The promise of productivity or a source of anxiety? Employees’ perspectives on GenAI and workplace outcomes

You’re listening to “The promise of productivity or a source of anxiety? Employees’ perspectives on GenAI and workplace outcomes,” by A.J. Mohamud and colleagues. Published in 2026.

ISSN: 2331-1975 (Online) Journal homepage: the linked source

The promise of productivity or a source of anxiety? Employees’ perspectives on GenAI and workplace outcomes

Abdulkadir Jeilani Mohamud, Abdifatah Nour Rage, Ali Dahir Mohamed & Adam Mohamed Mohamud

To cite this article: Abdulkadir Jeilani Mohamud, Abdifatah Nour Rage, Ali Dahir Mohamed & Adam Mohamed Mohamud (2026) The promise of productivity or a source of anxiety? Employees’ perspectives on GenAI and workplace outcomes, Cogent Business & Management, 13:1, 2706234, DOI: 10.1080/23311975.2026.2706234

© 2026 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group

View supplementary material

Published online: 28 Jul 2026.

Submit your article to this journal Article views: 620 View related articles View Crossmark data

Artificial Intelligence, Digitalization, and New Technologies | Research Article

The promise of productivity or a source of anxiety? Employees’ perspectives on GenAI and workplace outcomes

Abdulkadir Jeilani Mohamuda, Abdifatah Nour Rageb, Ali Dahir Mohamedc and Adam Mohamed Mohamudd aFaculty of Computer Science & Information Technology, Mogadishu University, Mogadishu, Somalia; bFaculty of Computer Science & IT, Salaam University, Mogadishu, Somalia; cFaculty of Education, Mogadishu University, Mogadishu, Somalia; dFaculty of Economics & Management Science, Mogadishu University, Mogadishu, Somalia

ABSTRACT.

The rapid advancement of generative artificial intelligence (GenAI) has accelerated its adoption in workplaces worldwide. While prior research highlights its productivity enhancing potential, growing concerns remain regarding its impact on employee well-being, particularly job disruption and anxiety. This study examines the mechanisms through which GenAI influences both work productivity and employees’ intention to use such technologies, addressing a critical gap in individual level adoption. Drawing on Technological Determinism Theory, Disruptive Innovation Theory, Innovation Resistance Theory and Digital Divide Theory, a multi-dimensional research model was developed.

Using a quantitative cross-sectional survey, data were collected between December 2025 and February 2026 from 368 employees working in AI-enabled organizations in Somalia and analyzed using a hybrid Structural Equation Modeling– Artificial Neural Network (SEM–ANN) approach. The findings reveal that perceived GenAI intensity significantly enhances work productivity and directly increases employees’ intention to use GenAI. Work productivity also partially mediates this relationship. Conversely, perceived job disruption significantly increases employee anxiety, highlighting the psychological challenges associated with GenAI adoption. Furthermore, training support positively influences intention to use GenAI but negatively moderates the relationship between GenAI intensity and productivity.

These results demonstrate the dual nature of GenAI as both a driver of productivity and a source of psychological strain. The study offers practical implications for organizations to balance performance gains with employee support mechanisms. Limitations include reliance on self-reported data and a single country context, suggesting opportunities for future research.

1. Introduction.

ARTICLE HISTORY

SUBJECTS Artificial Intelligence; Information & Communication Technology (ICT); Internet & Multimedia - Computing&IT

The rapid advancement of generative artificial intelligence (GenAI) is fundamentally reshaping the nature of work, organizational processes and employee experiences. Unlike earlier generations of automation technologies, GenAI systems such as large language models and AI-assisted content generation tools are capable of performing complex cognitive tasks, including writing, coding, data analysis and decision support. These capabilities extend automation beyond routine operations into knowledge intensive domains, enabling machines to generate human like content and interact using natural language. This shift has been widely documented in both academic and industry reports, which emphasize that GenAI is not merely a technical innovation but a socio-technical transformation influencing how employees interact with digital systems.

CONTACT Abdulkadir Jeilani Mohamud the email address Faculty of Computer Science & Information Technology, Mogadishu University, Mogadishu, Somalia

Recent evidence emphasizes both the scale and acceleration of this transformation. Generative AI has the technical potential to automate work activities that account for approximately 60–70% of employees’ time, particularly those involving information processing, communication and cognitive analysis. At the same time, its integration into the workplace is expected to reshape labor demand significantly, with up to 30% of current work hours potentially automated by 2030. Importantly, rather than simply replacing jobs, GenAI is transforming the composition and content of work. Knowledge intensive occupations such as those in business, legal services and Science Technology Engineering and Mathematics (STEM) fields are increasingly augmented by AI, allowing employees to enhance performance, creativity and decision-making rather than being fully substituted by technology.

Practical analyses from consulting firms such as McKinsey and PwC highlight that these changes are not uniform across industries, with knowledge intensive sectors experiencing augmentation rather than substitution, while routine based jobs face higher risks of displacement.

On one hand, GenAI is widely recognized as a powerful productivity enhancing tool. By automating repetitive tasks and augmenting human capabilities, it enables employees to complete tasks more efficiently, improve the quality of outputs and redirect their efforts toward higher value activities. At the macroeconomic level, the potential impact of GenAI is substantial, with estimates suggesting that it could contribute between $2.6 trillion and $4.4 trillion annually to the global economy. These projections are supported by practical case studies in marketing, customer service and software engineering, where GenAI applications are already demonstrating measurable improvements in efficiency.

However, the adoption of GenAI introduces significant challenges and psychological concerns for employees. The disruptive nature of this technology raises critical questions about job security, skill obsolescence and long-term career trajectories. Large-scale workforce transformations are already underway, with projections indicating that approximately 12 million occupational transitions may be required by 2030. These shifts disproportionately affect lower-wage and routine-based jobs, thereby intensifying concerns about inequality and job displacement. Consequently, employees may perceive GenAI as a threat, leading to heightened levels of anxiety, stress and resistance to technological change an emerging phenomenon often referred to as ‘AI anxiety’.

This duality where GenAI simultaneously enhances productivity while inducing disruption and psychological strain highlights a critical paradox in the evolving digital workplace. Theoretical perspectives such as Technological Determinism and Innovation Resistance Theory help explain why employees may embrace productivity gains yet resist adoption due to perceived risks. Structuring the introduction around this practical theoretical rationale clarifies the need for integrative research.

Importantly, the social and cultural context of Somalia provides a distinctive backdrop for this study. Somalia is undergoing rapid digital transformation, with increasing investments in ICT infrastructure, mobile penetration and digital entrepreneurship. Yet, disparities in access to technology, limited institutional support and uneven digital literacy levels create a unique environment where GenAI adoption may amplify both opportunities and challenges. Cultural attitudes toward technology adoption in Somalia shaped by historical instability, limited exposure to advanced digital systems, and strong community-based work practices make employee perspectives particularly valuable for understanding how GenAI is perceived in developing economies.

Despite the growing body of research on artificial intelligence in organizations, existing studies tend to adopt a fragmented perspective, focusing either on performance-related benefits or on the negative psychological consequences of technology adoption. Limited attention has been given to examining these outcomes simultaneously within a unified framework, particularly in the context of GenAI. Furthermore, much of the existing literature emphasizes organizational level outcomes, often overlooking the individual-level experiences of employees who directly interact with GenAI systems in their daily work. This gap underscores the need for a more comprehensive and integrative approach that captures both the enabling and disruptive effects of GenAI on employees.

To address this gap, the present study develops an integrative conceptual framework grounded in multiple theoretical perspectives, including Technological Determinism Theory (TDT),

Disruptive Innovation Theory (DIT), Innovation Resistance Theory (IRT) and Digital Divide Theory (DDT) (van Dijk, 2006). Through this multi-theoretical lens, employees’ responses to GenAI are conceptualized as operating through dual pathways. The first pathway captures the productivity-enhancement mechanism, whereby increased GenAI intensity improves perceived work productivity and strengthens employees’ intention to use the technology. The second pathway reflects the disruption anxiety mechanism, in which GenAI intensity heightens perceived job disruption, leading to increased employee anxiety and potential resistance to technology adoption.

In addition, this study examines the role of contextual and individual factors particularly AI skills readiness and organizational training support in shaping employees’ responses to GenAI. Given the accelerating pace of technological change, the ability of employees to acquire and apply relevant digital and AI-related skills is becoming a critical determinant of successful technology adoption. Evidence suggests that large-scale reskilling, continuous learning and organizational support mechanisms are essential to enable workers to transition into new roles and fully realize the benefits of GenAI.

