‘AI-Learner Partnership:’: Psychological Mechanisms and Developmental Trajectories of EFL Learners’ Writing Agency in GenAI-Assisted Courses

‘AI-Learner Partnership:’: Psychological Mechanisms and Developmental Trajectories of EFL Learners’ Writing Agency in GenAI-Assisted Courses | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article ‘AI-Learner Partnership:’: Psychological Mechanisms and Developmental Trajectories of EFL Learners’ Writing Agency in GenAI-Assisted Courses Lei Zhang, Chunli Jiang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7972095/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study investigates how EFL learners construct, negotiate, and develop writing agency when engaging with generative artificial intelligence (GenAI) tools in second language writing courses, conceptualizing these technologies as cultural mediational means within an integrated sociocultural and positive psychological theoretical framework. Eleven Chinese EFL university learners were observed over a 16-week period as they completed multiple academic writing tasks using GenAI assistance. Data collection employed a qualitative multi-method approach including screen recordings of writing sessions, stimulated recall interviews, reflective journals, and sequential drafts with AI interaction logs. Learners’ agency trajectories evolved through three developmental phases: initial exploration (characterized by novelty, dependency, and emergent self-efficacy), strategic adaptation (marked by selective tool use and growing writing-specific psychological capital), and deliberate appropriation (reflecting personalized integration and flourishing techno-authorial identity). Four patterns of agency manifestation emerged: transitional agency (shifting dependencies with emerging metacognitive awareness and increased resilience), distributed agency (strategic negotiation of textual authority fostering growth mindsets), reflective agency (critical evaluation of AI contributions coupled with self-determination), and transformative agency (reconceptualization of writer identities within human-AI collaboration leading to authentic engagement and eudaimonic well-being). Cognitive processes, affective dimensions, positive psychological resources, and task design characteristics significantly mediated how learners constructed agency through interactions with technological affordances. This study extends theoretical understanding of agency in GenAI-assisted educational contexts by illuminating the dynamic, developmental nature of writing agency and identifying specific patterns of agency manifestation in human-AI collaborative writing, while bridging sociocultural perspectives with positive psychological constructs that support optimal functioning. EFL writing generative artificial intelligence learner agency sociocultural theory human-AI collaboration academic writing Highlights This study explores how EFL learners construct and develop writing agency when engaging with generative AI (GenAI) tools in second language writing courses. Employing an integrated sociocultural and positive psychological framework , the study conceptualizes GenAI as a cultural mediational means shaping learners’ cognitive, affective, and agentic development. Learners’ agency followed a three-phase trajectory —exploration, adaptation, and appropriation—reflecting a dynamic developmental process of techno-authorial identity formation. Four distinct patterns of agency manifestation emerged: transitional, distributed, reflective, and transformative agency, each mediated by self-efficacy, resilience, and growth mindset. The findings advance theoretical understanding of agency in AI-assisted writing and offer pedagogical guidance for designing tasks that foster agentic engagement, well-being, and autonomous learning in GenAI-supported contexts. 1. Introduction Agency—the capacity to purposefully and meaningfully act within and upon one’s environment—remains a cornerstone of contemporary language teaching and learning scholarship (Ahearn, 2001 ; Deters et al., 2015 ; Duff, 2012 ; Larsen-Freeman, 2019 ; Tao & Gao, 2021 ). In second language (L2) education, agency is understood as socioculturally mediated: learners’ goal-directed actions are enabled, shaped, and at times constrained by the social, cultural, and increasingly technological environments they inhabit (Archer, 2010 ; Giddens, 1984 ; Priestley et al., 2015 ). As agency is relational and emergent—manifesting in negotiation with others, sociotechnical conditions, and the learner’s own evolving identity—its conceptualization and enactment are continuously revisited amid educational innovation (Larsen-Freeman, 2019 ; Roe & Perkins, 2024 ). The rapid integration of generative artificial intelligence (GenAI) in educational contexts has brought the question of learner agency to the forefront in new and complex ways (Niekerk et al., 2025 ; Lingard, 2023 ; Molenaar, 2022 ). GenAI technologies, powered by advanced natural language processing and machine learning, offer learners unprecedented resources for language production and revision, from sophisticated grammar correction and style adaptation to the generation of coherent, context-sensitive academic prose ((Niekerk et al., 2025 ; lingard, 2023 ; Molenaar, 2022 ; Chapelle & Sauro, 2017 ). These affordances promise deeper personalization and self-directed learning—but simultaneously raise concerns about overreliance, diminished independent thinking, and the erosion of learner autonomy (Lo et al., 2024 ; Zhang & Tur, 2024 ). Indeed, the permeability between human and machine contributions in writing, what Warschauer ( 2005 ) called the emergence of new “literacy ecologies,” compels educators and researchers to reexamine foundational notions of authorship, voice, and intentionality in digitally mediated settings. Underlying this transformation is a fundamental shift in the relational landscape of agency. Within Vygotsky’s sociocultural theory, all higher-order learning is mediated through cultural tools (Lantolf & Thorne, 2006 ). GenAI, in this sense, functions as an advanced mediational means that not only supports but transforms the very nature of the academic writing process—reshaping the division of labor (Engeström, 2001 ) between the actions of the learner and the affordances or constraints imposed by the technology (Wertsch, 1998 ). Unlike earlier digital tools, GenAI’s capacity to generate, critique, and even co-create text at a human-like level forces a reconsideration of where learner agency resides, how it is enacted, and how it evolves over time. It also raises the specter of “machine agency”—the degree to which the technology itself shapes, redirects, or even substitutes for human intent and authorship (Dattathrani & De’, 2023 ). Despite increasing theoretical attention to these dynamics (Szabo & Szoke, 2024 ; Roe & Perkins, 2024 ), empirical studies remain limited, especially regarding how language learners negotiate agency in everyday, longitudinal engagement with GenAI systems. Most existing literature has focused on initial user perceptions, functionality of AI tools, or teacher perspectives, leaving under-examined the temporal and processual aspects of agency construction as learners appropriate, resist, and adapt to GenAI in the context of complex academic tasks. Furthermore, little is known about the reciprocal interplay among human agency, technological affordances, sociocultural resources, and learners’ cognitive and affective processes—a gap recognized as pressing by educational policymakers and international organizations (UNESCO, 2023). To address this gap, this study adopts a sociocultural perspective to explore how EFL learners in China exercise, negotiate, and develop their writing agency over an extended period of GenAI-supported writing courses. 2. Literature Review 2.1 Writing Agency in Second Language Learning Agency has emerged as a key concept in second language acquisition, reflecting broader social turns in the field that recognize learners as active participants rather than passive recipients in educational processes (Duff, 2012 ; van Lier, 2008 ). Ahearn ( 2001 ) defines agency as “the socioculturally mediated capacity to act” (p. 112), emphasizing that individual intention always operates within socially structured constraints and affordances. In language learning contexts, agency manifests through learners’ strategic efforts to appropriate linguistic resources and participate meaningfully in various communities (Lantolf & Pavlenko, 2001; Lave & Wenger 1991 ; Rainio & Hilppö 2017 ). Writing agency specifically refers to the capacity of writers to make conscious choices, exercise control over the writing process, and express individual voice and identity through text (Tardy, 2016). For second language writers, this agency is complexly mediated by multiple factors, including linguistic proficiency, cultural-rhetorical traditions, institutional expectations, and available technological resources (Yi, 2013). Sociocultural approaches to writing agency emphasize that it emerges through participation in literacy practices, with writers gradually internalizing the values, strategies, and discourses associated with particular writing communities (Prior, 2006 ). Research on second language writing agency has highlighted several key dimensions. Tardy (2016) examined how L2 writers develop discoursal voice through strategic control of generic conventions. Baker ( 2014 ) investigated how writers navigate competing institutional demands while maintaining personal intentions. Yi & Angay-Crowder ( 2016 ) explored how digital technologies create new spaces for agency expression, particularly for multilingual writers who may feel constrained in traditional academic contexts. However, as Ding ( 2008 ) notes, agency in L2 academic writing is often characterized by tension between conformity to established discourse norms and the writer’s desire for self-expression. A significant gap in this literature concerns how emergent AI technologies—with their capacity to generate human-like text—transform the conditions for agency expression. Traditional views of writing agency presuppose human writers as primary agents who may employ various tools and strategies, but remain the central decision-makers in text production. As Haas (1996) argued in her examination of writing technologies, “writing is never simply a cognitive process, never simply a technological process, but is instead, always, a socio-technological-cognitive activity” (p. 27). GenAI technologies complicate this formulation by introducing a quasi-agentive element into the writing process itself. 2.2 Technology-Mediated Writing from the Lens of Sociocultural Theory Sociocultural theory provides a theoretical framework for understanding how tools and technologies fundamentally transform the nature of human activity rather than simply enhancing existing capabilities (Vygotsky, 1978 ; Wertsch, 1998 ). From this perspective, technologies are not merely instruments but mediational means that reshape cognitive processes, creating what Luria ( 1976 ) termed “historically formed biases of mind” through their regular use. Applied to writing technologies, sociocultural approaches examine how digital tools create new conditions for literacy development through mediated activity. Early work by Haas (1996) demonstrated how word processing technologies transformed writers’ conceptual and physical interactions with text. More recently, Zheng & Newgarden ( 2012 ) explored how digital environments create new affordances for language learners to develop agency through meaningful participation in valued activities. A central concept in sociocultural approaches to technology is the dialectical relationship between users and tools. As Wertsch ( 1998 ) argues, “agent and mediational means are best thought of as existing in a dialectical tension, in which neither can be analyzed or understood in isolation from the other” (p. 34). This perspective suggests that when EFL learners engage with GenAI tools, both student and technology mutually transform each other through their interaction. Learners appropriate technological affordances in individualized ways, while simultaneously adapting their writing practices to accommodate the tool’s constraints and possibilities. Several studies have applied sociocultural perspectives to examine technology-mediated writing in second language contexts. Jin & Deifell ( 2013 ) investigated how online translation tools function as mediational means in L2 writing, finding that learners developed strategic competence in tool use over time. Similarly, Li & Zhu ( 2013 ) examined collaborative writing in wiki environments, demonstrating how digital affordances transformed patterns of peer interaction and joint knowledge construction. The emergence of GenAI systems represents a qualitatively different category of mediational means in writing, however. Unlike previous technologies that primarily augmented specific aspects of the writing process, GenAI tools can participate in multiple dimensions simultaneously—generating ideas, structuring arguments, producing grammatical sentences, and even mimicking authorial voice. This multifunctionality creates what Engeström ( 2001 ) might describe as contradictions within the activity system of academic writing, as traditional divisions of labor and community norms are disrupted by the introduction of a new technological actor with quasi-agentive capabilities. 