How does employee–AI collaboration influence employee job performance? A multilevel analysis of cognitive flexibility and empowerment climate

In recent years, organisations have increasingly turned to artificial intelligence (AI) to augment capabilities that extend beyond the limits of traditional tools and human expertise (Pereira et al., 2023; Jia et al., 2024). Broadly defined, AI refers to a branch of computer science dedicated to creating systems that simulate human cognitive functions, including learning, problem-solving, and decision-making (Armenia et al., 2024). Unlike conventional information technologies, AI possesses the ability to mimic or even exceed human cognitive performance in specialised tasks. Its distinguishing features include autonomy, opacity (or “black-box” logic), and the ability to evolve through machine learning, which sets it apart from rule-based, pre-programmed IT systems (Golgeci et al., 2025). Powered by machine learning and deep learning, these systems process and interpret external data, learn from prior experiences, and make autonomous decisions without explicit human instructions (Jia et al., 2024). As such, these systems are not merely passive tools, they can address complex, ambiguous problems and even generate novel outputs in response to user inputs.

A growing body of research highlights AI's significant and multifaceted effects on employees’ cognition and behaviour, touching on aspects such as autonomy, job meaning, cognitive load, and perceptions of job security (Golgeci et al., 2025; Zheng et al., 2025). These impacts arise from AI's potential to reshape how tasks are executed, redefine job roles, and shift employees’ sense of control over their work. On one hand, AI has been found to improve employee performance by automating repetitive tasks, reducing mental strain, and providing personalised assistance, factors that can affect innovation and service quality (Li and Ding, 2026; Zhang et al., 2026). On the other hand, emerging evidence also raises concerns around reduced autonomy, increased monitoring, and perceived unfairness, often linked to algorithmic decision-making and surveillance mechanisms (Hai et al., 2025; Jeong and Jeong, 2025). These contrasting findings underscore the complex roles of AI in shaping the employee experience and highlight the need for a more nuanced understanding of its implications in the workplace. In contrast to traditional notions of replacement or augmentation, collaboration with AI implies a more dynamic relationship in which human superiority is not always assumed, and the technology is not necessarily subordinate (Danatzis et al., 2025). Rather, roles and responsibilities between humans and machines can shift fluidly throughout a work process, depending on task demands and the comparative strengths of each partner. Given these fundamental changes, it is vital and timely to recognize employee–AI collaboration as a critical element of contemporary work (Bankins et al., 2023). In this sense, this collaboration is not merely a user-tool interaction but a synergistic partnership in which humans and AI draw on their complementary capabilities to achieve outcomes neither could accomplish alone (Kong et al., 2023; Li et al., 2025).

Despite growing interest in AI at work, research on employee–AI collaboration has largely concentrated on its risks, such as job displacement, work intensification, and employee anxiety (Hai et al., 2025; Yang and Jiang, 2025). While these concerns are legitimate, they overlook the reality that many employees are already learning to collaborate productively with AI and, in doing so, are nurturing new competencies and mindsets that positively relate to employee job performance (i.e., in-role and extra-role performance) (Chen et al., 2023; Przegalinska et al., 2025). This highlights a critical yet underexplored area of research, as existing studies have only begun to investigate the contributions of employee–AI collaboration to employee performance (Bai et al., 2025; Li and Ding, 2026), leaving essential dimensions of this relationship insufficiently theorised and empirically examined. Unpacking the mechanisms through which AI collaboration influences employee job performance is therefore essential for advancing theory and informing managerial practice (Bankins et al., 2023; Yang and Jiang, 2025). In particular, scholars call for greater attention to mediating and moderating mechanisms that explain when and how AI collaboration fosters job performance (Mo et al., 2024; Kumar et al., 2025).

To examine how employee–AI collaboration shapes employees' job performance (e.g. in-role and extra-role), this study draws on social exchange theory (SET) (Blau, 1964) and socio-technical systems theory (STS) (Manz and Stewart, 1997). These two theories provide distinct yet complementary lenses for understanding the interpersonal and organisational dynamics of working with intelligent systems. While SET helps explain the interpersonal and motivational dynamics of collaboration (Cropanzano et al., 2017), STS sheds light on the broader system-level conditions that influence how technology and people work together (Makarius et al., 2020). We posit that SET helps conceptualise collaboration as a form of reciprocal exchange in which employees contribute oversight, contextual expertise, and adaptive inputs, while AI systems offer cognitive support, task augmentation, and efficiency gains (Anthony et al., 2023; Armenia et al., 2024). STS complements this relational view by situating collaboration within the broader organisational system. It emphasises the interdependence between technical systems and social structures, advocating for joint optimisation, the idea that both must be designed and managed together to achieve effective outcomes (Manz and Stewart, 1997). This perspective is particularly relevant in AI-enabled environments, where the introduction of intelligent technologies often alters job roles, workflows, and team dynamics (Yu et al., 2023). Together, we argue that SET and STS enable a more holistic understanding of the collaboration with AI will shape employees' cognition and job performance.

Given AI's evolving nature and its disruption of traditional competence models, working with AI demands continuous adaptation and active cognitive demands (Pereira et al., 2023; Blaurock et al., 2025). Prior research on human-AI collaboration has typically conceptualised adaptability within models that assume a clear division of labour and a largely one-directional flow of input (Marvi et al., 2025). In these approaches, tasks are either delegated to AI or retained by humans, and interaction tends to follow a sequential pattern: AI produces outputs, and humans evaluate, adjust, or implement them. However, collaboration in practice rarely works this way. Instead, employees and AI often engage in a back-and-forth process where ideas are generated, refined, questioned, and improved through multiple rounds of interaction (Luan et al., 2025). This ongoing exchange calls for more than general adaptability. We therefore argue that cognitive flexibility, which is defined as an individual's awareness of available options, willingness to adapt, and confidence in their ability to do so effectively, could serve as a mediator that channels employee–AI collaboration into improved employee job performance (Martin and Rubin, 1995; Kiss et al., 2020). Unlike fixed technical skills or surface-level emotional responses, cognitive flexibility represents a dynamic and transferable capability that enables employees to interpret AI outputs critically, integrate them with contextual and experiential knowledge, and make nuanced decisions across varying situations (Xu et al., 2026). Drawing on SET, we further argue that cognitive flexibility captures the reciprocal value through employee–AI collaboration, enabling employees to respond constructively to uncertainty and make more effective use of AI as a resource tasks (Bankins et al., 2023). Ultimately, it amplifies the benefits of collaboration, equipping employees to sustain high performance in increasingly complex organisational environments (Golgeci et al., 2025).