Empirically, this study contributes to the emerging literature by providing evidence from a developing economy context, where digital transformation is progressing rapidly but remains unevenly distributed. In such environments, disparities in access to technology, digital infrastructure and skills development opportunities may amplify both the opportunities and challenges associated with GenAI adoption. Understanding these dynamics is particularly important for designing inclusive organizational strategies that maximize productivity gains while minimizing adverse social and psychological impacts.

Overall, this study seeks to address a critical research question: Does GenAI primarily function as a driver of productivity or as a source of anxiety for employees in the workplace? By examining both outcomes simultaneously within a unified framework, the study offers a more balanced and nuanced understanding of GenAI’s impact. The findings are expected to contribute to theory by integrating multiple perspectives on technology adoption, disruption and resistance, and to practice by providing actionable insights for organizations seeking to implement GenAI in ways that enhance productivity while safeguarding employee well-being.

2. Literature review.

This study adopts a structured and integrative approach to the literature review to examine employees’ perspectives on Generative Artificial Intelligence (GenAI) in the workplace. Rather than presenting multiple disconnected theoretical perspectives, the review selectively focuses on four complementary and well-established theories Technological Determinism Theory (TDT), Disruptive Innovation Theory (DIT), Innovation Resistance Theory (IRT) and Digital Divide Theory (DDT). These theories are systematically integrated to explain both the positive (productivity enhancement) and negative (anxiety and disruption) outcomes associated with GenAI adoption.

By synthesizing recent empirical evidence, the literature review establishes the theoretical rationale for the hypotheses and clarifies the research gap: the need to examine both productivity outcomes and employees’ intention to use GenAI within a single conceptual model.

2.1. Technological determinism theory (TDT).

Technological Determinism Theory (TDT) posits those technological advancements act as a primary force shaping organizational structures, work practices and employee behaviors. Within the context of this study, TDT serves as the primary theoretical lens for explaining the productivity-enhancing effects of GenAI. The theory suggests that technology is not merely a passive tool but an active driver that transforms how work is designed, executed and evaluated. As digital technologies become increasingly embedded in organizations, employees’ productivity expectations, work routines and skill requirements are significantly altered. In the context of Generative Artificial Intelligence (GenAI), TDT is particularly relevant as GenAI systems automate cognitive tasks, enhance decision-making and augment employees’ analytical and creative capabilities.

Unlike traditional information systems, GenAI exhibits higher levels of autonomy and adaptability, thereby exerting a stronger influence on work processes. Prior studies have shown that advanced technologies improve perceived productivity and performance outcomes. Moreover, recent studies provide strong support for this perspective, demonstrating that GenAI adoption leads to measurable improvements in task efficiency, output quality and decision accuracy. Accordingly, TDT provides a strong theoretical basis for explaining how perceived GenAI intensity enhances employees’ work productivity and influences their intention to use GenAI.

2.2. Disruptive innovation theory (DIT).

Disruptive Innovation Theory explains how emerging technologies gradually transform and disrupt existing organizational practices, roles and skill structures. While such innovations improve efficiency and foster innovation, they also introduce uncertainty and instability, particularly for employees whose roles may be altered or replaced. GenAI represents a disruptive innovation as it increasingly performs tasks traditionally handled by knowledge workers, including writing, coding and problem-solving. As GenAI becomes more prevalent in organizations, employees may perceive threats to job security, role stability and long-term career prospects. Empirical studies have shown that technological disruption often leads to perceived job insecurity and uncertainty among employees.

Recent research highlights that GenAI adoption is associated with significant task reconfiguration, role ambiguity and evolving skill demands, which contribute to employees’ perceptions of job disruption. Therefore, DIT provides a suitable framework for explaining how perceived GenAI intensity contributes to employees’ perceptions of job disruption.

2.3. Innovation resistance theory (IRT).

Innovation Resistance Theory (IRT) explains why individuals develop resistance or negative reactions toward new technologies that disrupt established routines and competencies. Resistance may not always be expressed through direct opposition but can manifest emotional responses such as anxiety, stress and discomfort. In the proposed framework, IRT is positioned as the psychological mechanism linking perceived job disruption to employee anxiety. In the case of GenAI, employees may experience anxiety due to the rapid evolution of technology, its opaque decision-making processes, and its potential to outperform human capabilities in certain tasks. Prior research on technostress suggests that advanced technologies can create psychological strain when individuals perceive a lack of control or insufficient coping capacity.

Furthermore, perceived job disruption serves as a key antecedent of anxiety, as concerns over job loss and skill obsolescence undermine employees’ psychological security. Moreover, Recent studies emphasize that AI adoption is frequently associated with technostress, cognitive overload and emotional strain, particularly when individuals perceive limited control over technological outcomes. Thus, IRT is appropriate for explaining how GenAI-induced disruption translates into increased employee anxiety.

2.4. Digital divide theory (DDT).

Digital Divide Theory (DDT) emphasizes that individuals’ ability to benefit from technological advancements depends on unequal access to digital resources, skills and organizational support (van Dijk, 2006). In contrast to the previous theories, DDT plays a boundary-condition role in this study by explaining when and for whom GenAI leads to positive or negative outcomes. Recent research shows that AI-related skills and organizational training significantly influence whether employees perceive AI as an opportunity or a threat In modern workplaces, the digital divide extends beyond basic access to include disparities in advanced competencies such as AI-related skills. In the GenAI context, employees with higher levels of AI skills readiness are better equipped to understand and utilize AI tools effectively.

These employees are more likely to perceive GenAI as an opportunity rather than a threat. Conversely, employees with lower levels of digital skills may experience greater anxiety and perceive higher job disruption. Organizational support, including training and access to digital resources, further shapes employees’ experiences with GenAI. Prior studies suggest that such facilitating conditions enhance technology adoption and reduce resistance. Therefore, DDT provides a strong foundation for examining the moderating role or organizational training support.

2.5. Integrative theoretical perspective.

While prior studies have examined productivity, technostress and resistance in digital contexts, research focusing specifically on employees’ responses to GenAI remains limited. Existing studies often adopt a single-theory perspective or emphasize either positive or negative outcomes, overlooking the coexistence of both. By integrating TDT, DIT, IRT and DDT, this study provides a comprehensive framework to explain the dual impact of GenAI. Specifically, GenAI simultaneously enhances productivity (TDT), introduces job disruption (DIT), triggers anxiety (IRT) and is conditioned by digital capabilities and organizational support (DDT). This integrative perspective addresses an important gap in the literature by offering a balanced understanding of GenAI as both a productivity-enhancing tool and a potential source of employee anxiety.

3. Hypotheses development and research model.

This study develops its hypotheses by integrating Technological Determinism Theory (TDT), Disruptive Innovation Theory (DIT), Innovation Resistance Theory (IRT) and Digital Divide Theory (DDT) to explain employees’ responses to Generative Artificial Intelligence (GenAI). The hypotheses reflect a dual-path mechanism, where GenAI simultaneously enhances productivity while also triggering job disruption and anxiety an emerging phenomenon widely described in recent literature as the ‘productivity anxiety paradox’.

3.1. Perceived GenAI intensity and work outcomes.

According to Technological Determinism Theory (TDT), technological advancements play a central role in reshaping work processes, performance expectations and organizational outcomes. In the context of generative artificial intelligence (GenAI), increasing levels of technological intensity significantly enhance employees’ ability to automate routine cognitive tasks, support decision-making and generate higher-quality outputs. As GenAI tools become more embedded in daily work activities, employees are able to complete tasks more efficiently and focus on higher-value responsibilities. Recent empirical evidence supports this view, indicating that GenAI adoption leads to substantial improvements in individual productivity, time efficiency and task performance.

For example, large-scale workforce studies show that employees who frequently use GenAI report significantly higher productivity and improved work outcomes compared to non-users.

However, consistent with Disruptive Innovation Theory (DIT), the increasing intensity of GenAI also introduces structural changes in job roles, task composition and skill requirements. As GenAI systems increasingly perform tasks traditionally carried out by human employees, individuals may begin to perceive threats to job security, role stability and the relevance of their existing competencies. Empirical studies confirm that GenAI adoption is associated with significant shifts in task structures and an increasing demand for advanced cognitive and digital skills, which can heighten employees’ perceptions of job disruption. These changes do not necessarily imply immediate job loss but instead reflect growing uncertainty about future work arrangements and career trajectories.

Furthermore, as organizations intensify the adoption of GenAI, they often complement this transition with increased investments in training programs, digital infrastructure and skill development initiatives. Such organizational efforts are essential to ensure that employees can effectively utilize GenAI tools and adapt to evolving work environments. This perspective aligns with Digital Divide Theory (DDT), which emphasizes that individuals’ ability to benefit from technological advancements depends on their access to resources and opportunities for skill development. Recent evidence suggests that organizations adopting GenAI place greater emphasis on employee training and continuous learning to support technology integration and maximize productivity outcomes.