2.3 Re-conceptualizing agency in the age of GenAI: GenAI-Supported Writing The advent of generative artificial intelligence (GenAI) has catalyzed an essential reconsideration of agency in educational contexts, particularly in domains of academic writing where both human authorship and technological mediation converge. Recent theoretical advances increasingly recognize the need for an ecological and distributed perspective on agency—one that acknowledges complex interplays between human intention and technological participation in learning processes (Archer, 2010 ; Dattathrani & De’, 2023 ;Kaptelinin & Nardi, 2006 ; Ma & Chen, 2024 ). Theoretical frameworks from Giddens ( 1984 ) and Archer ( 2010 ) foreground the dynamic tension between structure and agency, wherein individuals’ capacities for meaningful action are simultaneously enabled and constrained by social structures—structures which are themselves recursively transformed through agentic practice. In digital environments, this dialectic becomes further nuanced as technologies such as GenAI function not merely as passive tools but as active participants in the construction and distribution of agency, giving rise to emergent socio-technical opportunity structures (Dattathrani & De’, 2023 ; Lee et al. 2023 ; Rose & Jones, 2005 ). This ecological conceptualization acknowledges agency as inherently relational and emergent—not a fixed property possessed by individuals, but a dynamic achievement intimately entangled with the affordances and constraints of technological and sociocultural environments (Emirbayer & Mische, 1998 ; Priestley et al., 2015 ). Moreover, agency in technological contexts is mediated by cognitive and affective dimensions—self-efficacy beliefs, motivational orientations, and emotional responses (Bandura, 2001 )—as well as by temporal-relational dynamics through which learners navigate between past experiences, present possibilities, and anticipated futures (Emirbayer & Mische, 1998 ). Thus, agency in GenAI-mediated language learning emerges as multifaceted and complex, constituted through intricate negotiations among human capacities, technological affordances, and sociocultural resources. The empirical examination of human-AI collaboration in writing contexts has begun to illuminate these dynamics. Cowan et al. (2021) demonstrated that AI language models, positioned as creative collaborators, can simultaneously inspire novel compositional pathways while inadvertently constraining creative exploration, contingent upon writers’ appropriation strategies. Clark et al. ( 2018 ) introduced the concept of “creative co-authorship” between humans and AI, emphasizing writers’ development of new metacognitive dimensions as they negotiate which AI-generated content to adopt, modify, or reject—a process that resonates with sociocultural notions of appropriation (Wertsch, 1998 ). In language education specifically, Feng et al. ( 2022 ) revealed how second language writers strategically deploy AI writing assistants while actively maintaining textual ownership, suggesting complex negotiations of control and assistance. Despite these important contributions, significant theoretical and empirical gaps persist in our understanding of agency in GenAI-supported academic writing. First, extant research has predominantly employed short-term experimental designs that capture initial interactions but fail to trace developmental trajectories as learners’ relationships with AI tools evolve over sustained engagement. As Warschauer ( 2005 ) contends, the transformative impact of new technologies on literacy development and agency can only be fully comprehended through longitudinal investigations that examine how users appropriate technological affordances within authentic contexts of meaningful practice. Second, studies have often conceptualized agency primarily through individualistic lenses focused on personal autonomy and control, neglecting the distributed, relational character of agency emerging through dynamic interactions among writers, AI systems, institutional contexts, and discourse communities (Wertsch et al., 1993). Third, research has typically examined either cognitive or sociocultural dimensions in isolation, without adequate attention to how these dimensions interact within integrated theoretical frameworks. Thus, the study addresses these limitations by conceptualizing GenAI writing tools as sophisticated cultural mediational means that transform the conditions of possibility for agency in academic writing. Through a longitudinal sociocultural lens, the study investigate how EFL university learners develop, negotiate, and redistribute agency in sustained engagement with GenAI-supported second language writing courses. Two research questions addressed: How do EFL students’ patterns of agency development and negotiation evolve over sustained engagement with GenAI in second language writing courses ? In what ways does the interplay of cognitive processes, affective dimensions, and contextual factors shape the construction and distribution of agency in GenAI-Supported Writing ? 3. Research methods This study employed a qualitative case study approach to investigate EFL learners’ writing agency construction in AI-supported academic writing. Case study methodology was selected for its capacity to examine complex phenomena within their authentic contexts (Yin, 2018 ), allowing for in-depth exploration of the “how” and “why” questions regarding learner agency in technology-mediated writing environments. The research design was informed by sociocultural principles, particularly attention to mediated action, historical development, and the interconnection between individual cognition and social practice (Lantolf & Thorne, 2006 ). 3.1 Research Context and Participants The study was conducted in an undergraduate English writing course at a university in China where students engaged in a series of GenAI-supported second language writing courses over the course of a semester. Participants included 11 EFL undergraduate students (7 males, 4 females). Participants were academically motivated students with intermediate to advanced English proficiency levels. All students had completed prior English writing courses but reported this course as their first experience using AI tools for academic writing purposes. The course instructor integrated GenAI tools (primarily Doubao) as optional resources for students to use throughout the writing process, from initial planning through drafting and revision. 3.2 Data Collection The primary data source for this study consisted of semi-structured interviews conducted with each participant at the conclusion of the semester-long AI-assisted writing course. The semi-structured interview protocol was designed specifically for this study. The full English version of the interview questions has been provided as a Supplementary File (Appendix A). Each interview lasted 30 to 45 minutes, was conducted in the participants’ native language (Mandarin Chinese), and yielded a total of 59,613 words of transcribed data. The interview protocol was designed to probe four key dimensions informed by the research questions: Temporal development of AI tool use and agency perceptions over the semester Cognitive and affective experiences when engaging with AI tools Responses to different AI-supported task designs Perceptions of human-AI collaboration and shared authorship In addition to interview data, supplementary sources were collected, including participants’ self-reported summaries of their emotional responses to AI-assisted writing activities and coded analyses of self-regulated writing strategies. The analysis of writing strategies categorized participants’ approaches into four dimensions: cognitive strategies (e.g., text processing, content retention), metacognitive strategies (e.g., idea planning, goal-setting, self-monitoring and evaluation), social behaviors (e.g., peer learning, feedback processing), and motivational regulation (e.g., interest enhancement, motivational self-talk, emotional control). This triangulated data collection approach provided a comprehensive understanding of learners’ agentic engagement and adaptive strategies in AI-supported academic writing contexts. 3.4 Data Analysis Interview data were analyzed using thematic analysis informed by sociocultural theory principles. Following Braun & Clarke’s ( 2006 ) six-phase approach, the process included: (1) familiarization with the data through repeated reading, (2) generation of initial codes, (3) searching for themes, (4) reviewing themes, (5) defining and naming themes, and (6) producing the report. Throughout this process, sociocultural concepts such as mediation, appropriation, internalization, and activity systems served as analytical lenses. The coding procedure followed a three-stage process—open coding, axial coding, and selective coding—to systematically develop themes from the raw data, as illustrated in Table 1 . Table 1 Three-Stage Coding Process with Examples from Interview Data Open Codes (with Interview Excerpts) Axial Codes Selective Codes (Themes) Most of the time, I directly let AI revise for me, then I look at why it made those revisions (S1-Interview-2503) Initial dependence Developmental Trajectories of Agency When I first used AI, I didn’t know how to write prompts, fortunately the teacher provided a guide document (S10-Interview-2503) Progression from initial exploration Lack of technical proficiency, users' unfamiliarity with AI use (S5-Interview-2503) through strategic adaptation Initially unfamiliar with AI use, at first the low scores made me think my writing level wasn't very high (S7-Interview-2503) to deliberate appropriation No, I select a few feedback suggestions to implement. Because I feel I should preserve some of my own elements (S10-Interview-2503) Critical evaluation Cognitive Mediations of Agency If I feel the sentence it provides deviates from what I want to express, I might not adopt it (S10-Interview-2503) Metacognitive awareness Critically accepting AI feedback (accepting all grammar suggestions, screening vocabulary) (S6-Interview-2503) strategic knowledge acquisition, Insisting on independent conception, maintaining reservations about AI outlines, believing one’s own thought process is clearer (S6-Interview-2503) critical evaluation, With AI I feel more relaxed because AI can help in many ways (S2-Interview-2503) Affective responses Affective Mediations of Agency The most obvious emotion would be a sense of achievement; there’s a notable feeling of accomplishment after improving an essay (S2–Interview-2503) Emotional responses to feedback, Negative emotions were fewer, mainly manifested as confusion and puzzlement over AI’s inaccurate or overly broad suggestions (S4-Interview-2503) changing self-efficacy beliefs, The use of AI enhanced the interviewee’s self-confidence and sense of achievement, changed their negative attitude toward writing (S3-Interview-2503) evolving motivational orientations This systematic coding approach enabled us to trace development patterns across participants while preserving individual variation in agency construction. The open codes, derived directly from interview excerpts, captured participants’ experiences in their own words. These were then grouped into conceptually related axial codes representing broader patterns of engagement with AI tools. Finally, selective codes integrated these patterns into comprehensive themes addressing our research questions about agency development in AI-assisted writing contexts. For agency trajectory analysis, individual case narratives were constructed for each participant, mapping their relationship with AI tools over time. Cognitive and affective mediators were identified through axial codes representing metacognitive processes, strategic behaviors, and emotional responses. Task design impacts were analyzed by comparing coded segments related to different writing assignments and AI interaction structures. Human-AI relationship dynamics were examined through selective codes reflecting perception, boundaries, and identity positions within technology-mediated writing practices. 