Additionally, given the inherently social nature of contemporary work, the effects of employee–AI collaboration are likely to be shaped by the immediate social context in which employees operate, particularly their team climate (Kumar et al., 2025; Erengin et al., 2025). In this study, we focus on empowerment climate as a key team-level moderator. Defined as a shared perception that a work group encourages autonomy, voice, and initiative (Seibert et al., 2011), empowerment climate reflects the tone of everyday interactions and the extent to which employees feel supported to take ownership of their work. This makes empowerment climate particularly relevant for employee–AI collaboration, which often demands judgment, interpretation, and experimentation, activities that flourish when employees feel empowered (Zhang et al., 2026). While contextual factors such as leadership style or AI governance structures are undoubtedly important (Tsai et al., 2022; Koponen et al., 2025), they may vary across hierarchical levels or may not be inconsistently experienced within teams (Erengin et al., 2025). Empowering leadership, for example, captures what leaders do in their interactions with employees and may help to cultivate empowerment climate (Han et al., 2020). Yet empowerment climate goes beyond individual leader behaviour and reflects a shared, collective understanding that autonomy and influence are genuinely embedded in the work environment (Seibert et al., 2011). It operates at the level where AI collaboration most frequently takes place within teams and offers a more proximal and consistent influence on how employees engage with AI (Alexiev et al., 2020; Seibert et al., 2011). From STS view, such climate supports the joint optimisation of the technical and social subsystems by allowing employees to adapt the technology meaningfully into their workflows. From a SET perspective, it also reinforces reciprocal norms: when employees feel trusted and supported, they may be more inclined to put effort and creativity into making AI collaboration work, trusting that their contributions will be recognised and reciprocated. Thus, we argue that empowerment climate is not only theoretically appropriate but also practically salient in shaping the strength of the relationship between employee–AI collaboration and employees' outcomes.

By proposing and testing the theoretical model, this study makes several important contributions to the management literature. First, we propose a novel multilevel framework that advances understanding of the outcomes of employee–AI collaboration within organisational contexts (Suseno et al., 2022; Kumar et al., 2025). In doing so, we extend the employee–AI collaboration literature shifting the focus to the psychological mechanisms that enable employees to derive growth-oriented outcomes. Our study employs a cross-level design to explain how reciprocal exchanges between employees and AI systems translate into meaningful workplace outcomes.

Second, we address an important gap by identifying cognitive flexibility as a central mediating mechanism linking employee–AI collaboration to both in-role performance and extra-role performance. In doing so, the study responds to calls for deeper theoretical exploration of how and when AI integration influences employees' attitudes and behaviours (Kong et al., 2023; Yang and Jiang, 2025). By clarifying this mediating pathway, we contribute to a more nuanced understanding of the transition processes through which collaboration with AI impacts employees' cognitive states and, ultimately, their performance within and beyond formal role boundaries.

Third, the study highlights the role of team empowerment climate as a critical boundary condition that shapes the effectiveness of employee–AI collaboration. By conceptualising empowerment climate as a contextual resource, we show how team climates that foster autonomy and shared accountability amplify direct effect of employee–AI collaboration on cognitive flexibility, as well as the indirect effects of employee–AI collaboration on employees' job performance. This cross-level insight enriches management research by revealing how climate dynamics jointly shape employee cognition and behaviour in human-AI work systems (Koponen et al., 2025). Practically, it underscores the importance of cultivating empowering environments that not only enhance individual flexibility but also generate collective benefits for organisations navigating AI integration (Bankins et al., 2023; Zheng et al., 2025).

To better understand how employee–AI collaboration shapes job performance, this study integrates Social Exchange Theory (SET) and Socio-Technical Systems Theory (STS). This dual-theoretical approach allows us to capture both the relational and systemic dynamics that characterise AI-enabled work. Rooted in organisational behaviour research, SET provides a relational lens for understanding collaboration develop through repeated exchanges of valued resources governed by norms of reciprocity (Blau, 1964; Cropanzano and Mitchell, 2005). When individuals perceive support, trust, or investment, they are motivated to reciprocate through discretionary effort and constructive engagement (Cropanzano et al., 2017). Extending this logic to AI-enabled work, we conceptualise employee–AI collaboration as a reciprocal exchange process in which employees contribute domain expertise, contextual interpretation, and effort, while AI provides analytical capability, structured insights, and novel alternatives. Employees' behavioural responses are therefore shaped not only by technological functionality but also by how AI integration is interpreted as a signal of organisational support, augmentation, or substitution. In this sense, collaboration outcomes emerge from relational evaluation rather than purely technical efficiency (Bankins et al., 2023). When this dynamic unfolds in teams with strong empowerment climates, with autonomy and responsibility are supported (Seibert et al., 2011), employees are even more likely to engage in meaningful exchanges with AI systems that enhance both in-role and extra-role performance.

While SET explains motivational and behavioural dynamics, it does not fully account for how technology is embedded within organisational systems. STS addresses this limitation by conceptualising organisations as composed of interdependent social and technical subsystems that must be jointly optimised (Manz and Stewart, 1997; Makarius et al., 2020). Rather than treating AI as an external tool or environmental input, STS views it as structurally integrated within workflows, roles, and routines. The effectiveness of AI therefore depends on alignment between technological capabilities and social arrangements, including work design, team processes, and skill configurations. Without appropriate support, this may lead to resistance, cognitive overload, or disengagement (Jooss et al., 2025; Golgeci et al., 2025). STS, therefore, highlights the importance of building flexibility into both their technical systems and social structures for organisations; for example, by creating an empowering environment where employees feel confident experimenting with new technologies and integrating them into their workflows (Zheng et al., 2025; Makarius et al., 2020). In other words, when alignment is achieved, AI can augment human judgment and performance. When misaligned, it may generate resistance, overload, or disengagement. STS thus foregrounds co-evolution and mutual adjustment, highlighting that AI integration is not a unilateral imposition but a systemic reconfiguration of work.

This integrated perspective advances dominant approaches that have interpreted AI primarily through the lenses of resource allocation or control. For instance, extensions of the Job Demands-Resources model (Zhang et al., 2026) conceptualise AI as either a job resource that enhances motivation or a demand that generates strain under uncertainty. Similarly, algorithmic control research emphasises how AI systems monitor performance and redistribute authority, focusing on surveillance and autonomy reduction (Golgeci et al., 2025). Although these perspectives offer valuable insights, they share a structural orientation in which AI acts upon employees as a resource, demand, or control mechanism. Employees are largely positioned as recipients of technological influence. However, contemporary employee–AI collaboration involves iterative, bidirectional exchanges in which AI participates in ideation, analysis, and decision-making (Blaurock et al., 2025). The central phenomenon is therefore not whether AI creates demands or exercises control, but how human and technical elements co-adapt over time and how employees reciprocate within these evolving arrangements (Li and Ding, 2026). By combining STS's systemic interdependence with SET's relational reciprocity, we reconceptualise AI not merely as a technological variable but as a collaborative actor embedded within a socio-relational system. This framework provides a more dynamic and interactional account of employee–AI collaboration than existing models allow.