Taken together, these arguments suggest that perceived GenAI intensity simultaneously enhances productivity, increases perceptions of job disruption, and encourages organizational investment in training and skill development. Accordingly, the following hypotheses are proposed:

H1: Perceived GenAI intensity has a positive relationship with perceived work productivity.

H2: Perceived GenAI intensity has a positive relationship with perceived job disruption.

H3: Perceived GenAI intensity has a positive relationship with organizational training support.

3.2. Productivity, training support and intention to use GenAI.

Perceived work productivity represents employees’ subjective evaluation of how effectively and efficiently they are able to complete their work tasks when supported by generative artificial intelligence (GenAI). It encompasses improvements in task speed, output quality, decision accuracy and the ability to focus on higher-value activities. Within the framework of Technological Determinism Theory (TDT), technologies such as GenAI are not merely supportive tools but transformative forces that reshape how work is performed, thereby directly influencing productivity outcomes. As GenAI systems automate routine cognitive tasks, generate content and assist in complex problem-solving, they enable employees to reallocate effort toward strategic and creative activities, ultimately enhancing perceived work performance.

For instance, organizational research indicates that GenAI tools significantly improve knowledge workers’ productivity by assisting with writing, coding, data analysis and decision-making tasks. Similarly, large-scale industry reports highlight that employee using GenAI report substantial time savings and increased efficiency, particularly in tasks involving information processing and content generation. These productivity gains not only enhance individual performance but also contribute to organizational effectiveness by enabling faster innovation cycles and improved service delivery. Moreover, recent studies suggest that perceived productivity improvements strengthen employees’ intrinsic motivation and perceived usefulness of AI systems, thereby reinforcing their intention to continue using GenAI.

When employees experience tangible benefits such as reduced cognitive load and improved task outcomes, they are more likely to integrate GenAI into their routine workflows. Conversely, if productivity gains are not clearly realized, employees may question the value of GenAI, leading to lower adoption and usage intentions.

In addition, emerging research highlights that GenAI-driven productivity is not limited to efficiency gains but also includes augmentation effects, where human AI collaboration enhances creativity and problem-solving capabilities. This suggests that perceived work productivity in the GenAI context extends beyond traditional performance metrics to include qualitative improvements in work processes and outcomes. Based on the above discussion, perceived work productivity is a critical mechanism through which GenAI influences employees’ behavioral intention to use AI technologies. Employees who perceive higher productivity benefits from GenAI are more likely to develop sustained usage intentions, reinforcing the role of productivity as a central driver in AI adoption.

In addition, organizational training support represents a critical facilitating condition influencing technology adoption. DDT suggests that employees with access to training and digital resources are more capable of utilizing advanced technologies effectively. Empirical evidence indicates that employees who receive adequate AI-related training demonstrate stronger engagement and higher adoption intentions. Beyond indirect effects, GenAI intensity itself directly influences intention to use. As employees interact more frequently with GenAI systems, familiarity and perceived usefulness increase, leading to stronger behavioral intentions. This aligns with recent findings showing that regular GenAI users are more optimistic and willing to integrate AI into their daily work activities.

Thus, the following hypotheses are proposed:

H4: Perceived work productivity has a positive relationship with intention to use GenAI.

H5: Perceived GenAI intensity has a positive relationship with intention to use GenAI.

H7: Perceived training support has a positive relationship with intention to use GenAI.

3.3. Job disruption and employee anxiety.

Innovation Resistance Theory (IRT) posits that individuals experience psychological discomfort when confronted with innovations that disrupt established routines, competencies and role identities. In the context of generative artificial intelligence (GenAI), perceived job disruption represents a critical cognitive appraisal through which employees evaluate the potential threat of technological change. As GenAI increasingly automates knowledge-intensive tasks and augments human decision-making, employees may perceive a misalignment between their existing skills and evolving job requirements, thereby triggering uncertainty and emotional strain.

Consistent with Disruptive Innovation Theory, the integration of GenAI into workplace processes alters task structures and redistributes responsibilities between humans and intelligent systems. This transformation often leads to heightened perceptions of job insecurity, role ambiguity and skill obsolescence. Empirical studies provide strong support for this relationship, demonstrating that employees who perceive higher levels of technological disruption are more likely to report increased anxiety and psychological distress. Similarly, recent evidence suggests that employees interpret GenAI as both an opportunity for performance enhancement and a potential threat to long-term employability, reinforcing the dual and often conflicting nature of AI adoption.

Furthermore, emerging research highlights that the rapid proliferation of AI tools can intensify cognitive demands and contribute to what is increasingly referred to as ‘AI fatigue’. This phenomenon reflects the mental exhaustion associated with continuously adapting to new systems, managing algorithmic outputs and maintaining performance expectations in AI-augmented environments. As employees struggle to cope with these demands, perceived job disruption becomes a salient stressor that undermines psychological well-being and increases workplace anxiety. In line with the Transactional Theory of Stress, such disruptions are interpreted as environmental stressors that exceed individuals’ coping capacities, resulting in heightened emotional responses.

When employees perceive that GenAI threatens their job stability or diminishes the value of their existing competencies, they are more likely to experience anxiety characterized by worry, uncertainty and reduced sense of control. This aligns with recent findings indicating that AI-driven transformations can negatively impact employees’ mental health, particularly when organizational support and clarity regarding role changes are insufficient.

Taken together, these arguments suggest that perceived job disruption functions as a key mechanism through which GenAI influences employees’ psychological outcomes. As the intensity of GenAI adoption increases, so does the likelihood that employees will perceive disruption in their work roles, which in turn elevates their levels of anxiety.

Accordingly, the following hypothesis is proposed:

H6: Perceived job disruption has a positive relationship with employee anxiety.

3.4. GenAI intensity and AI skills readiness.

Digital Divide Theory (DDT) posits that individuals’ ability to effectively benefit from emerging technologies is largely determined by their level of digital skills, competencies and access to learning opportunities. In the context of generative artificial intelligence (GenAI), this perspective is particularly relevant, as the effective use of GenAI tools requires not only basic digital literacy but also higher-order cognitive and technical capabilities. As GenAI becomes increasingly embedded in organizational workflows, employees are exposed to new forms of human–AI interaction, which can facilitate experiential learning and capability development over time.

Consistent with the ‘learning-by-doing’ perspective, continuous exposure to GenAI systems enables employees to gradually develop AI-related competencies, such as prompt engineering, critical evaluation of AI outputs and task augmentation strategies. Recent empirical studies indicate that employees who frequently interact with GenAI tools demonstrate higher levels of AI skills readiness and adaptability, as these technologies require active engagement and iterative learning. Moreover, GenAI adoption has been associated with an increasing demand for hybrid skill sets that combine domain expertise with digital and AI-related capabilities, further reinforcing the role of technology exposure in skills development.

Furthermore, positive relationship is observed between the use of chatbots and business outcomes, with a significant indirect effect through the redesign of routines, indicating that organizations must adapt to this disruptive technology for effective integration

In addition, organizations that intensify the use of GenAI often complement this adoption with formal training programs, reskilling initiatives and knowledge-sharing practices. Such organizational efforts enhance employees’ preparedness and confidence in using AI systems, thereby strengthening their overall AI skills readiness. This suggests that GenAI intensity does not merely influence work processes but also plays a developmental role in shaping employees’ competencies and long-term employability. Therefore, higher levels of perceived GenAI intensity are expected to positively influence employees’ AI skills readiness by increasing exposure, facilitating experiential learning and encouraging continuous skill development. Based on this reasoning, the following hypothesis is proposed:

H8: Perceived GenAI intensity has a positive relationship with AI skills readiness.

3.5. Mediating role of perceived work productivity.

From a theoretical perspective, TDT suggests that the impact of technology on behavioral outcomes is often mediated by performance-related mechanisms. In the case of GenAI, productivity gains serve as a key mechanism through which technology influences employees’ intention to use it. Empirical studies support this mechanism, showing that GenAI improves efficiency and task performance, which in turn increases employees’ willingness to adopt and continue using AI tools. However, recent research also indicates that productivity gains may not always directly translate into organizational outcomes, highlighting the importance of examining mediating pathways. Consistent with these arguments, this study proposes that perceived work productivity partially mediates the relationship between GenAI intensity and intention to use GenAI.

H9: Perceived work productivity mediates the relationship between perceived GenAI intensity and intention to use GenAI.