4. Findings The analysis reveals how EFL learners constructed, negotiated, and transformed their writing agency through engagement with AI tools in second language writing courses. 4.1 Developmental trajectories of writing agency Participants demonstrated distinct yet overlapping developmental trajectories in their relationships with AI tools and associated expressions of agency. Through the lens of sociocultural theory, these trajectories revealed the progressive internalization of AI-mediated writing practices and the gradual transformation from other-regulation to self-regulation (Lantolf & Thorne, 2006 ). In early encounters with AI tools, most participants exhibited limited agency, characterized by tentative exploration and high reliance on AI-generated content. This initial stage revealed significant variation in attitudes, with some approaching AI with curiosity while others expressed skepticism or technical discomfort. As one paritipant noted, “ When I first used AI, I didn’t know how to write prompts, fortunately the teacher provided a guide document ” (S10-Interview-2503). During this phase, participants typically deferred to AI authority, accepting suggestions with minimal critical evaluation—what Vygotsky ( 1978 ) would characterize as other-regulation, with AI functioning as the more capable other directing the writing process. As participants gained experience, they developed more strategic approaches characterized by selective adoption of AI suggestions and increased metacognitive awareness. This transition typically emerged after 2–3 writing tasks, when participants began critically evaluating AI contributions against their own intentions: “ I select a few feedback opinions to implement. Because I feel I should preserve some of my own elements, if I think mine is better, I keep it ” ((S10-Interview-2503)). This stage revealed the development of what Wertsch ( 1998 ) terms mastery—knowing how to use cultural tools effectively—as distinct from appropriation, which involves making tools one’s own. Participants developed evaluation criteria, typically accepting grammar corrections while being more selective about vocabulary and content recommendations. In later stages, many participants (particularly those reporting medium to high enjoyment) demonstrated sophisticated agency through deliberate appropriation of AI affordances for personalized goals. This stage featured intentional prompting strategies, metacognitive reflection, and integration of AI tools into individualized writing processes. As S7 described: After receiving feedback, I first look at which dimension I scored lowest in, and focus on modifying areas where I lost the most points, referring to AI feedback suggestions during revision. Then I send the modified content back to AI to see if the score has improved, and if not, I further question the AI, asking more specifically how I should revise this part. This multi-stage process reveals sophisticated agency through strategic tool use and proactive dialogue with AI. This developing expertise represents what Kaptelinin ( 2005 ) characterizes as the appropriation of technological affordances to expand one’s action capabilities. Individual trajectories varied significantly based on participants’ prior attitudes toward writing, technological self-efficacy, and perceived writing goals. High-enjoyment participants typically moved more quickly from exploration to appropriation, while others remained longer in the exploratory phase. Some exhibited non-linear trajectories, moving between stages based on task contexts or personal factors. This variation aligns with Rogoff’s (2003) view of development as transformation of participation in culturally valued activities, with individual pathways shaped by personal histories and contextual factors. 4.2 Cognitive and affective mediations of writing agency construction Participants’ writing agency construction was mediated by interconnected cognitive and affective processes that shaped their engagement with AI tools. These mediations reflect Vygotsky’s ( 1986 ) concept of perezhivanie—the unity of intellectual and emotional aspects in development—revealing how agency emerges through the dynamic interplay of thinking, feeling, and action in technology-mediated environments. The cognitive dimension of agency construction manifested through several interrelated processes that evolved as participants gained experience with AI writing tools. Metacognitive awareness emerged as a foundational element, with participants developing increasingly sophisticated understanding of their writing processes and how AI could support them. As S7 reflected: “I think about what issues AI feedback identified in my previous essays, and in my next writing, I try to avoid making those mistakes. I also consciously set writing goals for myself, such as avoiding repetitive words as much as possible in this essay, or incorporating some of the advanced sentence patterns I’ve accumulated.” (S7-interview-2503) This illustrates how AI feedback prompted metacognitive regulation through goal-setting and strategy development, creating what Flavell ( 1979 ) terms metacognitive experiences that shape future task approaches. As participants engaged with AI, they selectively internalized linguistic resources from these interactions, representing the Vygotskian transition from interpersonal to intrapersonal functioning. Many described deliberately studying AI-suggested vocabulary: “Yes, I will try to remember those good expressions ChatGPT gives me.” (S10-interview-2503) “I try to remember some of the language expressions AI provides, but I haven’t specifically organized or memorized the advanced expressions it offers.” (S7-interview-2503) This strategic knowledge acquisition transformed external AI suggestions into internal linguistic resources that participants could deploy independently. The development of critical evaluation capacities further distinguished more agentic participants, who articulated clear criteria for assessing AI suggestions rather than accepting them uncritically: “Insisting on independent thinking, maintaining reservations about AI outlines, believing one’s own thought process is clearer.” (S6-interview-2503) “For example, sometimes it gives me suggestions on how to write a certain sentence. But if I feel that the sentence it provides deviates from what I want to express, I might not adopt it.” (S10-interview-2503) This critical stance reflects Wertsch’s ( 1998 ) distinction between mastery and appropriation, with participants maintaining distance from aspects that conflicted with their intentions. The most sophisticated cognitive mediation involved personalized tool appropriation, where participants customized AI interactions to serve individual goals. S8 described learning to craft specific prompts to elicit targeted feedback, while S7 developed multi-stage processes alternating between self-revision and AI consultation. These approaches transformed generic AI affordances into individualized resources for writing development, reflecting what Kaptelinin ( 2005 ) characterizes as the appropriation of technological affordances to expand one’s action capabilities. Affective processes were equally crucial in mediating agency construction, intertwining with cognitive dimensions to shape engagement patterns. Emotional responses to AI feedback varied substantially but followed discernible patterns. Positive emotions—particularly satisfaction and reduced anxiety—typically accompanied experiences of increased efficiency and mastery: “Yes, with AI I feel more relaxed because AI can help in many ways, making the writing process less difficult. The most obvious emotion would be a sense of achievement; there’s a notable feeling of accomplishment after improving an essay.” (S2-interview-2503) “ Positive emotions dominate. Although initially I felt a bit anxious because low scores made me think my writing level wasn’t very high, with more use I discovered that AI can both point out my errors and help improve my writing level, while also saving considerable time. So compared to these advantages of AI, that small anxiety is negligible, and positive emotions are predominant.” (S7-interview-2503) These positive emotional experiences created reinforcing cycles of engagement with AI tools. Negative emotions typically emerged when participants encountered discrepancies between AI feedback and their intentions, or when they perceived AI limitations: “Negative emotions were fewer, mainly manifested as confusion and puzzlement over AI’s inaccurate or overly broad suggestions, with anxiety significantly reduced due to AI’s involvement.” (S4-interview-2503) Self-efficacy beliefs shifted over time, with many describing initial doubts followed by increased confidence as they learned to work effectively with the technology: “The use of AI enhanced the interviewee’s self-confidence and sense of achievement, changed their negative attitude toward writing, and made them more willing to engage in writing.” (S3-interview-2503) This trajectory aligns with sociocultural perspectives on self-efficacy as developing through successful tool-mediated activity rather than existing as a stable internal trait (Lantolf & Thorne, 2006 ). Motivational orientations evolved as participants developed more sophisticated relationships with AI tools. Several reported transitioning from extrinsic motivations (completing assignments efficiently) to more intrinsic orientations focused on writing improvement and self-expression: “My goal is to express ideas in very clear, precise language. I think in terms of expressing ideas with clear and accurate language, I have made progress.” (S10-interview-2503) This shift toward intrinsic motivation aligned with increased perceptions of agency, as participants moved from viewing AI as primarily reducing workload to seeing it as supporting their personal development as writers. The findings suggest that cognitive and affective processes operated in dialectical relationship, mutually reinforcing agency development. Participants who developed sophisticated metacognitive awareness typically also reported positive emotional experiences with AI tools, creating virtuous cycles of engagement. Conversely, those expressing persistent concerns about dependency often maintained more limited cognitive engagement with AI affordances, using tools primarily for surface-level editing rather than deeper writing development. This integrated cognitive-affective dimension underscores Vygotsky’s holistic view of development, emphasizing the inseparability of thought and emotion in the construction of agency within technology-mediated learning environments. 