The rapid advancement and widespread adoption of AI technologies have reshaped how employees interact with technology in their daily work (Armenia et al., 2024; Pereira et al., 2023). Rather than functioning as passive tools, AI systems now serve as responsive, semi-autonomous collaborators actively participating in tasks, offering data-driven insights and adapting over time. Once initiated, AI can autonomously generate content, propose novel alternatives, and actively shape problem-solving pathways (Anthony et al., 2023). This evolution moves beyond earlier conceptualisations of technology-enabled or augmented learning, which framed technology primarily as a cognitive aid that enhances human learning by improving access to information (Luan et al., 2025). In such models, learning remains an intrapersonal process, agency resides largely with the human actor, and technology functions as a supportive instrument rather than an active contributor. By contrast, AI fundamentally reconfigures the role of technology in work processes (Chen et al., 2023). Human learning, when mediated by AI, becomes a process of co-creation in which outputs emerge through iterative exchanges between human and intelligent system (Anthony et al., 2023; Luan et al., 2025). As a result, employee–AI collaboration is emerging as a distinct form of human-technology interaction, defined by mutual responsiveness, shared problem-solving, and dynamic role-shifting between human and AI actors (Jia et al., 2024; Kong et al., 2023). Given this important reciprocity, a nuanced understanding of the psychological and behavioural outcomes of employee–AI collaboration is essential to unlocking AI-adopted businesses' full potential (Przegalinska et al., 2025; Blaurock et al., 2025).

Unlike traditional automation, collaboration with AI involves a continuous interplay in which both parties contribute to task outcomes. Drawing on SET, we argue that employee–AI collaboration stimulates ongoing reciprocal interactions that augment one another. While AI contributes novel information and structure, employees must engage with interact with non-human agents that provide novel perspectives, probabilistic reasoning, and data-driven recommendations, which often divergent from employees' established ways of thinking (Li and Ding, 2026; Chen et al., 2023). Specifically, the continual pressure to keep pace with rapidly evolving AI technologies can lead to cognitive overload, as employees may struggle to process the sheer volume of information and the accelerating rate of change (Zhang et al., 2026). This ongoing interaction introduces cognitive demands that extend well beyond basic tool use or procedural compliance. Employees are required to constantly shift between mental models, interpret unfamiliar data inputs, and remain receptive to alternative viewpoints (Jia et al., 2024) – processes that lie at the heart of cognitive flexibility.

Cognitive flexibility refers to an individual's capacity to restructure cognitive processes in response to changing environments or demands (Martin and Rubin, 1995). It involves the ability to shift attention, revise strategies, and integrate diverse types of information (Cox et al., 2025). Although related to constructs such as adaptability, learning agility, and proactive behaviour, cognitive flexibility is conceptually distinct in that it captures real-time mental agility rather than broader behavioural change or meta-level learning orientations (Smith and Watkins, 2024; Dheer and Lenartowicz, 2019). This construct is particularly important in contexts characterised by uncertainty, novelty, and complexity, which are increasingly present in AI-mediated work environments (Bankins et al., 2023). As employees collaborate with AI systems often encountering unpredictable outputs or algorithmic suggestions, they must decide when to trust, when to override, and how to adapt their thinking in response (Danatzis et al., 2025). These situations naturally call for and can further nurture cognitive flexibility.

STS further contributes to this understanding by emphasising the need for joint optimisation, ensuring that the technical systems are aligned with the social and cognitive capacities of employees. Poorly designed or rigid AI systems may stifle flexibility by overwhelming or constraining human agency (Yang and Jiang, 2025). In contrast, when the human-AI system is well-aligned, it not only supports performance but encourages cognitive expansion, enabling employees to explore new approaches, test hypotheses, and shift between mental models as needed (Chowdhury et al., 2022; Jooss et al., 2025). Thus, the system's design and social integration play a crucial enabling role in fostering the cognitive conditions for collaboration.

In sum, both SET and STS suggest that employee–AI collaboration functions as a stimulus for cognitive flexibility, by requiring employees to continuously engage with evolving, sometimes ambiguous, information and adapt their thinking accordingly. Collaboration that is relationally reciprocal (SET) and systemically supported (STS) offers the ideal conditions for fostering this critical cognitive capability. Therefore, we hypothesize that.

To obtain a more comprehensive understanding of the effects of employee–AI collaboration, this study focuses on two categories of employee performance: in-role performance and extra-role performance. The concept of in-role performance refers to the extent to which employees fulfil their prescribed duties in accordance with organisational norms and regulations, typically assessed through performance evaluations (Williams and Anderson, 1991). By standardizing the boundaries of employee responsibilities, in-role performance ensures the rational allocation of tasks and underpins the achievement of organisational objectives (Probst et al., 2025). Meanwhile, extra-role performance refers to discretionary and voluntary behaviours that extend beyond formal role requirements but contribute significantly to organisational success (Blader and Tyler, 2009). These behaviours reflect employees' proactive efforts to exceed prescribed duties, such as offering assistance to colleagues, proposing improvements, or initiating innovations (Probst et al., 2025).

We argue that cognitive flexibility plays a critical role in enhancing in-role performance. Employees with high in cognitive flexibility are more capable of interpreting unfamiliar information, adjusting task execution to new inputs, and solving problems without deviating from organisational standards (Kiss et al., 2020; Martin and Rubin, 1995). This ability is especially important in settings where job roles are fluid, customer demands are variable, and technological systems regularly introduce new types of task-related information (Xu et al., 2026). In applied contexts such as hospitality, healthcare, and service sectors, cognitive flexibility can support employees in managing real-time challenges and maintaining performance amid uncertainty. These conditions reflect a growing need for employees not just to comply with existing procedures, but to interpret and adapt their responses based on emerging task demands (Dheer and Lenartowicz, 2019; Cox et al., 2025). We thus suggest that.