3.6. Moderating role of training support.

Digital Divide Theory further suggests that organizational support can influence how employees translate technology exposure into outcomes. Training support, in particular, plays a critical role in shaping employees’ experiences with GenAI. While prior studies generally assume that training strengthens the positive effects of technology, emerging evidence suggests a more nuanced relationship. For example, excessive reliance on structured training or multiple AI tools may introduce complexity, reduce autonomy and increase cognitive load, thereby weakening productivity gain. This perspective aligns with recent findings on the productivity paradox of GenAI, where increased technological support does not always lead to proportional productivity improvements. Instead, training may substitute for experiential learning, reducing the marginal benefits of GenAI intensity.

Thus, the following hypothesis is proposed:

H10: Perceived training support moderates the relationship between perceived GenAI intensity and perceived work productivity, such that the relationship is weaker at higher levels of training support.

3.7. Research model.

The proposed research model explains employees’ responses to generative artificial intelligence (GenAI) in the workplace through a dual-pathway mechanism, capturing both its enabling and disruptive effects. Rather than if GenAI adoption leads solely to positive or negative outcomes, the model posits that employees simultaneously experience productivity gains and psychological strain, depending on how GenAI reshapes their work processes and perceived job security. Importantly, this study advances prior models by explicitly examining how these positive and negative pathways are not independent but interrelated, thereby providing a more rigorous and realistic representation of employees’ experiences with GenAI.

First, drawing on Technological Determinism Theory (TDT), the model proposes a productivity-enhancing pathway. Specifically, perceived GenAI intensity (i.e. the extent to which GenAI is embedded in daily work activities) directly improves employees’ perceived work productivity. This occurs because GenAI automates routine cognitive tasks, augments decision-making and enhances efficiency and output quality. In turn, increased work productivity strengthens employees’ intention to use GenAI, reflecting the idea that technologies perceived as performance-enhancing are more likely to be continuously adopted. In addition, GenAI intensity also directly influences intention to use, indicating that employees may develop favorable usage intentions based on both direct exposure and perceived performance benefits.

Second, informed by Disruptive Innovation Theory (DIT), the model introduces a disruption–anxiety pathway. As GenAI becomes more pervasive, it alters traditional job roles, task structures and skill requirements. This leads employees to perceive higher levels of job disruption, particularly in terms of job security and skill obsolescence. Perceived job disruption subsequently increases employee anxiety, reflecting concerns about future employability and role stability. Consistent with Innovation Resistance Theory (IRT), such anxiety represents a psychological response to technological change and may reduce employees’ willingness to fully embrace GenAI, thereby potentially undermining its productivity benefits.

Crucially, extending prior research, the model explicitly connects the negative pathway (job disruption and anxiety) with the positive pathway (productivity and intention to use). Specifically, employee anxiety is expected to weaken the positive influence of perceived work productivity on intention to use GenAI, as psychologically strained employees may be less willing to translate productivity gains into sustained usage behavior. Moreover, heightened anxiety may also attenuate the direct effect of GenAI intensity on intention to use, reflecting resistance or hesitation despite frequent exposure. This integrated perspective ensures that the model captures the dynamic tension between performance gains and psychological costs, rather than treating them as isolated outcomes.

Third, the model incorporates AI skills readiness and organizational training support as critical contextual enablers, grounded in Digital Divide Theory (DDT). These factors capture employees’ ability to effectively engage with GenAI technologies. AI skills readiness reflects employees’ perceived competence in using AI tools, while training support represents the extent to which organizations provide resources, guidance and learning opportunities. These factors not only directly influence employees’ intention to use GenAI but also shape how GenAI intensity translates into outcomes. Training support plays a moderating role in the relationship between GenAI intensity and perceived work productivity, indicating that organizational support conditions the extent to which employees can convert GenAI exposure into productivity gains.

Furthermore, training support may also mitigate the negative pathway by reducing perceived job disruption and anxiety, as employees who receive adequate organizational support are better equipped to adapt to technological changes. This highlights the dual role of training support as both an enabler of productivity and a buffer against psychological strain, thereby reinforcing the interconnected nature of the model.

Positioning perceived work productivity as a mediator reflects the argument that GenAI improves behavioral outcomes (i.e. intention to use) primarily through its impact on employees’ performance perceptions. Similarly, positioning perceived job disruption as an antecedent of anxiety highlights the behavioral mechanism through which technological disruption translates into psychological responses. The inclusion of training support as a moderator further emphasizes that organizational interventions do not operate uniformly but instead shape the strength of technology–outcome relationships. Additionally, by incorporating the interaction between positive and negative pathways, the model moves beyond linear assumptions and captures the complex, contingent nature of GenAI adoption in organizational settings.

Overall, the conceptual model consists of six core constructs:

1. Perceived GenAI Intensity.

2. Perceived Work Productivity (Mediator).

3. Perceived Job Disruption.

4. Employee Anxiety.

5. AI Skills Readiness.

6. Organizational Training Support (Moderator).

7. Intention to Use GenAI (Outcome).

By explicitly integrating both enabling and inhibiting mechanisms, as well as their interrelationships, the model provides a more comprehensive and theoretically robust framework for understanding employees’ responses to GenAI. It captures direct, indirect and conditional effects while also accounting for the interplay between productivity gains and psychological strain. This integrative approach advances the emerging literature on GenAI by demonstrating that employees’ adoption decisions are shaped not only by performance benefits but also by the extent to which technological change generates uncertainty and anxiety (see Table 1 and Figure 1).

4. Research methodology.

A quantitative research approach was adopted with a positivist philosophical standpoint to examine employees’ responses to generative artificial intelligence (GenAI) in the workplace. The study targeted employees working in organizations where GenAI tools are being used or introduced, particularly within the context of developing economies such as Somalia. A purposive non-probability sampling technique was employed due to the absence of a comprehensive sampling frame and to ensure that respondents possessed relevant experience with digital technologies and AI-enabled tools. This approach is appropriate for studies that require information-rich participants capable of evaluating specific technological phenomena, thereby enhancing the internal validity of the findings despite potential limitations in generalizability.

The inclusion criteria required respondents to be currently employed, to use or be exposed to digital tools in their work environment, and to have at least basic familiarity with AI-enabled applications. Responses that did not meet these criteria were excluded during the screening process.

A five-point Likert scale online self-administered questionnaire was adopted (with 1 = strongly disagree and 5 = strongly agree) to measure each item. The measurement instrument was systematically developed by adopting and adapting items from validated scales in prior studies, ensuring alignment between theoretical constructs and empirical measurement. Specifically, perceived GenAI intensity items were adapted from Tarafdar et al. (2019) and Viswanath et al. (2012); perceived work productivity from Palvalin (2017); intention to use GenAI from Venkatesh et al. (2003); perceived job disruption from Brougham and Haar (2018) and Makarius et al. (2020); employee anxiety from McCarthy et al. (2016) and Tarafdar et al. (2019); AI skills readiness from Compeau and Higgins (1995) and van Laar et al. (2017); and organizational training support from Rai et al. (2019) and Venkatesh et al. (2003).

Each construct was operationalized using multiple reflective indicators explicitly derived from its theoretical definition, ensuring content validity and conceptual consistency between the research model and measurement items. Appendix A presented each construct, and it is items.

The structural model was developed based on an integration of Technological Determinism Theory (TDT), Disruptive Innovation Theory (DIT), Innovation Resistance Theory (IRT) and Digital Divide Theory (DDT). These theoretical foundations guided the specification of both direct and indirect relationships among constructs, including mediating and moderating effects. Specifically, productivity-related pathways were grounded in TDT, while disruption and anxiety pathways were derived from DIT and IRT. Contextual enablers such as AI skills readiness and training support were incorporated based on DDT. This theory-driven model specification ensures that the hypothesized relationships are not only empirically testable but also theoretically justified.

Prior to the main data collection, content validity was assessed through expert review by two academic specialists, followed by a pilot test involving 45 respondents. The pilot results indicated that all measurement items were reliable, clear and appropriate for the study context, leading to minor refinements in wording and structure. Data were collected through an online survey between December 2025 and February 2026. After collecting responses from 380 participants, a rigorous data screening process was conducted to remove incomplete, inconsistent, or ineligible responses, resulting in a final sample of 368 valid questionnaires used for analysis.

To determine the minimum required sample size, GPower 3.1 was utilized following established procedures in prior research. Based on the parameters of effect size (f2) = 0.15, significance level (α) = 0.05, statistical power (1 – β) = 0.95, and six predictors, the minimum required sample size was calculated to be 146 respondents. In addition, this estimation was complemented using Cohen’s guidelines and recent SEM literature, which suggest a minimum sample size of at least 200 for models of moderate complexity. The final sample size of 368 therefore exceeds both statistical and methodological thresholds, ensuring adequate power, stability of parameter estimates and robustness of the results.