4.3 Task design influences on agency manifestation Task design significantly influenced how participants exercised and developed agency in AI-supported writing. This influence reflects activity theoretical perspectives on how structural elements of learning environments constrain and enable particular forms of participation (Engeström, 2001 ). Through systematic analysis of participant experiences, three key aspects of task design emerged as particularly consequential for agency development: the degree of structure in AI-assisted activities, the positioning of AI within the writing process, and the nature of evaluation mechanisms. The degree of structure in tasks created varied affordances for agency expression across the learning trajectory. Highly structured activities with specific AI interaction protocols provided necessary scaffolding for initial agency development but sometimes limited more advanced expressions of agency. As one participant explained: “I gradually used AI software functions under the teacher’s guidance, such as grammar checking, sentence diversification, and so on." (S4-interview-2503) This structured approach helped novice users develop fundamental competencies with AI tools through guided practice. In contrast, less structured tasks that invited exploratory AI use revealed greater variation in agency manifestations. Some participants flourished with this autonomy, developing sophisticated and personalized interaction strategies: “During writing class, I first ask it to write an outline, then write a complete essay based on the outline, and afterward have it polish my work, checking if there’s anything that needs revision.” (S10-interview-2503) However, not all students thrived with minimal structure. Those with less technological self-efficacy sometimes struggled with unstructured assignments: “Lack of technical proficiency, users’ unfamiliarity with AI use, leads to confusion and difficulties during the usage process.” (S5-interview-2503) These differential responses suggest that optimal task structuring for agency development may follow what Collins et al. (1989) described as a scaffold-and-fade approach, with initial structure gradually giving way to greater autonomy as students develop expertise in AI interaction. The positioning of AI within the writing process—whether as planning aid, drafting assistant, or revision tool—significantly affected how participants manifested agency in their writing practices. Students generally reported stronger expressions of agency when tasks positioned AI as a planning resource or revision aid rather than as a primary text generator. This preserved students’ sense of authorial control while leveraging AI’s analytical capabilities: “I first think about the logical flow of the essay myself, then send the essay topic requirements to AI to see what logical structure it generates, supplementing my existing logical framework with AI’s suggestions. The logical structure provided by AI is just auxiliary.” (S7-interview-2503) This approach reflects what Wertsch ( 1998 ) termed “authoritative stance," with participants maintaining primary authorship while using AI as a resource. In contrast, participants who began by generating full AI drafts often reported weaker agency expressions initially: “AI will give me some materials and such that I can directly use." (S2-interview-2503) This reliance on AI-generated content positioned students more as editors than authors. However, many participants developed increasingly agentic approaches over time even when using AI for full draft generation, learning to critically evaluate and substantially revise AI content as their expertise grew. Task designs also varied in their evaluation mechanisms, with significant consequences for agency development and metacognitive awareness. AI scoring and detailed feedback provided valuable information for self-regulation but sometimes created dependency when positioned as the authoritative evaluation source: “I don’t think AI scoring is very good, though I might be misjudging. But my intuition is that the essay, honestly, was one that AI helped me revise, and then it gave it a low score, though I thought the essay was actually quite good. But I don’t really understand AI.” (S1-interview-2503) This quote reveals the confusion that can arise when AI simultaneously serves as both writing assistant and evaluator, creating contradictions that undermine students’ developing sense of agency. In contrast, tasks that combined AI feedback with teacher assessment or peer discussion typically supported stronger agency development by creating opportunities for critical evaluation of AI contributions: “Feedback handling: Critically accepting AI feedback (accepting all grammar suggestions, screening vocabulary); addressing AI deficiencies through teacher communication.” (S6-interview-2503) This multi-source feedback approach encouraged students to develop their own evaluative criteria rather than deferring entirely to AI assessment. Tasks that explicitly prompted reflection on the human-AI writing process—rather than focusing exclusively on the final product—further supported metacognitive awareness and agency development: “I consciously set writing goals for myself, such as avoiding repetitive words as much as possible in this essay, or incorporating some advanced sentence patterns I’ve accumulated.” (S7-interview-2503) The findings on task design align with Deci & Ryan’s (2000) self-determination theory, suggesting that tasks supporting autonomy, competence, and relatedness foster more agentic engagement with AI writing tools. The optimal approach appears to be tasks that provide sufficient structure to support competence while allowing space for autonomy and encouraging reflective awareness of the human-AI relationship. Such designs can scaffold the transition from other-regulation to self-regulation in technology-mediated writing environments, supporting what Vygotsky termed the zone of proximal development—the gap between what learners can accomplish independently and what they can achieve with appropriate support. This socio-culturally informed approach to task design recognizes that agency is not simply an individual attribute but emerges through the dynamic interaction between learners and the structured environments in which they operate, including the technological tools that mediate their writing activities. 4.4 Negotiated agency in Human-AI Partnerships The study found that participants’ views and practices towards AI were both complex and dynamic, directly shaping their agency manifestations. Most commonly, learners perceived AI as a powerful tool that could efficiently address much of their writing needs: “ Using AI definitely has more benefits than drawbacks; it can solve at least 60–70% of the problems at hand, and I can find solutions for the remaining issues myself” (S1-interview-2503)—and this instrumental approach often corresponded with a clear sense of authorial control and selective adoption of AI suggestions. Some, particularly those who engaged more deeply with AI, developed a more collaborative stance, describing dialogical processes where they negotiated meaning and refined ideas in partnership with AI: “ AI’s feedback speed...I find the feedback very detailed ”(S10-interview-2503). Others, especially in early stages, deferred to AI as an authority, accepting revisions with little critique: “ Most of the time, I directly let AI revise for me... ” (S1-interview-2503). Notably, many participants’ perspectives evolved over time, blending these roles as they became more familiar with AI’s strengths and limits. Negotiation of contribution boundaries emerged as a central theme. Learners increasingly distinguished between aspects of writing to keep personal and those to delegate to AI, such as reserving content and logic for themselves but using AI for language enhancement and revision. As one participant noted, “ Independently conceptualizing logical thought patterns with AI as auxiliary support, monitoring writing progress through feedback and goal setting” (S7-interview-2503), while another balanced independent ideation with AI-provided scoring and revision support (S8). This selective, often task-dependent appropriation reflected growing evaluative criteria and confidence— “ No, I select a few feedback suggestions to implement. If I think mine is better, I keep it and don’t change it ” (S10-interview-2503). Identity positioning was also fluid. Some learners insisted on maintaining independence and clear boundaries— “ Insisting on independent conception, maintaining reservations about AI outli nes” (S6-interview-2503)—while others viewed themselves as part of a productive partnership, e.g., “ AI can help me correct grammar errors and polish my completed essays... ” (S7-interview-2503). Feelings of dependency, particularly early on, were voiced: “ Feeling highly dependent...may create user dependency, affecting independent thinking ability ” (S5-interview-2503). However, most evolved toward more autonomous or collaborative identities as experience grew. Overall, agency in human-AI writing partnerships was the outcome of ongoing negotiation—participants continuously redefined the role of AI, the boundaries of its contribution, and their own identity as writers. As participants’ experience with AI deepened, their agency shifted from passive acceptance toward strategic, reflective, and evaluative engagement, revealing agency as an emergent, contextual, and relational process shaped by both personal intention and the affordances of AI technology. 5. Discussion The findings illuminate how EFL learnersconstruct, negotiate, and develop writing agency when engaging with GenAI tools in academic writing contexts. In this section, we interpret these findings through sociocultural theoretical lenses, discussing their implications for understanding learning dynamics in AI-mediated writing environments. 5.1 The evolution of writing agency in prolonged GenAI engagement (Addressing RQ1) Our findings demonstrate that EFL students’ writing agency in GenAI-supported contexts is a dynamic and developmental phenomenon, not a static trait. Learners usually progressed through a three-phase trajectory: 1) initial exploration, marked by reliance and experimentation; 2) strategic adaptation, characterized by selective and thoughtful tool use; and 3) deliberate appropriation, where GenAI becomes an integrated, personalized cognitive tool. This pattern aligns with Vygotsky’s ( 1978 ) principle that higher-order functions emerge through socially mediated activity and the gradual internalization of external tools (Lantolf & Thorne, 2006 ). However, the process is highly individualized. Students’ movement across these phases was neither unidirectional nor uniform, being shaped by factors such as prior writing experience, technological self-efficacy, and institutional expectations. This observation reflects Valsiner’ s (1997) “bounded indeterminacy” and echoes recent findings by Lee & Wang ( 2024 ) and Aydin & Akyüz ( 2022 ), who noted that agency development in AI-rich environments is marked by personalized, recursive shifts rather than linear progressions. The data support the argument by Chen et al. (2024) that flexible, adaptive pedagogies are necessary to cater to such diversified development. Moreover, the evolution of agency was not merely the result of tool exposure but also continual negotiation and praxis within authentic academic activities. This challenges the binary view of agency as either present or absent and instead reveals it as a situated, evolving achievement that occurs as learners appropriate GenAI’s mediational means into their own writing practices (Sun et al., 2023 ; Trust et al., 2023 ). 