Drawing on SET and STS, we suggest that employee–AI collaboration contributes to in-role performance by fostering employees' cognitive flexibility (Kiss et al., 2020; Probst et al., 2025). More specifically, when collaborating with AI, employees must shift between multiple cognitive tasks that are all within the boundaries of their formal responsibilities. For example, a customer service representative supported by an AI-powered chatbot system may need to shift quickly between resolving client inquiries, interpreting AI-suggested responses, and personalising communication to meet customer needs. SET helps explain the reciprocal nature of this collaboration, in which both parties provide inputs in a responsive mechanism. STS complements this view by highlighting the importance of joint optimisation between technological systems and social structures. Cognitive flexibility flourishes when the AI system is embedded in an environment that supports employee agency, interpretive discretion, and learning (Xu et al., 2026). When the socio-technical system is well-aligned, employees are not merely adapting reactively to technology but engaging in active sensemaking and continuously updating mental models, which enhances their ability to fulfil core job duties effectively (Makarius et al., 2020). Taken together, these arguments suggest that employee–AI collaboration cultivates cognitive flexibility, which in turn equips them to integrate AI-generated insights into their day-to-day responsibilities and fulfilling their in-role performance. The following hypothesis is thus proposed.

We also propose that cognitive flexibility further strengthens employees' extra-role performance. Employees with high cognitive flexibility are better equipped to explore alternative perspectives, identify latent opportunities, and generate novel solutions that support both their own tasks and broader team objectives (Dheer and Lenartowicz, 2019; Kiss et al., 2020). These individuals are more responsive to the needs of others, more likely to take initiative, and more effective at integrating diverse stakeholder perspectives, including managers, customers, and colleagues, into constructive and optimistic responses. In turn, such responses enable them to engage more fully in discretionary acts such as assisting colleagues, initiating improvements, and contributing to innovation (Li et al., 2025; Kim et al., 2022).

Consequently, employees who possess strong cognitive flexibility are therefore more likely to take initiative, exhibit a heightened sense of responsibility, and contribute meaningfully beyond their formal job roles (Martin and Rubin, 1995; Cox et al., 2025). In this way, cognitive flexibility fosters the conditions necessary for employees to engage in extra-role performance. Therefore.

Drawing on SET and STS, we further suggest that cognitive flexibility serves as a key mechanism through which employee–AI collaboration enhances extra-role performance. Working with AI often requires employees to engage with outputs that are generated through non-linear or unfamiliar decision rules and may not align neatly with their prior assumptions or expectations (Anthony et al., 2023; Przegalinska et al., 2025). Unlike traditional systems that reinforce established routines, AI often challenges employees to reconsider how they interpret tasks, evaluate problems, or arrive at decisions. This process can function as a cognitive prompt, encouraging employees to think more expansively and question pre-existing mental models (Zhang et al., 2026; Golgeci et al., 2025). Over time, repeated exposure to such cognitive challenges may reduce reliance on confirmatory biases and strengthen employees’ capacity to shift perspectives, integrate new information, and respond creatively to changing demands (Xu et al., 2026; Cox et al., 2025). In this way, employees who collaborate effectively with AI are less likely to experience cognitive overload or feel overwhelmed by the volume of AI-generated information, and are more likely to engage proactively, generate creative ideas, and support colleagues in dynamic work contexts (Blaurock et al., 2025). Thus, while STS underscores the need for an enabling environment that supports such exploration, SET reinforces the reciprocate mechanism through which employee–AI collaboration that enables employees to extend their contributions beyond prescribed duties, thereby enhancing extra-role performance in ways that create additional organisational value. We hypothesize that.

While employee–AI collaboration functions as an enabling resource that can stimulate cognitive flexibility, its effectiveness depends on the broader social context in which it occurs (Bankins et al., 2023). To capture this dimension, we suggest that team's empowerment climate moderates the relationship between employee–AI collaboration and cognitive flexibility. Team empowerment climate is defined as a shared perception of the extent to which a team supports, rewards, and expects employee empowerment (Seibert et al., 2004). In such climates, responsibility is allocated in ways that enable collective goal attainment, and sensitive information is openly shared across team levels (Chen et al., 2007; Seibert et al., 2011). Employees at lower levels enjoy substantial autonomy, with clearly defined areas of responsibility and accountability for decisions distributed beyond senior management alone (Kim et al., 2016; Seibert et al., 2004). By diffusing accountability, clarifying organisational vision, and reinforcing shared goals, empowerment climate frees up resources at the top while fostering a “negotiated order” in which interdependencies are transparent and managed constructively (Alexiev et al., 2020).

In climates characterised by high empowerment, employees are more likely to feel trusted, supported, and efficacious (Seibert et al., 2011). They are therefore less concerned about appearing incompetent or dependent when collaborating with AI and more willing to experiment, learn, and adapt (Yang and Jiang, 2025). Conversely, in low-empowerment climates, where autonomy is restricted and decision-making concentrated at higher levels, employees may perceive AI collaboration as risky or threatening. Under such conditions, employees may experience diminished awareness of available options, reluctance to adjust, and reduced confidence in their capabilities (Schmager et al., 2025), thereby weakening the positive influence of employee–AI collaboration on cognitive flexibility. This reasoning is consistent with SET, which posits that when organisations extend resources such as empowerment, employees reciprocate by investing greater effort and demonstrating enhanced cognitive flexibility in return (Cropanzano et al., 2017; Kiss et al., 2020). At the same time, STS highlights the need for an alignment in social and technical contexts. Employees will collaborate with AI more effectively once they are in a climate that supports for the integration of the technology (Zhang et al., 2026; Makarius et al., 2020). Accordingly, empowerment climate acts as a critical resource that strengthens the effect of employee–AI collaboration by fostering security, autonomy, and shared responsibility. In such environments, employees are more inclined to use AI actively, engage in collaborative learning processes, and leverage their autonomy to experiment with new approaches. Over time, these dynamics reinforce the reciprocal exchange between humans and AI, enabling employees to cultivate and expand their cognitive flexibility (see Figure 1).

Thus far, we have hypothesized the indirect effects of employee–AI collaboration on in-role and extra-role performance in Hypotheses 3 and 5. Combining this with Hypothesis 6, which proposes that empowerment climate moderates the relationship between employee–AI collaboration and cognitive flexibility, we suggest that empowerment climate also moderates the indirect effects of employee–AI collaboration on in-role and extra-role performance through cognitive flexibility, constituting a moderated mediating effect. Taken together, the following hypotheses are formulated.

This study focuses on service firms in Vietnam, which presents a theoretically meaningful context for examining employee–AI collaboration. As a rapidly digitising emerging economy, Vietnam is witnessing accelerated AI adoption across key service sectors such as finance, telecommunications, and e-commerce (Fang et al., 2026). These sectors demand high levels of customer responsiveness and require frontline employees to integrate AI-generated insights into judgment-based and interpersonal work, making human-AI collaboration both salient and consequential (Jooss et al., 2025; Wirtz and Stock-Homburg, 2025). Additionally, Vietnam's institutional and cultural context adds complexity to how such collaboration unfolds. Rooted in a collectivist and moderate power-distance culture, Vietnamese organizations often emphasise hierarchical control and deference to authority (Dang and Bertrandias, 2023), which may constrain critical factors for effective AI use and integration, such as employee autonomy and psychological empowerment.