Covariance-based structural equation modeling (CB-SEM) using AMOS was employed to analyze the data, following a two-step analytical procedure recommended in the literature. First, confirmatory factor analysis (CFA) was conducted to assess the measurement model in terms of reliability and validity. Indicator reliability was evaluated using standardized factor loadings, with a recommended threshold of 0.60 or higher. Internal consistency reliability was assessed using composite reliability (CR), with values above 0.70 indicating acceptable reliability. Convergent validity was evaluated using average variance extracted (AVE), where values exceeding 0.50 indicate that constructs explain more than half of the variance in their indicators.

Discriminant validity was assessed using both the Fornell–Larcker criterion and the heterotrait–monotrait ratio (HTMT), with HTMT values below 0.85 (strict threshold) or 0.90 (liberal threshold) indicating adequate discriminant validity. Second, the structural model was evaluated to test the hypothesized relationships, including direct, mediating and moderating effects. Model fit was assessed using multiple goodness-of-fit indices, including χ2/df (<3.0), CFI, TLI and IFI (>0.90), and RMSEA and SRMR (<0.08), following established SEM guidelines. This approach is appropriate for theory testing and allows for the estimation of complex relationships among latent constructs.

To ensure methodological rigor, additional diagnostic tests were performed. The risk of common method bias (CMB) was assessed using the full collinearity approach, where variance inflation factor (VIF) values below 3.3 indicate the absence of CMB. All VIF values in this study were below the recommended threshold, confirming that CMB is not a concern (see Table 2). Moreover, Harman’s single-factor test was conducted to examine whether a single factor accounts for the majority of the variance in the data. Potential endogeneity issues were examined using the Gaussian copula approach, while unobserved heterogeneity was assessed using multi-group analysis; both tests confirmed the robustness and stability of the model across subsamples.

To enhance transparency and replicability, the technical specifications of the study, including research design, sampling strategy, measurement scale and analytical techniques, are summarized in Table 3.

This study was conducted in accordance with established ethical principles for research involving human participants. The research protocol was reviewed and approved by the Research Ethics Committee (REC) at Salam University, Mogadishu, Somalia (Approval No: 2025/SU-REC/AMSHS/P0378; Date of Approval: November 21, 2025). All participants involved in this study were adults, and their participation was entirely voluntary. Written informed consent was obtained through the online survey platform, where participants were provided with an information sheet outlining the purpose of the study, their rights as participants, and assurances of confidentiality and anonymity. Participants were required to indicate their consent before proceeding to the questionnaire and were informed that they could withdraw from the study at any time without any consequences.

Overall, the methodological approach adopted in this study ensures rigor in sampling, instrument development, data collection and statistical analysis, thereby enhancing the reliability and validity of the research findings.

5. Result and discussion.

This section presents the empirical results and provides an integrated discussion of the findings in relation to the proposed research model and its underlying theoretical foundations. The analysis proceeds systematically through several stages, beginning with an overview of respondents’ demographic characteristics, followed by a comprehensive evaluation of the measurement model to establish reliability and validity. Subsequently, the structural model is assessed to test the hypothesized relationships, including mediation and moderation mechanisms that capture the dynamic interplay between productivity outcomes and psychological responses. To further enhance analytical robustness, artificial neural network (ANN) analysis is employed as a complementary technique to SEM, enabling the detection of potential non-linear patterns that may not be captured through traditional parametric methods.

Collectively, these analyses provide a nuanced and multidimensional understanding of employees’ responses to GenAI, demonstrating how its increasing intensity simultaneously drives performance gains while generating perceptions of disruption and anxiety in the workplace.

5.1. Respondents’ profile.

A total of 368 valid responses were included in the final analysis. The key demographic characteristics of the respondents are summarized in Figure 2. Overall, the sample is predominantly male (76.1%) and composed largely of early- to mid-career employees, with the majority aged between 26 and 35 years (72.5%). In terms of education, most respondents hold at least a bachelor’s degree (49.2%) or master’s degree (32.1%), indicating a relatively well-educated workforce capable of engaging with advanced technologies such as generative artificial intelligence (GenAI). Regarding organizational characteristics, more than half of the respondents have 5 years or less tenure (56.3%), and the majority report 1–3 years of work experience (67.4%), suggesting that the sample is largely composed of relatively less experienced employees.

Participants were drawn from diverse sectors, with the largest representation from services, education, public sector and healthcare, enhancing the contextual relevance of the findings across multiple industries. To improve clarity and readability, key demographic distributions are also illustrated using pie charts (see Figure 2).

5.2. Measurement model evaluation.

The measurement model specifies the relationships between the latent constructs and their observed indicator variables. As part of the measurement model assessment, the standardized factor loadings of all indicators were examined to evaluate indicator reliability. Indicators with factor loadings below the recommended threshold of 0.60 are typically removed to ensure adequate measurement quality. During this evaluation, four items ASR1, ASR2, ASR3 and PTS5 were removed from the analysis due to low factor loadings that did not meet the recommended threshold. After removing these items, the remaining indicators demonstrated acceptable standardized factor loadings ranging from 0.600 to 0.934, indicating satisfactory indicator reliability.

The first component of the measurement model evaluation is construct reliability, which was assessed using composite reliability (CR). The recommended cutoff value for CR is 0.70, indicating satisfactory internal consistency among the items measuring each construct. As shown in Table 3, all constructs exceed this threshold, with CR values ranging from 0.813 to 0.940, confirming that the measurement scales exhibit strong internal consistency.

SMC = Squared Multiple Correlation; GenAI Intensity = Perceived Generative AI Intensity; PWP = Perceived Work Productivity; ITU = Intention to Use GenAI; PJD = Perceived Job Disruption; EAX = Employee Anxiety; ASR = AI Skills Readiness; PTS = Training Support.

The second component is convergent validity, which was evaluated using the average variance extracted (AVE). The acceptable threshold for AVE is 0.50, indicating that the construct explains at least 50% of the variance in its indicators. The AVE values reported in Table 4 range from 0.521 to 0.759, demonstrating that all constructs achieve adequate convergent validity.

CR AVE MSV MaxR(H) ASR PAI EAX PTS PWP ITU PJD

CR = Composite Reliability; AVE = Average Variance Extracted; MSV = Maximum Shared Variance; MaxR(H) = Maximum Reliability; GenAI Intensity = Perceived Generative AI Intensity; PWP = Perceived Work Productivity; ITU = Intention to Use GenAI; PJD = Perceived Job Disruption; EAX = Employee Anxiety; ASR = AI Skills Readiness; PTS = Training Support. The bold values represent the square root of the Average Variance Extracted (AVE) for each construct, reported according to the Fornell–Larcker criterion for assessing discriminant validity. The diagonal (bold) values are expected to be greater than the corresponding inter-construct correlations in the same row and column, indicating that each construct shares more variance with its own indicators than with other constructs. The bold formatting is used solely to distinguish these diagonal values for ease of interpretation.

Finally, discriminant validity was assessed using the Fornell–Larcker criterion and the comparison between maximum shared variance (MSV) and AVE. According to Fornell and Larcker (1981), discriminant validity is established when the square root of AVE for each construct is greater than its correlations with other constructs. In addition, MSV values should be lower than the corresponding AVE values. The results indicate that these conditions are satisfied for all constructs, confirming that each construct is empirically distinct from the others. Furthermore, discriminant validity was assessed using the heterotrait–monotrait ratio (HTMT). All HTMT values are below the recommended threshold of 0.85 (strict criterion) and 0.90 (liberal criterion), indicating that discriminant validity is established among the constructs (See Tables 4–6 and Figure 3).

Overall, the results demonstrate that the measurement model satisfies the recommended criteria for indicator reliability, construct reliability, convergent validity and discriminant validity, indicating that the measurement scales used in this study are both reliable and valid for subsequent structural model analysis

5.3. Structure model.

The structural model was analyzed to test the hypothesized relationships among perceived GenAI intensity, perceived work productivity, perceived job disruption, employee anxiety, digital access and training support, AI skills readiness and intention to use GenAI. The results indicate that perceived GenAI intensity has a significant positive effect on perceived work productivity (β = 0.576, SE = 0.056, CR = 10.203, p < 0.001), supporting H1. This finding suggests that higher exposure to GenAI technologies in the workplace enhances employees’ perceptions of their productivity and efficiency. Similarly, perceived GenAI intensity significantly predicts perceived job disruption (β = 0.346, SE = 0.055, CR = 6.321, p < 0.001), supporting H2.

This result indicates that while employees recognize the productivity benefits associated with GenAI technologies, they simultaneously perceive potential risks related to job displacement or role changes due to automation. The results also show that perceived GenAI intensity positively influences digital access and training support (β = 0.515, SE = 0.043, CR = 11.960, p < 0.001), supporting H3. This finding suggests that higher levels of GenAI adoption within organizations are associated with increased availability of training opportunities for employees. Regarding behavioral outcomes, perceived work productivity significantly increases employees’ intention to use GenAI (β = 0.127, SE = 0.027, CR = 4.753, p < 0.001), supporting H4. This result indicates that employees are more willing to adopt GenAI technologies when they perceive that such tools improve their work performance.