5.2 The interplay of cognitive, affective, and contextual factors (Addressing RQ2) The construction and distribution of agency in EFL students’ AI-supported writing developed through the intertwined influences of cognitive, affective, and contextual factors. Cognitively, learners who exhibited strong metacognitive awareness (such as evaluating GenAI feedback, transferring learning across contexts, and strategic task planning) were able to exert greater agency, moving from passive recipients of AI output to critical co-authors. Recent studies (Lin & Lee, 2022 ; Zou et al., 2021 ) confirm that metacognition is a key predictor of effective and agentive AI tool usage in writing. Affective experiences like confidence, anxiety, and technological motivation proved equally salient. Positive emotional responses—often linked to mastery moments or successful negotiation with GenAI—facilitated reflective engagement, while ongoing anxiety or fear of over-reliance restricted agentive growth. These findings are consistent with Saghafian et al. ( 2022 ) and Kim & Reeves ( 2023 ), who highlight the mediating effects of affect on technology adoption and creative autonomy in academic literacy contexts. The contextual and relational dimension was no less critical. The design of writing assignments, positioning of AI within the process, and the quality of scaffolding either expanded or limited students’ opportunities for agentive action. Tasks that began with tightly structured AI use but faded support as learners gained expertise created productive “zones of proximal development” (Kohnke et al., 2022 ; Reich et al., 2023 ). Furthermore, when GenAI was positioned as a revision aid or dialogic collaborator rather than a text generator, learners demonstrated stronger sense of authorship and self-regulation—reflecting Bakhtin’s ( 1981 ) ideal of internally persuasive discourse and recent distributed cognition frameworks (de Bot & Zhang, 2022 ; Zheng et al., 2022 ). In sum, agency in GenAI-Supported writing emerges through a dialectical movement: learners continuously navigate and negotiate personal goals, social roles, technological affordances, and educational constraints. This confirms the argument by Brynjolfsson et al. ( 2023 ) that distributed agency in AI-supported contexts is always contingent, relational, and open for pedagogical optimization. 6. Conclusion This study reveals that the development of writing agency among EFL learners in GenAI-assisted academic contexts is a dynamic, non-linear process, shaped by sustained engagement with technological tools, task design, and sociocultural relationships. Agency emerges not as a fixed trait but through evolving trajectories characterized by initial dependency, strategic adaptation, and, for some, the deliberate appropriation of AI affordances. Both cognitive and affective mediators—such as metacognitive awareness, critical evaluation, and emotional responses—interact dialectically, enabling learners to negotiate meaningful partnerships with AI while maintaining autonomy and personal authorship. These findings underscore the necessity of flexible pedagogical scaffolding and thoughtfully designed tasks that facilitate reflective, dialogic interaction with AI. Effective writing instruction in AI-assisted environments should recognize the relational and distributed nature of agency, allowing for diverse learner pathways and critical engagement with technology. As GenAI continues to evolve, educators are encouraged to foster technological literacy, encourage reflective tool use, and design adaptive supports that nurture empowered and adaptive agency in academic writing. Declarations Funding The author(s) received no financial support for the research. Author Information Authors and Affiliations School of Foreign Languages, Beijing University of Technology, Beijing, People’s Republic of China, Lei Zhang & Chunli Jiang Contributions Lei Zhang: conceptualization, investigation, and draft the original manuscript writing; Chunli Jiang: reviewing the manuscript. All authors reviewed the findings and approved the final version of the manuscript. Corresponding author Correspondence to Chunlijiang Ethics Declarations Ethics approval and consent to participate This study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the University Ethics Committee for Research Involving Humans at Beijing University of Technology. All participants received detailed information about the study and voluntarily provided written informed consent prior to participation. Consent to participant Informed consent was obtained from every participant provided. Human ethics and consent to participate declarations Not applicable Consent to publication Not applicable Conflict interest The authors have no potential conflicts of interest to disclose. Competing interest The authors declare no competing interests. References Ahearn, L. M. (2001). Language and agency. Annual Review of Anthropology , 30(1), 109-137. Archer, M. S. (2010). Routine, reflexivity, and realism: Three lectures on structure and agency . Oxford University Press. Aydin, S. & Akyüz, S. (2022). Learner agency in digital language education: A systematic review. ReCALL , 34(1), 23-44. Baker, W. (2014). “It’s not their job to share content”: A case study of the role of teaching assistants in the diffusion of knowledge and the enhancement of research-teaching linkages in undergraduate courses. Higher Education, 67(5), 581-594. Bakhtin, M. M. (1981). The dialogic imagination: Four essays (M. Holquist, Ed.; C. Emerson & M. Holquist, Trans.). University of Texas Press. Bandura, A. (2001). Social cognitive theory: An agentic perspective. Annual Review of Psychology , 52(1), 1–26. Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology , 3(2), 77-101. Brynjolfsson, E., Ikeda, S., & Liang, J. (2023). Augmenting human learning and agency with AI: Emerging evidence and considerations. Harvard Business Review , 101(3), 74-79. Chapelle, C. A., & Sauro, S. (Eds.). (2017). The handbook of technology and second language teaching and learning . John Wiley & Sons. Chen, W., & Xie, H. (2024). Digital literacy and agency in AI-enhanced academic writing: Evidence from ESL classrooms. Language Learning & Technology , 28(1), 33-48. Clark, E., Ross, A. S., Tan, C., Ji, Y., & Smith, N. A. (2018). Creative writing with a machine in the loop: Case studies on slogans and stories . In 23rd International Conference on Intelligent User Interfaces (pp. 329-340). Dattathrani, S., & De’, A. (2023). Technomediated agency: Reconceptualizing authorship in human-AI collaborative writing. Computers and Composition , 68, 102814. De Bot, K., & Zhang, L. (2022). Distributed cognition in digital academic writing: Human–machine partnerships in practice. Language Teaching Research , 26(4), 524-538. Deters, P., Gao, X., Miller, E. R., & Vitanova, G. (Eds.). (2015). Theorizing and analyzing agency in second language learning: Interdisciplinary approaches . Multilingual Matters. Ding, H. (2008). Negotiating voice and agency in L2 academic writing: Tensions between discourse conformity and self-expression. Journal of Second Language Writing , 17(3), 101-120. Duff, P. A. (2012). Identity, agency, and second language acquisition. Routledge. Emirbayer, M., & Mische, A. (1998). What is agency? American Journal of Sociology , 103(4), 962-1022. Engeström, Y. (2001). Expansive learning at work: Toward an activity theoretical reconceptualization. Journal of Education and Work , 14(1), 133-156. Feng, L., Wong, L. H., & Chen, W. (2022). Effects and moderating factors of automated writing evaluation on L2 learners’ writing self-efficacy and behaviors. Computers & Education , 176, 104356. Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist , 34(10), 906-911. Jin, L., & Deifell, E. (2013). Foreign language learners’ use and perception of online dictionaries: A survey study. MERLOT Journal of Online Learning and Teaching , 9(4), 515-533. Giddens, A. (1984). The constitution of society: Outline of the theory of structuration . Polity Press. Kaptelinin, V. (2005). The object of activity: Making sense of the sense-maker. Mind, Culture, and Activity, 12(1), 4-18. Kaptelinin, V., & Nardi, B. A. (2006). Acting with technology: Activity theory and interaction design . MIT Press. Kim, Y., & Reeves, T. (2023). Affective agency in digital literacy practices. Language Learning & Technology, 27(2), 121-138. Kohnke, L., Zou, D., & Zhang, R. (2022). Scaffolded feedback and student agency in AI-supported EFL writing. ReCALL, 34(2), 165-182. Lantolf, J. P., & Thorne, S. L. (2006). Sociocultural theory and the genesis of second language development . Oxford University Press. Larsen-Freeman, D. (2019). On language learner agency: A complex dynamic systems theory perspective. Modern Language Journal , 103(S1), 61-78. Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation . Cambridge University Press. Lee, E., Kim, H., & Reeves, T. C. (2023). Student motivation and affect in technology-rich language writing. Computer Assisted Language Learning , 36(2), 131-149. Lee, J., & Wang, K. (2024). Student agency in AI-integrated EFL writing: A longitudinal study. Computers and Composition , 72, 102760. Li, M., & Zhu, W. (2013). Patterns of computer-mediated interaction in small writing groups using wikis. Computer Assisted Language Learning , 26(1), 61-82. Lin, M., & Lee, J. (2022). Metacognition in AI-supported writing tasks: A review. TESOL Quarterly , 56(3), 828-845. Lingard, L. (2023). Reconfiguring agency in the age of artificial intelligence: Dialogic possibilities for language classrooms. TESOL Quarterly , 57(3), 901–924. Lo, C. K., Chan, W. H., Ng, T. K., & Wong, K. Y. (2024). Generative AI in education: Balancing personalized learning and learner autonomy erosion. Computers & Education , 215, 105025. Luria, A. R. (1976). Cognitive development: Its cultural and social foundations . Harvard University Press. Ma, Y., & Chen, M (2024). AI-empowered applications effects on EFL learners’ engagement in the classroom and academic procrastination. BMC Psychol 12, 739. Molenaar, I. (2022). Scaffolding agency in AI-enhanced learning environments: A dynamic systems approach . Paper presented at the International Conference on Artificial Intelligence in Education, Durham, United Kingdom. Niekerk, J. F., Schmidt, T. L., van der Merwe, A. R., & Chen, X. (2025). Generative AI and learner agency in digital learning ecologies: A sociotechnical perspective. Educational Technology Research and Development, Advance online publication. Priestley, M., Biesta, G., & Robinson, S. (2015). Teacher agency: An ecological approach . In M. Priestley & G. Biesta (Eds.), Reinventing the curriculum: New trends in curriculum policy and practice (pp. 1–20). Bloomsbury Academic. Prior, P. (2006). A sociocultural theory of writing . In C. A. MacArthur, S. Graham, & J. Fitzgerald (Eds.), Handbook of writing research (pp. 54-66). The Guilford Press. Rainio, A. P., & Hilppö, J. (2017). The dialectics of agency in educational ethnography. Ethnography and Education , 12(1), 78-94. Reich, J., Ito, M., & Richards, J. (2023). Task design for engagement in AI-mediated instruction. Computers & Education , 193, 104698. Roe, A. B., & Perkins, C. D. (2024). A gency in educational innovation: Negotiating identity and sociotechnical change . In T. J. Li & G. S. Park (Eds.), Advances in language education research (pp. 145–167). Springer. Rose, J., & Jones, M. (2005). The architecture of opportunity: Socio-technical systems and the distribution of agency in digital environments. MIT Press. Saghafian, H., Hayati, K., & Eskandari, Z. (2022). Learners’ emotions, motivation, and self-efficacy in AI-integrated language learning: A mixed-methods study. Language Teaching Research , 26(6), 944–962. Szabo, T. L., & Szoke, I. (2024). Theorizing machine agency in language learning environments: A posthumanist lens . Technology, Pedagogy and Education, Advance online publication. Sun, Y., Yin, Y., Wang, F., & Chen, N.S. (2023). The role of affective factors in AI-based language learning. Computers & Education: Artificial Intelligence , 4, 100175. Tao, J., & Gao, X. (2021). Language teacher agency in emergency remote teaching: A longitudinal narrative inquiry. System , 103, 102660. Trust, T., et al. (2023). Promoting reflective digital practices in AI writing contexts. British Journal of Educational Technology , 54(1), 42-62. Valsiner, J. (1997). Culture and the development of children’s action: A theory of human development (2nd ed.). John Wiley & Sons. van Lier, L. (2008). Agency in the classroom. In J. P. Lantolf & M. E. Poehner (Eds.), Sociocultural theory and the teaching of second languages (pp. 163-186). Equinox. Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (M. Cole, V. John-Steiner, S. Scribner, & E. Souberman, Eds.). Harvard University Press. Vygotsky, L. S. (1986). Thought and language (A. Kozulin, Trans.). MIT Press. Warschauer, M. (2005). Sociocultural perspectives on CALL . In J. Egbert & G. M. Petrie (Eds.), CALL research perspectives (pp. 41-51). Lawrence Erlbaum Associates. Wertsch, J. V. (1998). Mind as action . Oxford University Press. Yi, Y., & Angay-Crowder, T. (2016). Multimodal pedagogies for teacher education in TESOL. TESOL Quarterly , 50(4), 988-998. Yin, R. K. (2018). Case study research and applications: Design and methods (6th ed.). SAGE Publications. Zhang, Y., & Tur, G. (2024). Agency at risk? Generative AI and the paradox of self-directed learning. Australasian Journal of Educational Technology , 40(3), 1-17. Zheng, B., Warschauer, M., & Farkas, G. (2022). Engaging with AI: Agency and accountability in learning with emerging technologies. British Journal of Educational Technology , 53(4), 786-802. Zheng, D., & Newgarden, K. (2012). Rethinking language learning: Virtual worlds as a catalyst for change. International Journal of Learning and Media , 3(2), 13-36. Zou, D., Lin, J., & Sun, J. (2021). Metacognition and AI literacy in EFL writing. ReCALL , 33(1), 113-130. Additional Declarations No competing interests reported. Supplementary Files Supplementaryfile.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7972095","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":546188140,"identity":"ae1851fb-ef9c-4e37-aff2-224805a69806","order_by":0,"name":"Lei Zhang","email":"","orcid":"","institution":"Beijing University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Lei","middleName":"","lastName":"Zhang","suffix":""},{"id":546188141,"identity":"31eb390f-f69b-4125-b1d9-6a2747aaa3ba","order_by":1,"name":"Chunli Jiang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIiWNgGAWjYBACPmbmBoaEiv9ybPyNDQYfDGzkCGphY2ZsYHhwhtmYX+LwgcIZBWnGhLUwMDYwPmxjTpzZkJbwmefD4UTCWtgZ2yQS29gSNxw4Y7jZxoA5gYH98NENBBzWJpFwjsd4w+EeY+McA7Y8Bp60tBuEtZRJyAJtMQNq4SlmkOAxI0ILmwHjhgM55r8tDCQSG4jT0pagCPK+MYOBAVFami0SzhwAB7Jhj0GCMRshv/DzHz5480fFAUhU/vjzX46f/fAxvFqAgEUC1V4CykGA+QMRikbBKBgFo2AkAwAK3EqibZk1zwAAAABJRU5ErkJggg==","orcid":"","institution":"Beijing University of Technology","correspondingAuthor":true,"prefix":"","firstName":"Chunli","middleName":"","lastName":"Jiang","suffix":""}],"badges":[],"createdAt":"2025-10-28 17:26:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7972095/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7972095/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96160314,"identity":"60e86b8f-53df-4f09-ad17-cf71983af0cc","added_by":"auto","created_at":"2025-11-18 08:47:50","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":65275,"visible":true,"origin":"","legend":"","description":"","filename":"RevisedAnonymisedManuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/f220cb5d4efa99c965bea9d7.docx"},{"id":96251113,"identity":"256affc2-dee5-433f-a612-842c611ec231","added_by":"auto","created_at":"2025-11-19 07:39:20","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5264,"visible":true,"origin":"","legend":"","description":"","filename":"5921a0f7c08848749ba23a8ac4f3db19.json","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/26014e8a8b4bd4c72c1c760c.json"},{"id":96251043,"identity":"f1b5e5ec-7685-4ddb-b13d-6bcb63a74609","added_by":"auto","created_at":"2025-11-19 07:39:15","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16756,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfile.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/e779d821b923a7703b1c7c9c.docx"},{"id":96160315,"identity":"6585b0d0-8f5e-4eee-a248-e35db8d55e49","added_by":"auto","created_at":"2025-11-18 08:47:50","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14654,"visible":true,"origin":"","legend":"","description":"","filename":"CoverletterBMCPsychologyd1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/2fd75055418bb14abe726832.docx"},{"id":96252020,"identity":"c8007b61-4548-4173-be85-637da2081024","added_by":"auto","created_at":"2025-11-19 07:40:20","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":129064,"visible":true,"origin":"","legend":"","description":"","filename":"5921a0f7c08848749ba23a8ac4f3db191enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/cdbd31af5056701cc4695c1d.xml"},{"id":96160321,"identity":"21b593f1-2ba8-4c97-8042-9abf58a2e9ae","added_by":"auto","created_at":"2025-11-18 08:47:50","extension":"xml","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":125374,"visible":true,"origin":"","legend":"","description":"","filename":"5921a0f7c08848749ba23a8ac4f3db191structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/54741e66f9d23ef299b48914.xml"},{"id":96160319,"identity":"ae8f927b-6d89-427b-87c1-7cae3e1df6d7","added_by":"auto","created_at":"2025-11-18 08:47:50","extension":"html","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":135347,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/8f4b676c76df52b306cce5c8.html"},{"id":97895703,"identity":"d11441a4-8800-4b99-823f-32db8b6eeaf6","added_by":"auto","created_at":"2025-12-10 15:34:45","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1059763,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/0f10b682-9416-4405-b515-25cb1e79f26b.pdf"},{"id":96160317,"identity":"6d5b5029-d86d-467a-9e62-767306e74873","added_by":"auto","created_at":"2025-11-18 08:47:50","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":16756,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfile.docx","url":"https://assets-eu.researchsquare.com/files/rs-7972095/v1/fd7a524f3f1bf0e1f9352b20.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"‘AI-Learner Partnership:’: Psychological Mechanisms and Developmental Trajectories of EFL Learners’ Writing Agency in GenAI-Assisted Courses","fulltext":[{"header":"Highlights","content":"\u003cul\u003e\n \u003cli\u003eThis study explores how EFL learners construct and develop \u003cstrong\u003ewriting agency\u003c/strong\u003e when engaging with \u003cstrong\u003egenerative AI (GenAI) tools\u003c/strong\u003e in second language writing courses.\u003c/li\u003e\n \u003cli\u003eEmploying an \u003cstrong\u003eintegrated sociocultural and positive psychological framework\u003c/strong\u003e, the study conceptualizes GenAI as a \u003cstrong\u003ecultural mediational means\u003c/strong\u003e shaping learners\u0026rsquo; cognitive, affective, and agentic development.\u003c/li\u003e\n \u003cli\u003eLearners\u0026rsquo; agency followed a \u003cstrong\u003ethree-phase trajectory\u003c/strong\u003e\u0026mdash;exploration, adaptation, and appropriation\u0026mdash;reflecting a dynamic developmental process of techno-authorial identity formation.\u003c/li\u003e\n \u003cli\u003eFour distinct \u003cstrong\u003epatterns of agency manifestation\u003c/strong\u003e emerged: transitional, distributed, reflective, and transformative agency, each mediated by self-efficacy, resilience, and growth mindset.\u003c/li\u003e\n \u003cli\u003eThe findings advance theoretical understanding of \u003cstrong\u003eagency in AI-assisted writing\u003c/strong\u003e and offer pedagogical guidance for designing tasks that foster \u003cstrong\u003eagentic engagement, well-being, and autonomous learning\u003c/strong\u003e in GenAI-supported contexts.\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"1. Introduction","content":"\u003cp\u003eAgency\u0026mdash;the capacity to purposefully and meaningfully act within and upon one\u0026rsquo;s environment\u0026mdash;remains a cornerstone of contemporary language teaching and learning scholarship (Ahearn, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2001\u003c/span\u003e; Deters et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Duff, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Larsen-Freeman, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Tao \u0026amp; Gao, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In second language (L2) education, agency is understood as socioculturally mediated: learners\u0026rsquo; goal-directed actions are enabled, shaped, and at times constrained by the social, cultural, and increasingly technological environments they inhabit (Archer, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Giddens, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1984\u003c/span\u003e; Priestley et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). As agency is relational and emergent\u0026mdash;manifesting in negotiation with others, sociotechnical conditions, and the learner\u0026rsquo;s own evolving identity\u0026mdash;its conceptualization and enactment are continuously revisited amid educational innovation (Larsen-Freeman, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Roe \u0026amp; Perkins, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe rapid integration of generative artificial intelligence (GenAI) in educational contexts has brought the question of learner agency to the forefront in new and complex ways (Niekerk et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Lingard, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Molenaar, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). GenAI technologies, powered by advanced natural language processing and machine learning, offer learners unprecedented resources for language production and revision, from sophisticated grammar correction and style adaptation to the generation of coherent, context-sensitive academic prose ((Niekerk et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; lingard, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Molenaar, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Chapelle \u0026amp; Sauro, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). These affordances promise deeper personalization and self-directed learning\u0026mdash;but simultaneously raise concerns about overreliance, diminished independent thinking, and the erosion of learner autonomy (Lo et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhang \u0026amp; Tur, \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Indeed, the permeability between human and machine contributions in writing, what Warschauer (\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2005\u003c/span\u003e) called the emergence of new \u0026ldquo;literacy ecologies,\u0026rdquo; compels educators and researchers to reexamine foundational notions of authorship, voice, and intentionality in digitally mediated settings. Underlying this transformation is a fundamental shift in the relational landscape of agency. Within Vygotsky\u0026rsquo;s sociocultural theory, all higher-order learning is mediated through cultural tools (Lantolf \u0026amp; Thorne, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). GenAI, in this sense, functions as an advanced mediational means that not only supports but transforms the very nature of the academic writing process\u0026mdash;reshaping the division of labor (Engestr\u0026ouml;m, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) between the actions of the learner and the affordances or constraints imposed by the technology (Wertsch, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e). Unlike earlier digital tools, GenAI\u0026rsquo;s capacity to generate, critique, and even co-create text at a human-like level forces a reconsideration of where learner agency resides, how it is enacted, and how it evolves over time. It also raises the specter of \u0026ldquo;machine agency\u0026rdquo;\u0026mdash;the degree to which the technology itself shapes, redirects, or even substitutes for human intent and authorship (Dattathrani \u0026amp; De\u0026rsquo;, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Despite increasing theoretical attention to these dynamics (Szabo \u0026amp; Szoke, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Roe \u0026amp; Perkins, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), empirical studies remain limited, especially regarding how language learners negotiate agency in everyday, longitudinal engagement with GenAI systems.