While much of the existing research on employee–AI collaboration has been conducted in Western cultures with low power-distance (Chowdhury et al., 2022; Marvi et al., 2025), contexts as Vietnam offers a valuable contrast. It enables exploration of how employees navigate AI use in environments where hierarchical norms remain strong (Hofstede, 2001), yet where technological change increasingly calls for individual initiative and adaptability. In such settings, empowerment climates may differ in salience and meaning, requiring employees to navigate established power structures while learning to collaborate with intelligent systems (Nguyen and Nguyen, 2025). This tension makes cognitive flexibility especially important, as employees must not only process AI outputs but also reconcile them with managerial expectations, team norms, and customer needs, often without clear guidelines or precedents (Li and Ding, 2026). By examining these dynamics in the Vietnamese service sectors, this study contributes a more context-sensitive perspective on employee–AI collaboration and extends existing theory beyond Western-centric settings. It deepens our understanding of how culture, empowerment, and technology intersect to shape employee behaviour, while also offering insights relevant to other emerging economies undergoing similar digital transformations.

To ensure the sample's representativeness, we focused on Vietnamese service firms that had adopted AI technologies for a minimum of six months. The sampling frame was drawn from a professional Vietnamese human resources association comprising over 350 members, many of whom hold senior managerial roles, including C-level executives, business owners, vice-presidents, directors, and senior managers. The research team contacted CEOs or HR managers in the services industry through a combination of formal outreach methods, including phone calls, email invitations, virtual meetings, and in-person interactions. During these communications, participants were provided with detailed information sheets and the complete survey instrument to ensure transparency and alignment with the study's objectives. Upon receiving organisational approval, the survey was distributed to both managers and employees across various departments. Recruitment and data collection were conducted over a three-month period, from October to December 2024.

To balance methodological rigour with participant convenience, the survey was offered in both online and paper formats (Saunders et al., 2019; Neuman, 2011). Participants who preferred the digital option received a personalised link to the Qualtrics platform, where they could complete the questionnaire independently. For those more comfortable with a physical format, printed copies were delivered in person and later collected in sealed envelopes to ensure privacy and data protection. Each survey began with a cover letter explaining the purpose of the study, its practical implications, and the steps taken to protect confidentiality. To match responses between employees and their supervisors while preserving anonymity, each participant was assigned a five-digit identification code by a research team member. Before starting the main questionnaire, all participants were given a clear explanation of what AI is and how it is typically used in workplace settings. This ensured a shared understanding of the topic. They were then asked to indicate their experience with using AI in their job. Participants who reported no such experience were excluded from the rest of the survey to maintain the relevance and integrity of the data.

To address the research questions and reduce the risk of common method variance, we employed a multi-source data collection approach. By matching the employee-leader pairs of questionnaires, a total of 230 complete dyadic responses from 58 leaders and 230 employees made up our final research samples, yielding response rates of 72% of the 400 questionnaires distributed. The survey was conducted in two waves, spaced two weeks apart. During the first phase (Time 1), employees completed measures capturing demographic information, perceptions of the empowerment climate, and self-assessments related to employee–AI collaboration and cognitive flexibility. The average age of the employee participants was 30.05 years. Of the 230 employees, 89 (38.7%) identified as male and 139 (60.4%) as female. Their average organisational tenure was 3.35 years. In terms of education, most held a bachelor's degree (76.5%), followed by postgraduate degrees (14.3%), vocational qualifications (5.2%), and high school diplomas or below (3.5%). In the second phase (Time 2), leaders provided demographic details, along with information on their team's size and tenure. They also rated employees' in-role and extra-role performance. On average, managers were 36.5 years old and had an organisational tenure of 5.56 years. Educationally, 31 leaders (53.4%) held university degrees, while 24 (41.4%) had completed graduate education, reflecting the qualifications typically expected for managerial positions. The average team size was 3.97 members, and the average team tenure was 3.35 years (SD = 1.37). Descriptive statistics for the employee and leader samples are summarised in Table 1 and 2.

All the scales used in this study were developed and validated by previous research. Since the survey was conducted in Vietnamese and the measurements were developed in English, a back-translation process was performed to ensure the translations were equivalent (Brislin, 1986). The research applies a five-point Likert scale for all items in the questionnaire to measure, ranging from “1 = strongly disagree” to “5 = strongly agree.”

A five-item scale developed by Kong et al. (2023) was utilized to gauge employee–AI collaboration, with a sample item being: “AI participates in my problems, opportunities, or risk recognition process” (α = 0.867).

Cognitive flexibility was evaluated with a five-item scale developed by Kruglanski et al. (1993), with a sample item stating: “I always see many possible solutions to problems I face” (α = 0.855).

In-role performance was assessed using a five-item scale established by Janssen and Van Yperen (2004), with a sample item: “This employee meets all the formal performance requirements of the job” (α = 0.878).

Extra-role performance was measured with a three-item scale developed by Blader and Tyler (2009), with a sample item: “This employee often volunteers to do helpful things that are not required by the job description” (α = 0.868).

We measured empowerment climate through a three-item scale validated by Alexiev et al. (2020), with a sample item being: “In my team, we are allowed to define our own role and to pursue different roles” (α = 0.877).

Control variables: Demographic information at the individual level (i.e., age, gender, education background and working tenure) were measured for employees, as suggested in previous research (Mo et al., 2024). Meanwhile, team characteristics (i.e., team tenure and team size) were further taken as controlling variables.

To mitigate common method variance (CMV) (Podsakoff et al., 2003), we applied a combination of methodological safeguards during both the survey design and data collection phases. Our approach included the use of multiple data sources, time-lagged data collection, and several procedural and statistical remedies aimed at minimising bias.

First, all survey items were drawn from widely used and empirically validated scales, helping to ensure measurement reliability and reduce potential bias associated with newly developed or organisation-specific items (Creswell and Creswell, 2018).

Second, the survey design incorporated procedural techniques to mitigate common method bias. These included collecting data from both employees and their direct supervisors, and temporally separating the measurement of independent and dependent variables to reduce the likelihood of consistency motifs or patterned responses (Podsakoff et al., 2003). Moreover, responses were collected anonymously, encouraging honest feedback and reducing the impact of social desirability bias (Saunders et al., 2019; Creswell and Creswell, 2018).