In addition, perceived GenAI intensity directly influences intention to use GenAI (β = 0.129, SE = 0.038, CR = 3.427, p < 0.001), supporting H5. This suggests that increased exposure to GenAI technologies in the workplace encourages employees to continue using these tools. Furthermore, perceived job disruption has a significant positive effect on employee anxiety (β = 0.496, SE = 0.046, CR = 10.660, p < 0.001), supporting H6. This finding indicates that employees who perceive higher levels of job disruption due to GenAI technologies are more likely to experience anxiety in the workplace. The results also reveal that perceived training support significantly predict intention to use GenAI (β = 0.470, SE = 0.035, CR = 13.448, p < 0.001), supporting H7.

This suggests that employees are more likely to adopt GenAI technologies when organizations provide adequate digital resources and training support. Finally, perceived GenAI intensity does not significantly predict AI skills readiness (β = 0.002, SE = 0.024, CR = 0.098, p = 0.922). This finding indicates that exposure to GenAI

GenAI = Perceived GenAI Intensity; PWP = Perceived Work Productivity (PWP); UGenAI = Intention to Use GenAI (ITU); PJD = Perceived Job Disruption; Anxiety = Employee Anxiety; Skills = AI Skills Readiness; Access = Digital Access and Training Support.

GenAI = Perceived GenAI Intensity; PWP = Perceived Work Productivity; ITU = Intention to Use GenAI.

technologies in the workplace alone is insufficient to enhance employees’ readiness to develop AI-related skills.

Overall, the structural model results demonstrate that GenAI adoption in the workplace is associated with both productivity-related benefits and psychological concerns, highlighting the complex role of GenAI technologies in shaping employees’ perceptions, behaviors and emotional responses. The result represented in Table 7 and Figure 4.

5.4. Mediation analysis.

A mediation analysis was conducted to examine whether perceived work productivity (PWP) mediates the relationship between perceived GenAI intensity and employees’ intention to use GenAI (ITU). The results are presented in Table 8 and Figure 5. The findings show that perceived GenAI intensity significantly predicts perceived work productivity (β = 0.576, p < 0.001). In turn, perceived work productivity significantly influences employees’ intention to use GenAI (β = 0.198, p < 0.001). Additionally, the direct effect of perceived GenAI intensity on intention to use GenAI remains significant (β = 0.330, SE = 0.040, CR = 8.259, p < 0.001). To further examine the mediation mechanism, the indirect effect was assessed using bootstrapping 5,000.

The results indicate that the indirect effect of perceived GenAI intensity on intention to use GenAI through perceived work productivity is significant (β = 0.114, p < 0.001). Because both the direct effect and the indirect effect are significant, the findings provide evidence of partial mediation. This result suggests that perceived GenAI intensity not only directly encourages employees to adopt GenAI technologies but also indirectly promotes adoption by enhancing employees’ perceptions of their work productivity.

5.5. Moderation analysis.

A moderation analysis was conducted to examine whether training support moderates the relationship between perceived Generative AI intensity (GenAI) and perceived work productivity (PWP). The analysis specifically tested whether the interaction between GenAI intensity and organizational training support significantly influences employees’ productivity perceptions.

The results presented in Table 9 and Figure 6 show that the interaction effect between GenAI intensity and training support on perceived work productivity is statistically significant (β = −0.167, SE = 0.077, CR = −2.175, p = 0.030). Since the p-value is below the conventional significance threshold of 0.05, Hypothesis H10 is supported, indicating that training support significantly moderates the relationship between GenAI intensity and perceived work productivity.

The negative interaction coefficient suggests that the positive relationship between GenAI intensity and perceived work productivity weakens as the level of training support increases. In other words, while GenAI intensity generally contributes to higher productivity perceptions, the strength of this relationship becomes less pronounced among employees who receive higher levels of organizational training support. This may indicate that employees with substantial training support rely less on GenAI tools alone to enhance productivity, as training programs may already equip them with the necessary skills and knowledge to perform their tasks effectively.

5.6. Goodness-of-fit indices.

The goodness-of-fit indices were examined to determine whether the proposed measurement model adequately fits the observed data. As shown in Table 10, the results indicate that the model demonstrates an acceptable and satisfactory fit according to commonly recommended SEM thresholds.

First, the chi-square (χ2) value of 621.272 with 381 degrees of freedom produced a χ2/df ratio of 1.631, which is below the recommended cutoff value of 3.0, indicating a good model fit. Although the p-value is significant (p < 0.001), this outcome is common in structural equation modeling, particularly when the sample size is relatively large, and therefore the chi-square statistic should be interpreted alongside other fit indices.

Second, the absolute fit indices suggest that the model adequately represents the observed data. The Goodness-of-Fit Index (GFI = 0.902) exceeds the recommended threshold of 0.90, indicating a good level of model fit. The Adjusted Goodness-of-Fit Index (AGFI = 0.880) is slightly below the recommended cutoff of 0.90, but it remains within an acceptable range and does not indicate serious model misfit.

Third, the incremental fit indices demonstrate strong model fit. The Normed Fit Index (NFI = 0.917) and Relative Fit Index (RFI = 0.905) both exceed the recommended value of 0.90, indicating satisfactory improvement of the proposed model compared with the independence model. Similarly, the Incremental Fit Index (IFI = 0.966), Tucker–Lewis Index (TLI = 0.961) and Comparative Fit Index (CFI = 0.966) are all well above the recommended threshold of 0.90, indicating excellent model fit.

Finally, the error-based fit indices also support the adequacy of the model. The Root Mean Square Error of Approximation (RMSEA = 0.041) is well below the recommended cutoff value of 0.08, and the Standardized Root Mean Square Residual (SRMR = 0.040) is also below the acceptable threshold of 0.08. These values indicate that the model provides a close approximation to the observed covariance matrix.

Overall, the combination of absolute, incremental and error-based fit indices suggests that the proposed measurement model demonstrates a good fit to the data, indicating that the hypothesized relationships between the latent constructs and their observed indicators are adequately supported.

Therefore, the measurement model is considered suitable for proceeding to the structural model analysis and hypothesis testing.

5.7. Artificial neural network (ANN).

An artificial neural network (ANN) analysis was conducted to complement the structural equation modeling results and to capture potential non-linear relationships among the study variables. A total of five ANN models were developed, each corresponding to key relationships in the conceptual framework. The dataset consisted of 368 valid cases, which were randomly divided into training samples (67.7%–72.6%) and testing samples (27.4%–32.3%), ensuring robust model validation and generalizability.

Model 1 examined the combined effects of perceived GenAI intensity (GenAII), perceived work productivity (PWP) and training support on intention to use GenAI (UGenAI). The model utilized four hidden neurons, indicating a relatively more complex structure to capture the joint influence of multiple predictors. The relatively balanced split between training (68.2%) and testing (31.8%) samples suggests good model stability and predictive capability.

Models 3 focused on the direct effects of GenAII on perceived job disruption (PJD) and employee anxiety, respectively. Both models employed a single hidden neuron, reflecting simpler relationships. The results indicate that GenAI intensity alone is a strong predictor of both job disruption and anxiety, consistent with the structural model findings.

Model 4 examined the influence of GenAI intensity on AI skills readiness, with two hidden neurons. This indicates that while the relationship is relatively straightforward, there are still underlying complexities in how exposure to GenAI contributes to the development of employees’ AI-related skills.

Across all models, the use of the hyperbolic tangent activation function in the hidden layer and the identity function in the output layer is consistent with best practices for continuous outcome prediction. The consistent performance across training and testing datasets suggests that the ANN models demonstrate good predictive accuracy and generalizability.

Overall, the ANN results corroborate the findings from the structural equation modeling analysis, while also highlighting the presence of non-linear relationships among key variables. In particular, the models confirm that GenAI intensity is a central predictor influencing productivity, job disruption, anxiety and AI skills readiness, thereby reinforcing its pivotal role in shaping employees’ responses to GenAI in the workplace (see Table 11 and Figures 7–10).

6. Discussion.

This study examined the dual consequences of generative artificial intelligence (GenAI) adoption in the workplace by simultaneously investigating its productivity-enhancing potential and its psychological implications for employees. Unlike much of the existing literature, which predominantly portrays GenAI as either a technological opportunity or a workplace threat, the present study demonstrates that these two perspectives coexist. The findings reveal that GenAI simultaneously promotes higher work productivity and stronger intention to use AI while also increasing perceptions of job disruption and employee anxiety. This evidence supports the view that organizational AI adoption should be understood as a socio-technical transformation rather than merely a technological implementation.