\u003c/p\u003e\u003cp\u003eMost existing literature has focused on initial user perceptions, functionality of AI tools, or teacher perspectives, leaving under-examined the temporal and processual aspects of agency construction as learners appropriate, resist, and adapt to GenAI in the context of complex academic tasks. Furthermore, little is known about the reciprocal interplay among human agency, technological affordances, sociocultural resources, and learners\u0026rsquo; cognitive and affective processes\u0026mdash;a gap recognized as pressing by educational policymakers and international organizations (UNESCO, 2023). To address this gap, this study adopts a sociocultural perspective to explore how EFL learners in China exercise, negotiate, and develop their writing agency over an extended period of GenAI-supported writing courses.\u003c/p\u003e"},{"header":"2. Literature Review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Writing Agency in Second Language Learning\u003c/h2\u003e\u003cp\u003eAgency has emerged as a key concept in second language acquisition, reflecting broader social turns in the field that recognize learners as active participants rather than passive recipients in educational processes (Duff, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; van Lier, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). Ahearn (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) defines agency as \u0026ldquo;the socioculturally mediated capacity to act\u0026rdquo; (p. 112), emphasizing that individual intention always operates within socially structured constraints and affordances. In language learning contexts, agency manifests through learners\u0026rsquo; strategic efforts to appropriate linguistic resources and participate meaningfully in various communities (Lantolf \u0026amp; Pavlenko, 2001; Lave \u0026amp; Wenger \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e1991\u003c/span\u003e; Rainio \u0026amp; Hilpp\u0026ouml; \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWriting agency specifically refers to the capacity of writers to make conscious choices, exercise control over the writing process, and express individual voice and identity through text (Tardy, 2016). For second language writers, this agency is complexly mediated by multiple factors, including linguistic proficiency, cultural-rhetorical traditions, institutional expectations, and available technological resources (Yi, 2013). Sociocultural approaches to writing agency emphasize that it emerges through participation in literacy practices, with writers gradually internalizing the values, strategies, and discourses associated with particular writing communities (Prior, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Research on second language writing agency has highlighted several key dimensions. Tardy (2016) examined how L2 writers develop discoursal voice through strategic control of generic conventions. Baker (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) investigated how writers navigate competing institutional demands while maintaining personal intentions. Yi \u0026amp; Angay-Crowder (\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) explored how digital technologies create new spaces for agency expression, particularly for multilingual writers who may feel constrained in traditional academic contexts. However, as Ding (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2008\u003c/span\u003e) notes, agency in L2 academic writing is often characterized by tension between conformity to established discourse norms and the writer\u0026rsquo;s desire for self-expression.\u003c/p\u003e\u003cp\u003eA significant gap in this literature concerns how emergent AI technologies\u0026mdash;with their capacity to generate human-like text\u0026mdash;transform the conditions for agency expression. Traditional views of writing agency presuppose human writers as primary agents who may employ various tools and strategies, but remain the central decision-makers in text production. As Haas (1996) argued in her examination of writing technologies, \u0026ldquo;writing is never simply a cognitive process, never simply a technological process, but is instead, always, a socio-technological-cognitive activity\u0026rdquo; (p. 27). GenAI technologies complicate this formulation by introducing a quasi-agentive element into the writing process itself.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Technology-Mediated Writing from the Lens of Sociocultural Theory\u003c/h2\u003e\u003cp\u003eSociocultural theory provides a theoretical framework for understanding how tools and technologies fundamentally transform the nature of human activity rather than simply enhancing existing capabilities (Vygotsky, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e1978\u003c/span\u003e; Wertsch, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e). From this perspective, technologies are not merely instruments but mediational means that reshape cognitive processes, creating what Luria (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e1976\u003c/span\u003e) termed \u0026ldquo;historically formed biases of mind\u0026rdquo; through their regular use. Applied to writing technologies, sociocultural approaches examine how digital tools create new conditions for literacy development through mediated activity. Early work by Haas (1996) demonstrated how word processing technologies transformed writers\u0026rsquo; conceptual and physical interactions with text. More recently, Zheng \u0026amp; Newgarden (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2012\u003c/span\u003e) explored how digital environments create new affordances for language learners to develop agency through meaningful participation in valued activities.\u003c/p\u003e\u003cp\u003eA central concept in sociocultural approaches to technology is the dialectical relationship between users and tools. As Wertsch (\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e) argues, \u0026ldquo;agent and mediational means are best thought of as existing in a dialectical tension, in which neither can be analyzed or understood in isolation from the other\u0026rdquo; (p. 34). This perspective suggests that when EFL learners engage with GenAI tools, both student and technology mutually transform each other through their interaction. Learners appropriate technological affordances in individualized ways, while simultaneously adapting their writing practices to accommodate the tool\u0026rsquo;s constraints and possibilities. Several studies have applied sociocultural perspectives to examine technology-mediated writing in second language contexts. Jin \u0026amp; Deifell (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) investigated how online translation tools function as mediational means in L2 writing, finding that learners developed strategic competence in tool use over time. Similarly, Li \u0026amp; Zhu (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) examined collaborative writing in wiki environments, demonstrating how digital affordances transformed patterns of peer interaction and joint knowledge construction.\u003c/p\u003e\u003cp\u003eThe emergence of GenAI systems represents a qualitatively different category of mediational means in writing, however. Unlike previous technologies that primarily augmented specific aspects of the writing process, GenAI tools can participate in multiple dimensions simultaneously\u0026mdash;generating ideas, structuring arguments, producing grammatical sentences, and even mimicking authorial voice. This multifunctionality creates what Engestr\u0026ouml;m (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) might describe as contradictions within the activity system of academic writing, as traditional divisions of labor and community norms are disrupted by the introduction of a new technological actor with quasi-agentive capabilities.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Re-conceptualizing agency in the age of GenAI: GenAI-Supported Writing\u003c/h2\u003e\u003cp\u003eThe advent of generative artificial intelligence (GenAI) has catalyzed an essential reconsideration of agency in educational contexts, particularly in domains of academic writing where both human authorship and technological mediation converge. Recent theoretical advances increasingly recognize the need for an ecological and distributed perspective on agency\u0026mdash;one that acknowledges complex interplays between human intention and technological participation in learning processes (Archer, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Dattathrani \u0026amp; De\u0026rsquo;, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e;Kaptelinin \u0026amp; Nardi, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Ma \u0026amp; Chen, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Theoretical frameworks from Giddens (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1984\u003c/span\u003e) and Archer (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) foreground the dynamic tension between structure and agency, wherein individuals\u0026rsquo; capacities for meaningful action are simultaneously enabled and constrained by social structures\u0026mdash;structures which are themselves recursively transformed through agentic practice. In digital environments, this dialectic becomes further nuanced as technologies such as GenAI function not merely as passive tools but as active participants in the construction and distribution of agency, giving rise to emergent socio-technical opportunity structures (Dattathrani \u0026amp; De\u0026rsquo;, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Lee et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Rose \u0026amp; Jones, \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2005\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis ecological conceptualization acknowledges agency as inherently relational and emergent\u0026mdash;not a fixed property possessed by individuals, but a dynamic achievement intimately entangled with the affordances and constraints of technological and sociocultural environments (Emirbayer \u0026amp; Mische, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Priestley et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Moreover, agency in technological contexts is mediated by cognitive and affective dimensions\u0026mdash;self-efficacy beliefs, motivational orientations, and emotional responses (Bandura, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2001\u003c/span\u003e)\u0026mdash;as well as by temporal-relational dynamics through which learners navigate between past experiences, present possibilities, and anticipated futures (Emirbayer \u0026amp; Mische, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e1998\u003c/span\u003e). Thus, agency in GenAI-mediated language learning emerges as multifaceted and complex, constituted through intricate negotiations among human capacities, technological affordances, and sociocultural resources.\u003c/p\u003e\u003cp\u003eThe empirical examination of human-AI collaboration in writing contexts has begun to illuminate these dynamics. Cowan et al. (2021) demonstrated that AI language models, positioned as creative collaborators, can simultaneously inspire novel compositional pathways while inadvertently constraining creative exploration, contingent upon writers\u0026rsquo; appropriation strategies. Clark et al. (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) introduced the concept of \u0026ldquo;creative co-authorship\u0026rdquo; between humans and AI, emphasizing writers\u0026rsquo; development of new metacognitive dimensions as they negotiate which AI-generated content to adopt, modify, or reject\u0026mdash;a process that resonates with sociocultural notions of appropriation (Wertsch, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e). In language education specifically, Feng et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) revealed how second language writers strategically deploy AI writing assistants while actively maintaining textual ownership, suggesting complex negotiations of control and assistance.