Furthermore, we conducted confirmatory factor analyses and assessed the reliability and discriminant validity of the measurement model (Podsakoff et al., 2003). Specifically, both Fornell and Larcker (1981) criterion and the heterotrait-monotrait (HTMT) ratio indicated strong discriminant validity (Kline, 2023). As shown in Table 3, the square root of the AVE for each construct exceeded its inter-construct correlations, and all HTMT values remained well below the 0.85 threshold. We also checked for CMV using variance inflation factor (VIF) scores. All VIFs fell between 1.61 and 2.66, well under the recommended threshold of 3.3, suggesting minimal multicollinearity or common method bias (Kline, 2023). Finally, model comparisons in CFA revealed poorer fit when constructs were forced into fewer factors, further supporting discriminant validity (see Table 4). Taken together, CMV is not a major concern in this study.

To evaluate whether individual responses on empowerment climate could be meaningfully aggregated to the team level, we calculated the within-group agreement index (R_wg) and intra-class correlation coefficients [ICC (1) and ICC (2)]. The average R_wg value was 0.89, indicating acceptable within-group consensus (Humphrey and LeBreton, 2019). The ICC (1) was 0.16, and ICC (2) was 0.42. Although no universally accepted thresholds exist, both R_wg≥0.70 and ICC(1) > 0.05 are commonly cited as sufficient indicators to support aggregation (Bliese, 1998; Humphrey and LeBreton, 2019). While an ICC(2) value of 0.50 is sometimes referenced as a guideline for acceptable group-level reliability, it is not uncommon that previous studies aggregate individual-level data to the team level with considerably low ICC(2) values, with value around 0.30 (e.g. Jiang et al. (2024), Chatzi et al. (2023)). In addition, these less reliable means may arise from relatively small unit sizes (i.e., average of five team members) (Bliese, 1998). Therefore, it is widely recognised that lower ICC(2) values do not necessarily preclude aggregation, particularly when theoretical justification and strong within-group agreement are present (Bliese et al., 2018). In this study, aggregation was both conceptually grounded and empirically supported by the R_wg values, in line with practices adopted in previous multilevel research (Humphrey and LeBreton, 2019; Tsai et al., 2022).

Given the hierarchical structure of the dataset, where employees are nested within teams and the inclusion of mediating variables in the hypothesised model, we employed Mplus (version 8.0) to conduct multilevel structural equation modelling (MSEM) (Zhang et al., 2009; Eid et al., 2024). The analysis involved multilevel confirmatory factor analysis (CFA) to establish convergent and discriminant validity, followed by multilevel path analysis to evaluate the hypothesised relationships.

To assess cross-level mediation, we applied Monte Carlo simulation procedures with 20,000 replications to generate 95% bias-corrected confidence intervals, following the guidelines outlined by Preacher et al. (2010). All analyses were performed using Mplus 8.0 (Muthén and Muthén, 1998–2010), with robust maximum likelihood (MLR) estimation.

Model fit was assessed using multiple indices, including the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardised Root Mean Square Residual (SRMR), reported separately at the within-group (individual) and between-group (team) levels. To compare competing multilevel models, we conducted scaled chi-square difference tests in line with Kline (2023) recommendations.

As shown in Table 3, all CR values exceeded 0.70, and all AVE values were above the 0.50 threshold (Hair et al., 2019), indicating satisfactory convergent validity.

A multilevel CFA was conducted with four individual-level variables (i.e., employee–AI collaboration, cognitive flexibility, in-role performance and extra-role performance); and one group-level variable (i.e., empowerment climate). Results as shown in the Table 5 revealed the 5-factor model to have good fit (χ² = 3.982, df = 2; CFI = 0.993, TLI = 0.944, RMSEA = 0.066, SRMR for individual level = 0.036, SRMR for team level = 0.059).

This hypothesized model fits the data better than other alternative models (refer to Table 4). Since the model fits worsened as more variables were combined, discriminant validity was also confirmed.

Hypothesis 1 proposes that employee–AI collaboration is positively related to cognitive flexibility. As shown in the Table 5, the relationship between employee–AI collaboration and cognitive flexibility was statistically positive and significant (γ = 0.183, SE = 0.0.15, p = 0.000). Thus, Hypothesis 1 was supported.

Next, the study proposes that cognitive flexibility mediates the relationship between employee–AI collaboration and in-role performance. As shown in the Table 5, cognitive flexibility is positively and significantly related to in-role performance (γ = 0.159, SE = 0.069, p = 0.022), providing evidence consistent with Hypothesis 2. To further assess the indirect effect, we conducted a Monte Carlo bootstrapping procedure with 20,000 replications. As reported in Table 6, the estimated indirect effect was 0.029 (SE = 0.013, p = 0.025, 95% CI = [0.004, 0.055]). Since the confidence interval does not include zero, the results suggest a statistically significant mediating role of cognitive flexibility, offering support for Hypothesis 3.

Additionally, we propose that cognitive flexibility mediates the relationship between employee–AI collaboration and extra-role performance. As shown in the Table 5, cognitive flexibility is significantly related to extra-role performance (γ = 0.228, SE = 0.107, p = 0.033), providing support for Hypothesis 4. In addition, we conducted a bootstrapping test examine the indirect effect of employee–AI collaboration on extra-role performance through cognitive flexibility. The result in Table 6, based on Monte Carlo bootstrap simulation with 20,000 replications, showed that the estimated indirect effect was 0.042 (SE = 0.020, p = 0.039, 95% CI = [0.002, 0.082]. Since the confidence interval does not include zero, a mediating effect of cognitive flexibility is confirmed-supporting Hypothesis 5.

Hypothesis 6 posited that empowerment climate moderates the relationship between employee–AI collaboration and cognitive flexibility. As shown in Table 5, the interaction effect was statistically significant and positive (γ = 0.642, SE = 0.025, p = 0.000). The study further plotted the impact of employee–AI collaboration on cognitive flexibility at higher and lower degrees of empowerment climate (see Figure 2). The slope analysis showed that this relationship is stronger under high empowerment climate (simple slope = 1.056, p < 0.001) than under low empowerment climate (slope = 0.594, p < 0.001). This indicates that empowerment climate enhances the benefits of employee–AI collaboration on cognitive flexibility. The moderation effect supports the notion that contextual empowerment amplifies the adaptive potential of AI-enabled work. Therefore, empowerment climate was a moderator and Hypothesis 6 was supported.