By integrating Technological Determinism Theory (TDT), Disruptive Innovation Theory (DIT), Innovation Resistance Theory (IRT) and Digital Divide Theory (DDT), this study explains how technological, organizational and psychological mechanisms interact to shape employees’ responses to GenAI. Consequently, the study extends existing research by offering a more comprehensive explanation of workplace AI adoption that incorporates both positive and unintended consequences.

The study found a significant positive impact of perceived GenAI intensity on perceived work productivity. The results support this hypothesis and are consistent with recent studies that demonstrate the productivity-enhancing potential of generative AI technologies in knowledge work environments. For example, experimental research shows that employees using generative AI tools such as large language models can complete writing, coding and analytical tasks faster and with improved quality compared with those who do not use AI assistance. However, the present study extends these findings in an important way.

Whereas most previous studies have been conducted in technologically advanced economies, this research demonstrates that comparable productivity benefits can also emerge within an emerging economy such as Somalia, where AI adoption remains at an early stage and digital infrastructure is comparatively limited. This suggests that employees’ perceptions of AI-enabled productivity improvements are not solely dependent on technological maturity but also on how effectively organizations integrate AI into everyday work processes. Similarly, field studies indicate that generative AI improves problem-solving efficiency and decision-making capabilities among knowledge workers.

These findings reinforce the assumptions of Technological Determinism Theory, which suggests that technological innovations reshape organizational practices and improve work outcomes by enabling new forms of efficiency and automation. Thus, employees who perceive higher levels of exposure to GenAI tools tend to report improved productivity and work performance.

The study also found a significant relationship between perceived GenAI intensity and perceived job disruption. The results validate this hypothesis and align with recent research suggesting that generative AI technologies can transform traditional job roles and task structures within organizations. Rather than indicating immediate job displacement, the findings suggest that employees perceive GenAI as accelerating organizational change, thereby increasing uncertainty regarding future work arrangements, required competencies and career progression. This distinction is important because perceptions of disruption may emerge long before actual workforce restructuring occurs. In line with Disruptive Innovation Theory, employees interpret AI technologies as catalysts for organizational transformation that challenge existing work practices and require continuous adaptation.

The findings therefore extend prior research by demonstrating that employees’ perceptions of disruption constitute an important psychological response to AI implementation, even in organizations where large-scale workforce replacement has not yet occurred.

The study also demonstrates that perceived job disruption significantly increases employee anxiety. This finding is consistent with previous studies reporting that technological uncertainty contributes to psychological stress and concerns regarding career stability. More importantly, the results illustrate the mechanism through which GenAI affects employee well-being. Rather than GenAI directly creating anxiety, employees first perceive disruption to their work environment, which subsequently generates emotional strain. This finding advances Innovation Resistance Theory by illustrating that resistance toward emerging technologies may originate from anticipated workplace changes rather than from the technology itself.

Employees appear to experience anxiety because GenAI challenges established routines, increases uncertainty regarding future employability, and requires continuous adaptation to evolving digital competencies. Consequently, organizations implementing GenAI should recognize that technological transformation inevitably involves psychological adaptation, making employee well-being a central component of successful AI implementation.

The findings further reveal that perceived work productivity significantly enhances employees’ intention to continue using GenAI technologies. This result aligns with technology adoption literature demonstrating that perceived performance improvements remain among the strongest predictors of behavioral intention. Beyond confirming previous evidence, the present study demonstrates that productivity functions as an important psychological mechanism linking AI exposure to future AI adoption. Employees appear willing to embrace GenAI not merely because the technology is available but because they experience tangible improvements in their daily work performance. This finding suggests that organizations seeking sustainable AI adoption should prioritize demonstrating measurable productivity gains rather than focusing exclusively on technological implementation.

The study also found that organizational training positively influences perceived work productivity. This finding supports recent research emphasizing that structured digital learning enables employees to integrate AI technologies more effectively into workplace activities. However, the findings also indicate that training serves a broader organizational function beyond improving technical competence. Effective training appears to reduce employees’ uncertainty, strengthen confidence in AI usage, and facilitate smoother integration of AI into routine work processes. Consequently, organizational learning should be viewed not simply as skill development but as an essential mechanism supporting successful digital transformation.

One of the most noteworthy findings of this study is that perceived GenAI intensity does not significantly influence AI skills readiness. This result contrasts with an implicit assumption frequently found in the digital transformation literature that increased exposure to advanced technologies naturally develops employee capabilities. Instead, the findings suggest that operational use of GenAI and genuine AI competency represent two distinct phenomena. Employees may extensively use AI systems to complete work tasks without acquiring deeper conceptual understanding, critical evaluation skills, or transferable AI related competencies. This distinction represents an important theoretical contribution because it extends Digital Divide Theory beyond issues of technological access to emphasize the importance of capability development.

In other words, access to AI technologies alone is insufficient to create AI ready employees unless organizations simultaneously invest in structured learning opportunities, continuous digital upskilling and long-term capability development strategies.

Another important contribution concerns the moderating role of organizational training support. Contrary to the conventional expectation that training consistently strengthens the positive effects of technology adoption, the findings reveal that training weakens the relationship between perceived GenAI intensity and work productivity. Although this result initially appears counterintuitive, several plausible explanations exist. Employees receiving extensive training may gradually develop greater confidence in performing complex tasks independently rather than relying heavily on AI assistance. Training may therefore shift employees from AI dependence toward AI complementarity, where AI serves as one resource among many rather than the primary driver of productivity.

Alternatively, employees undergoing intensive training may initially experience increased cognitive demands as they learn new systems and workflows, temporarily reducing the incremental productivity benefits associated with AI use. This finding therefore contributes to the emerging literature by demonstrating that organizational support mechanisms do not always amplify technology outcomes; instead, they may fundamentally reshape how employees interact with AI technologies.

Beyond the individual hypotheses, this study makes several broader contributions to the workplace AI literature. First, it develops and empirically validates an integrated theoretical framework combining TDT, DIT, IRT and DDT to explain both the technological and psychological consequences of GenAI adoption. Previous studies have typically relied on a single theoretical perspective to explain either AI acceptance or employee resistance. By integrating four complementary theories, the present study provides a more comprehensive explanation of how productivity gains, job disruption, employee anxiety, organizational support and behavioral intention interact within AI-enabled workplaces. Second, the study contributes valuable empirical evidence from Somalia, a context that remains significantly underrepresented in AI and organizational research.

Demonstrating similar patterns of productivity enhancement alongside increased employee anxiety suggests that the dual nature of GenAI extends beyond highly developed digital economies and is also evident within resource constrained environments undergoing early stages of digital transformation. This contextual contribution enhances the external relevance of existing AI adoption theories and broadens their applicability across diverse organizational settings.

Overall, the findings demonstrate that GenAI should neither be viewed exclusively as a productivity enhancing technology nor solely as a source of workplace disruption. Instead, its organizational impact is inherently multidimensional, simultaneously generating operational efficiencies while creating psychological, organizational and capability related challenges. Sustainable organizational value from GenAI therefore depends not only on technological implementation but also on employees’ preparedness, organizational learning, responsible change management, and continuous investment in workforce capability development. By presenting this balanced perspective, the study advances the emerging discourse on responsible workplace AI adoption and provides insights of how organizations can maximize AI’s benefits while mitigating its unintended consequences.

7. Managerial implications.

Based on the findings, the enhancement of employees’ work productivity through GenAI is critical, as it directly strengthens their intention to use GenAI technologies in the workplace. However, the results also demonstrate that this productivity gain coexists with psychological strain arising from perceived job disruption and anxiety. This duality implies that GenAI implementation is not merely a technological upgrade but a socio-technical transformation requiring integrated managerial strategies. Accordingly, organizations should approach GenAI adoption not only as an operational efficiency initiative but also as a workforce transformation process requiring strategic human resource management, organizational learning and employee well-being support.

First, organizations should actively invest in structured training programs and continuous learning initiatives. Although GenAI improves efficiency, employees may experience uncertainty and anxiety due to changing job roles. Providing hands-on training, AI literacy programs and upskilling opportunities can enhance employees’ confidence and reduce resistance toward GenAI. However, the moderation results of this study reveal that training support may weaken the positive impact of GenAI on productivity, suggesting that excessive or poorly designed training may overwhelm employees or disrupt workflow efficiency. Therefore, organizations should adopt targeted and practical training approaches that are aligned with employees’ job needs rather than generic or overly complex programs.

