiding Employee Illiteracy and Threatening U.S. Corporate Productivity with a Silent Crisis. Al Emarat Al Youm. [6] Salem, M. A., & Khalil, Z. A. (2026). Integrating Semantic NLP and PLS-SEM for AI-Enabled Strategic Decision Support: An Explainable Framework for Assessing Organisational AI Illiteracy. Information. [7] Rodríguez Castelán, C., & Winkler, H. (2025). Artificial Intelligence Is Transforming Middle-Class Jobs. Can It Also Help the Poor? World Bank Blogs. [8] Tan, W., Nawaz, M., Shu, T., & Ramzan, B. (2026). Collaborative Artificial Intelligence Literacy and Employee Performance: Task–Technology Fit and Technostress in a Moderated Mediation Model. South African Journal of Business Management, 57(1). [9] Boston University. (2026). Why Business Leaders Need AI Fluency, Not Just AI Tools. [10] AI Literacy Assessment Revisited: A Task-Oriented Approach Aligned with Real-World Occupations. (2025). arXiv. [11] The Augmentation Trap: AI Productivity and the Cost of Cognitive Offloading. (2026). arXiv.