The findings indicated that the indirect effect of employee–AI collaboration on in-role performance through cognitive flexibility was statistically significant under conditions of both high empowerment climate (B = 0.244, SE = 0.106, 95% CI [0.037, 0.451]) and low empowerment climate (B = 0.220, SE = 0.096, 95% CI [0.032, 0.409]). The index of moderated mediation (Difference 1) was also significant (B = 0.034, SE = 0.015, 95% CI [0.005, 0.063]), thereby supporting Hypothesis 7 (see Table 7).

Similarly, the indirect effect of employee–AI collaboration on extra-role performance via cognitive flexibility was statistically significant when empowerment climate was high (B = 0.331, SE = 0.166, 95% CI [0.006, 0.656]) as well as when it was low (B = 0.299, SE = 0.151, 95% CI [0.003, 0.594]). The index of moderate mediation (Difference 2) was significant (B = 0.046, SE = 0.023, 95% CI [0.001, 0.091]), confirming Hypothesis 8.

While these effects reached statistical significance, their magnitude should be interpreted with appropriate caution. The association between employee–AI collaboration and cognitive flexibility (γ = 0.183, SE = 0.015, p < 0.001), as well as the indirect effects on in-role (0.029) and extra-role performance (0.042), suggest that these relationships are meaningful but incremental. These patterns likely reflect the multifaceted nature of human-AI collaboration, where outcomes are shaped not only by technological interaction but also by individual differences and organisational context. Importantly, cognitive flexibility showed a stronger association with extra-role performance (γ = 0.228) than with in-role performance (γ = 0.159), suggesting that its relevance may be particularly pronounced for discretionary and proactive behaviours. (see Figure 3)

Drawing on a multilevel design with multisource and time-lagged data, this study investigated a moderated mediation model to clarify how employee–AI collaboration influences job performance through cognitive flexibility and empowerment climate. Underpinned by SET and STS, the study suggests that employee–AI collaboration when reinforced by an empowerment climate provides employees with valuable resources, including greater awareness of choices, willingness to adapt, and confidence in their capabilities. In turn, these resources contribute to higher levels of cognitive flexibility, which ultimately strengthens employees' job performance.

Although our findings support the proposed model, they should be interpreted within several boundaries. Participation was voluntary, and respondents with stronger interest in or experience with AI may have been more inclined to take part, raising the possibility of self-selection bias and relatively AI-ready contexts. Moreover, AI implementation can disrupt established routines and generate shifting role expectations (Kong et al., 2023; Zhang et al., 2026), which may independently influence employees' psychological states and behavioural outcomes beyond the mechanisms captured in our model. Our contribution therefore lies in clarifying how collaboration may translate into performance through cognitive flexibility, rather than fully accounting for antecedent selection or broader role restructuring processes. These considerations delineate the scope of the study and point to additional dynamics that warrant further investigation.

This study contributed to the employee–AI collaboration literature significantly. First, this research extends a body of research that has predominantly emphasized the negative consequences of AI integration, such as expediency, job insecurity, and burnout (Hai et al., 2025; Jeong and Jeong, 2025) by highlighting how positive psychological states, specifically cognitive flexibility, can emerge from collaboration with AI. This shift adds balance to the discourse, illustrating that collaboration with AI is not only a potential source of threat but also a catalyst for better job performance when implemented thoughtfully. In line with prior work (Przegalinska et al., 2025; Yang and Jiang, 2025), the findings suggest that when organizations design AI systems to complement and expand employee roles, employees are more likely to approach AI with an optimistic and flexible mindset. This argument resonates with emerging evidence from Marvi et al. (2025), reinforcing the view that cultivating positive attitudes is key to sustaining productive employee–AI collaboration. Moreover, the findings also address recent calls for research to identify and explain the mechanisms through which AI collaboration contributes to employees' ability to fulfil their job performance (Bankins et al., 2023; Bai et al., 2025).

Additionally, by identifying cognitive flexibility as a key mediating mechanism linking employee–AI collaboration to job performance, the study provides a deeper understanding of the potential positive benefits employees can derive from working with AI (Bankins et al., 2023; Schmager et al., 2025). This deepens our theoretical understanding of how AI collaboration translates into meaningful outcomes. While existing scholarship has often focused on cognitive overload and fragmentation (Kumar et al., 2025; Jooss et al., 2025), our findings suggest that AI can also enrich cognition via stimulating flexible, creative thought patterns that support both routine and discretionary work. These finding challenges prevailing assumptions that AI primarily displaces or constrains human cognition, supporting a more dynamic view in which collaboration with intelligent systems can promote employee learning, sensemaking, and adaptability over time (Marvi et al., 2025).

By adopting a multilevel perspective, this study extends our understanding of how team-level empowerment climate cascades to shape individual attitudes and behaviours in the context of AI adoption. Specifically, our findings provide evidence that empowerment climate functions as a salient resource that interacts with employees' cognitive flexibility, thereby creating the conditions necessary for effective employee–AI collaboration (Alexiev et al., 2020; Bankins et al., 2023). These dynamic advances our understanding of how supportive social contexts can buffer the potential constraints of hierarchical structures, especially in AI-augmented work environments where cognitive authority is increasingly distributed between human and machine actors (Makarius et al., 2020). It also responds to recent calls to examine team-level dynamics, which remain underexplored in human-AI systems (Erengin et al., 2025), yet are crucial for understanding how employees engage with intelligent tools in their day-to-day work.

Additionally, by combining SET and STS, this study advances a more integrated theoretical approach to employee–AI collaboration, addressing a gap in the literature where single-theory applications may fall short. While frameworks such as the Resource-Based View (Barney, 1991), the Job Demands-Resources model (Bakker et al., 2023), and the Technology-Organisation-Environment framework (Baker et al., 2011) have provided valuable insights into technology adoption and organisational adaptation, they often focus on either structural or individual-level mechanisms in isolation. Importantly, these models were primarily developed to explain human-human or human-technology interactions in contexts where both agents were presumed to be human or passive tools (Danatzis et al., 2025). As AI systems increasingly exhibit agency including adapting, responding, and learning over time, traditional theories may overlook the complexity of interactions in which one actor is non-human, yet still perceived as a collaborative partner (Zhang et al., 2026). Therefore, by integrating SET and STS, this study contributes to the theory in three ways. First, it reframes AI as a co-performing agent rather than a passive tool, enabling more nuanced investigation of mutual dependence and dynamic role adaptation. Second, it bridges micro-level and macro-level perspectives, linking interpersonal processes (e.g. employee cognition and behaviour) with system-wide factors (e.g. team climate), highlighting how their interaction shapes collaborative outcomes. Third, this integrative approach supports the development of a context-sensitive framework (Makarius et al., 2020; Suseno et al., 2022), recognising that collaboration outcomes are contingent on both the quality of exchange and the alignment between technological features and the surrounding social system. In doing so, the study responds to recent calls for research that captures the evolving, multilevel nature of human-AI relationships at work and contributes to a growing body of scholarship on how intelligent technologies are reshaping not only how we work, but with whom we work (Danatzis et al., 2025).