This finding provides an important managerial insight for organizations implementing AI systems: training effectiveness depends not only on availability but also on relevance, timing and alignment with employees’ actual work tasks. This finding is particularly relevant for developing economies where training resources are often limited and must be carefully optimized for impact.

Second, management should ensure clear communication regarding the role of GenAI in the workplace. Transparency about how AI will complement rather than replace employees’ roles can help reduce perceived job disruption and associated anxiety. This is especially important in contexts similar to Somalia and other emerging economies, where technological uncertainty and employment insecurity are more pronounced due to fragile labor market structures. Managers should therefore adopt participatory change-management approaches that involve employees in AI integration processes and clarify how AI technologies support rather than threaten long-term career development.

Third, organizations should foster a supportive digital work environment by providing adequate technological resources and encouraging a culture of experimentation and innovation. Employees should feel psychologically safe to explore and use GenAI tools without fear of failure or job insecurity. Such an environment is critical in low-resource settings where digital transformation is still emerging, and employee readiness varies significantly across sectors. The findings further suggest that organizations should establish internal AI governance mechanisms, including ethical AI usage guidelines, employee support systems and continuous monitoring of the psychological effects of AI-driven work transformation.

Finally, policymakers and organizational leaders should consider developing inclusive digital strategies that address disparities in digital skills and access. While training and AI readiness are important, this study suggests that they may not fully eliminate anxiety related to technological disruption. Therefore, broader initiatives such as organizational change management, career development support, and job redesign are necessary to ensure sustainable GenAI adoption. These implications are transferable to other developing countries undergoing early-stage AI adoption, where structural constraints amplify the socio-psychological effects of technological change.

From a policy perspective, governments and industry stakeholders should collaborate to develop national AI workforce strategies that prioritize digital inclusion, workforce reskilling, and responsible AI adoption practices to ensure that the benefits of AI transformation are broadly distributed across society.

8. Theoretical implications.

This study contributes to the emerging literature on generative artificial intelligence (GenAI) in the workplace by providing a comprehensive dual-pathway framework that integrates productivity enhancement and psychological disruption mechanisms. The key theoretical novelty of this study lies in demonstrating that GenAI operates simultaneously as an enabling and destabilizing force, rather than producing uniform positive or negative outcomes as assumed in traditional technology adoption models. In doing so, the study responds to recent scholarly calls for more context-sensitive and human-centered AI research that explains how technological transformation reshapes both organizational performance and employee experiences.

First, the findings provide empirical support for Technological Determinism Theory (TDT) by demonstrating that GenAI intensity significantly enhances employee productivity. This confirms that technological advancements shape work processes and performance outcomes. However, this study extends TDT by showing that technological effects are not deterministic in a linear sense, but are mediated by employee perceptions of disruption and anxiety. Thus, the study contributes to the evolving debate on digital transformation by demonstrating that the organizational consequences of AI technologies depend on both technological capability and employee psychological adaptation.

Second, the study extends Disruptive Innovation Theory (DIT) by demonstrating that GenAI does not only disrupt organizational structures at a macro level but also generates micro-level psychological disruption at the employee level, influencing perceived job security and emotional stability. This represents a refinement of DIT by incorporating human affective responses into disruption theory. Accordingly, the findings shift the theoretical focus of disruption research from purely structural and market-level outcomes toward employee-centered organizational consequences of AI-driven change.

Third, the study advances Innovation Resistance Theory (IRT) by identifying employee anxiety as a central mechanism explaining resistance to GenAI adoption. Unlike prior studies that treat resistance as behavioral non-adoption, this study conceptualizes resistance as an emotional and cognitive response embedded within active usage environments, where employees continue using GenAI despite psychological discomfort. This contribution is theoretically important because it demonstrates that technology acceptance and psychological resistance may coexist simultaneously within AI-enabled workplaces.

Fourth, the study contributes to Digital Divide Theory (DDT) by demonstrating that AI skills readiness and training support do not operate uniformly as enabling factors. Instead, their effects are context-dependent and may produce diminishing or substitution effects under high GenAI exposure. Moreover, the non-significant relationship between GenAI intensity and AI skills readiness challenges assumptions that technology exposure automatically leads to digital capability development. This extends DDT beyond access and capability to include ‘adaptive digital inequality’, where outcomes differ based on how individuals cognitively and behaviorally integrate AI tools.

Importantly, the study introduces and empirically supports the concept of a ‘productivity–anxiety paradox’ in GenAI adoption, where performance improvements coexist with psychological strain. This challenges conventional technology acceptance models (e.g. TAM and UTAUT), which generally assume that usefulness leads to uniformly positive behavioral intentions. The paradox identified in this study contributes to the growing scholarship on responsible AI and socio-technical systems by emphasizing that successful AI implementation requires balancing productivity outcomes with employee well-being and organizational sustainability.

Overall, by integrating TDT, DIT, IRT and DDT, this study provides a multi-layered theoretical explanation that bridges macro-level technological change with micro-level employee psychology. This integrated framework is particularly relevant for both developing and developed economies, where AI adoption is uneven and shaped by institutional readiness, workforce skill levels and organizational culture differences. Consequently, the study contributes to advancing interdisciplinary AI scholarships by linking technology adoption, organizational behavior, workforce transformation and digital inequality within a unified explanatory framework.

9. Limitations and future research directions.

The present study has several limitations that provide opportunities for future research. First, the generalizability of the findings may be limited, as data were collected within a specific developing economy context (Somalia). Given that AI adoption trajectories differ significantly across countries, particularly between developed, emerging and fragile economies, future studies should conduct cross-country comparative analyses to test the external validity of the proposed dual-pathway model.

Second, cross-sectional design limits the ability to capture temporal changes in employee perceptions and behavior. Since GenAI adoption is an evolving process, longitudinal research is needed to examine how productivity gains and anxiety evolve over time as employees progress from initial exposure to routine use.

Third, the study relied on self-reported survey data, which may introduce common method bias and limit behavioral precision. Future research should adopt multi-method approaches, including experimental designs, qualitative interviews and digital trace data (e.g. system usage logs), to triangulate findings and improve measurement robustness.

Fourth, although the model explains substantial variance in key constructs, it does not fully capture organizational and institutional-level factors that may shape GenAI adoption outcomes. Future studies should incorporate additional variables such as leadership style, organizational culture, regulatory environment and task complexity.

Fifth, while this study examined training support and AI skills readiness, the findings suggest these variables may operate in non-linear or counterintuitive ways. Future research should explore alternative boundary conditions such as psychological safety, trust in AI, perceived organizational support, and change readiness to better explain when GenAI produces positive versus negative employee outcomes.

Finally, future research should extend this work by examining sector-specific and country-specific differences. For example, industries such as healthcare, education and public administration may experience distinct patterns of GenAI adoption due to differences in risk sensitivity, regulatory constraints and skill requirements. Such comparative research would further refine the theoretical generalizability of the dual-pathway model across global contexts.

CRediT: Abdulkadir Jeilani Mohamud: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration; Abdifatah Nour Rage: Data curation, Investigation, Methodology, Project administration, Supervision, Validation; Ali Dahir Mohamed: Data curation, Formal analysis, Project administration, Supervision; Adam Mohamed Mohamud: Data curation, Investigation, Methodology, Supervision.

Disclosure statement appeared to influence the work reported in this paper.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not for profit sectors.

About the authors

Dr. Abdulkadir Jeilani Mohamud is a Senior Lecturer in the Faculty of Computer Science at Mogadishu University and the Department of Computer Science at Somali National University. He holds a PhD in Information and Communication Technology from Asia e University, Malaysia. His research interests include artificial intelligence, cybersecurity, information systems, educational technology, digital transformation, and information security management. He has published more than 22 articles in leading international journals, and his current research focuses on AI adoption, digital literacy, cybersecurity awareness, and technology-enabled learning in developing countries.

Mr. Abdifatah Nour Rage is a Senior Lecturer in the Faculty of Computer Science at Mogadishu University and the Department of Computer Science at Salaam University. He holds master’s degrees in Computer Science and International Relations and currently serves as Deputy Rector for Research and Development at Salaam University. His research interests include computer science, software development, higher education, digital transformation, and institutional development.

Dr. Ali Dahir Mohamed holds a PhD in Educational Psychology with a specialization in Personality Psychology. He is an experienced researcher and trainer whose work focuses on educational psychology, personality development, learning behavior, motivation, and academic performance. His research aims to enhance teaching and learning through evidence-based psychological approaches in educational settings.

Mr. Adam Mohamed Mohamud holds an MBA from Asia e University, Malaysia, and a bachelor’s degree in Economics from Mogadishu University. He is the Deputy Dean of the Faculty of Economics and Management Science at Mogadishu University and serves as a part-time lecturer. His research interests include economics, management, human resource management, organizational development, and higher education administration.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.