Our findings suggest three actionable strategies for day-to-day management to promote effective collaboration in AI-integrated work environments. First, managers must ensure that empowering climates offer more than rhetorical autonomy and provide tangible support for employees to navigate ongoing technological change. While AI enhances efficiency in structured tasks, human judgment remains critical to avoid overreliance on AI in situations requiring empathy, creativity, or contextual sensitivity (Armenia et al., 2024; Blaurock et al., 2025). Managers should therefore cultivate team norms that legitimise the questioning, interpretation, and refinement of AI outputs, empowering employees to apply discretion and intervene when appropriate (Yang and Jiang, 2025). At the same time, while our findings highlight the benefits of employee–AI collaboration within empowering climates, practical implementation is unlikely to be uniformly distributed across organisations. Access to empowering climates is often uneven across roles and teams, and exposure to AI tools may be concentrated in resource-rich units (Zhang et al., 2026). Such asymmetries can exacerbate existing skill gaps, limit participation in AI-driven innovation, and contribute to skill polarization (Li and Ding, 2026; Bankins et al., 2023). Managers should proactively identify empowerment and resource disparities and address them through inclusive training, participatory system design, and structured feedback mechanisms (Golgeci et al., 2025; Danatzis et al., 2025).

Additionally, collaboration with AI requires sustained cognitive effort. In less empowered climates, employees may interpret AI as a control mechanism rather than a developmental resource, heightening resistance and disengagement. The continuous adaptation demanded by evolving AI systems may also deplete employees' time and cognitive resources, reducing their willingness to experiment or explore novel applications (Zhang et al., 2026). To mitigate these risks, managers should allocate protected time for learning and experimentation, frame AI augmentation as supportive rather than punitive, maintain human oversight in key decision points, especially in high-stakes or customer-facing roles. Models such as “human-in-the-loop” allow for a healthy balance between automation and accountability (Wirtz and Stock-Homburg, 2025). HR practices can play a vital role in reinforcing employees' sense of ownership and strengthening their commitment via tailored support, such as structured onboarding and mentoring for employees with low AI confidence, and greater autonomy and resources for high-efficacy employees (Blaurock et al., 2025).

Third, effective human-AI collaboration requires rethinking job design and long-term capability development. Training should extend beyond technical instruction to include data literacy, ethical reasoning, and critical thinking, enabling employees to engage reflexively rather than defer automatically to algorithmic outputs (Chen et al., 2023; Yang and Jiang, 2025). Beyond organisational boundaries, policymakers and educators must invest in equitable lifelong learning systems to prevent widening employability gaps (Li and Ding, 2026; Kumar et al., 2025). These efforts are especially critical in emerging economies, where digital skill gaps and uneven access to technological tools can limit both individual and organisational adaptability (Duarte Alonso et al., 2025). In rapidly developing economies where digital transformation is accelerating but unevenly distributed, public-private partnerships are especially vital in extending access to training and technological resources. Without such systemic investment, the promise of AI-enabled collaboration risks benefiting only those already positioned to capitalise on it.

While this study makes meaningful contributions, it is not without limitations, which in turn highlight opportunities for future research. Firstly, our findings emerge from Vietnam, an emerging economy characterised by hierarchical and collectivist norms (Nguyen and Nguyen, 2025; Hofstede, 2001). Future work should examine whether the mechanisms identified here hold in lower power-distance or more individualistic contexts, such as the United States or Western Europe (Marvi et al., 2025).

Secondly, although our study focused primarily on the potential positive consequences of employee–AI collaboration through the cognitive flexibility pathway, prior research suggests that this collaboration can also have negative or null effects, such as cognitive overload or role ambiguity (Jeong and Jeong, 2025; Wirtz and Stock-Homburg, 2025). The observed positive relationships may also reflect unmeasured individual differences, such as psychological safety, or trust in AI, which could predispose employees to engage more constructively with intelligent systems. Future studies should therefore examine both enabling and disruptive pathways of employee–AI collaboration, ideally using longitudinal and cross-industry designs.

Third, regarding our measurement of employee–AI collaboration, while we employed a widely used scale in previous research (e.g. Kong et al. (2023), Li et al. (2025), Marvi et al. (2025)), we acknowledge that it does not fully differentiate between various forms or intensities of collaboration (e.g. decision support vs. automation). Given the complexity and evolving nature of AI roles in practice, future research would benefit from more fine-grained measures that distinguish between specific types of AI involvement.

Fourth, although the time-lagged design introduced temporal separation and helped reduce common method bias, the two-week interval may be insufficient to capture long-term developmental changes in cognitive flexibility or performance. In this study, we conceptualised cognitive flexibility as including context-sensitive shifts in thinking and performance as proximal behavioural enactment, both of which may vary in response to recent employee–AI interactions. Accordingly, our findings should be interpreted as reflecting short-term variations rather than long-term transformation. Future research employing extended longitudinal or multi-wave designs could further examine sustained cognitive and performance development over time.

Finally, although aggregation of empowerment climate was supported by strong theoretical rationale and satisfactory within-group agreement (Rwg), the relatively modest ICC(2) suggests caution in interpreting cross-level effects. Given that ICC(2) is highly sensitive to team size, which was naturally limited in our field setting, this limitation represents a common but important challenge in organisational research. We therefore encourage future studies to replicate in settings with larger or more stable team structures, which would allow for stronger cross-level inference and enhance the generalisability of our conclusions.

Taken together, future research should move beyond establishing whether employee–AI collaboration improves outcomes and instead examine how it evolves within broader socio-technical systems. Scholars might explore how AI-induced role restructuring and shifting decision authority reshape professional identity, how empowerment climates interact with unequal access to AI resources to affect participation and skill polarization, and how repeated human-AI exchanges contribute to the longer-term development of cognitive flexibility.

Thao Nguyen: Conceptualization, Data curation, Formal analysis, Methodology, Software, Visualization, Writing- original draft, Writing- review and editing. Trung Nguyen: Conceptualization, Investigation, Resources, Supervision, Validation, Writing- review and editing. Giang Hoang: Conceptualization, Methodology, Software, Supervision, Validation, Writing- review and editing. Alvedi Sabani: Conceptualization, Investigation, Supervision, Validation, Writing- review and editing.

Ethical considerations were integrated throughout the research design and implementation, including obtaining ethical clearance from the appropriate institutional review board and ensuring informed consent and anonymity during the survey deployment.