\u003c/p\u003e\u003cp\u003eDespite these important contributions, significant theoretical and empirical gaps persist in our understanding of agency in GenAI-supported academic writing. First, extant research has predominantly employed short-term experimental designs that capture initial interactions but fail to trace developmental trajectories as learners\u0026rsquo; relationships with AI tools evolve over sustained engagement. As Warschauer (\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2005\u003c/span\u003e) contends, the transformative impact of new technologies on literacy development and agency can only be fully comprehended through longitudinal investigations that examine how users appropriate technological affordances within authentic contexts of meaningful practice. Second, studies have often conceptualized agency primarily through individualistic lenses focused on personal autonomy and control, neglecting the distributed, relational character of agency emerging through dynamic interactions among writers, AI systems, institutional contexts, and discourse communities (Wertsch et al., 1993). Third, research has typically examined either cognitive or sociocultural dimensions in isolation, without adequate attention to how these dimensions interact within integrated theoretical frameworks. Thus, the study addresses these limitations by conceptualizing GenAI writing tools as sophisticated cultural mediational means that transform the conditions of possibility for agency in academic writing. Through a longitudinal sociocultural lens, the study investigate how EFL university learners develop, negotiate, and redistribute agency in sustained engagement with GenAI-supported second language writing courses. Two research questions addressed:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eHow do EFL students\u0026rsquo; patterns of agency development and negotiation evolve over sustained engagement with GenAI in second language writing courses ?\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eIn what ways does the interplay of cognitive processes, affective dimensions, and contextual factors shape the construction and distribution of agency in GenAI-Supported Writing ?\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Research methods","content":"\u003cp\u003eThis study employed a qualitative case study approach to investigate EFL learners\u0026rsquo; writing agency construction in AI-supported academic writing. Case study methodology was selected for its capacity to examine complex phenomena within their authentic contexts (Yin, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), allowing for in-depth exploration of the \u0026ldquo;how\u0026rdquo; and \u0026ldquo;why\u0026rdquo; questions regarding learner agency in technology-mediated writing environments. The research design was informed by sociocultural principles, particularly attention to mediated action, historical development, and the interconnection between individual cognition and social practice (Lantolf \u0026amp; Thorne, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Research Context and Participants\u003c/h2\u003e\u003cp\u003eThe study was conducted in an undergraduate English writing course at a university in China where students engaged in a series of GenAI-supported second language writing courses over the course of a semester. Participants included 11 EFL undergraduate students (7 males, 4 females). Participants were academically motivated students with intermediate to advanced English proficiency levels. All students had completed prior English writing courses but reported this course as their first experience using AI tools for academic writing purposes. The course instructor integrated GenAI tools (primarily Doubao) as optional resources for students to use throughout the writing process, from initial planning through drafting and revision.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Data Collection\u003c/h2\u003e\u003cp\u003eThe primary data source for this study consisted of semi-structured interviews conducted with each participant at the conclusion of the semester-long AI-assisted writing course. The semi-structured interview protocol was designed specifically for this study. The full English version of the interview questions has been provided as a Supplementary File (Appendix A). Each interview lasted 30 to 45 minutes, was conducted in the participants\u0026rsquo; native language (Mandarin Chinese), and yielded a total of 59,613 words of transcribed data. The interview protocol was designed to probe four key dimensions informed by the research questions:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eTemporal development of AI tool use and agency perceptions over the semester\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eCognitive and affective experiences when engaging with AI tools\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eResponses to different AI-supported task designs\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003ePerceptions of human-AI collaboration and shared authorship\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eIn addition to interview data, supplementary sources were collected, including participants\u0026rsquo; self-reported summaries of their emotional responses to AI-assisted writing activities and coded analyses of self-regulated writing strategies. The analysis of writing strategies categorized participants\u0026rsquo; approaches into four dimensions: cognitive strategies (e.g., text processing, content retention), metacognitive strategies (e.g., idea planning, goal-setting, self-monitoring and evaluation), social behaviors (e.g., peer learning, feedback processing), and motivational regulation (e.g., interest enhancement, motivational self-talk, emotional control). This triangulated data collection approach provided a comprehensive understanding of learners\u0026rsquo; agentic engagement and adaptive strategies in AI-supported academic writing contexts.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.4 Data Analysis\u003c/h2\u003e\u003cp\u003eInterview data were analyzed using thematic analysis informed by sociocultural theory principles. Following Braun \u0026amp; Clarke\u0026rsquo;s (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2006\u003c/span\u003e) six-phase approach, the process included: (1) familiarization with the data through repeated reading, (2) generation of initial codes, (3) searching for themes, (4) reviewing themes, (5) defining and naming themes, and (6) producing the report. Throughout this process, sociocultural concepts such as mediation, appropriation, internalization, and activity systems served as analytical lenses. The coding procedure followed a three-stage process\u0026mdash;open coding, axial coding, and selective coding\u0026mdash;to systematically develop themes from the raw data, as illustrated in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eThree-Stage Coding Process with Examples from Interview Data\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOpen Codes (with Interview Excerpts)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAxial Codes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSelective Codes (Themes)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMost of the time, I directly let AI revise for me, then I look at why it made those revisions (S1-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003e\u003cb\u003eInitial dependence\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003eDevelopmental Trajectories of Agency\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWhen I first used AI, I didn\u0026rsquo;t know how to write prompts, fortunately the teacher provided a guide document (S10-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eProgression from initial exploration\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLack of technical proficiency, users' unfamiliarity with AI use (S5-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ethrough strategic adaptation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInitially unfamiliar with AI use, at first the low scores made me think my writing level wasn't very high (S7-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eto deliberate appropriation\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo, I select a few feedback suggestions to implement. Because I feel I should preserve some of my own elements (S10-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003e\u003cb\u003eCritical evaluation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003eCognitive Mediations of Agency\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIf I feel the sentence it provides deviates from what I want to express, I might not adopt it (S10-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMetacognitive awareness\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCritically accepting AI feedback (accepting all grammar suggestions, screening vocabulary) (S6-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003estrategic knowledge acquisition,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInsisting on independent conception, maintaining reservations about AI outlines, believing one\u0026rsquo;s own thought process is clearer (S6-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ecritical evaluation,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eWith AI I feel more relaxed because AI can help in many ways (S2-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003e\u003cb\u003eAffective responses\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\u003cb\u003eAffective Mediations of Agency\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eThe most obvious emotion would be a sense of achievement; there\u0026rsquo;s a notable feeling of accomplishment after improving an essay (S2\u0026ndash;Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eEmotional responses to feedback,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNegative emotions were fewer, mainly manifested as confusion and puzzlement over AI\u0026rsquo;s inaccurate or overly broad suggestions (S4-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003echanging self-efficacy beliefs,\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eThe use of AI enhanced the interviewee\u0026rsquo;s self-confidence and sense of achievement, changed their negative attitude toward writing (S3-Interview-2503)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eevolving motivational orientations\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThis systematic coding approach enabled us to trace development patterns across participants while preserving individual variation in agency construction. The open codes, derived directly from interview excerpts, captured participants\u0026rsquo; experiences in their own words. These were then grouped into conceptually related axial codes representing broader patterns of engagement with AI tools. Finally, selective codes integrated these patterns into comprehensive themes addressing our research questions about agency development in AI-assisted writing contexts. For agency trajectory analysis, individual case narratives were constructed for each participant, mapping their relationship with AI tools over time. Cognitive and affective mediators were identified through axial codes representing metacognitive processes, strategic behaviors, and emotional responses. Task design impacts were analyzed by comparing coded segments related to different writing assignments and AI interaction structures. Human-AI relationship dynamics were examined through selective codes reflecting perception, boundaries, and identity positions within technology-mediated writing practices.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Findings","content":"\u003cp\u003eThe analysis reveals how EFL learners constructed, negotiated, and transformed their writing agency through engagement with AI tools in second language writing courses.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Developmental trajectories of writing agency\u003c/h2\u003e\u003cp\u003eParticipants demonstrated distinct yet overlapping developmental trajectories in their relationships with AI tools and associated expressions of agency. Through the lens of sociocultural theory, these trajectories revealed the progressive internalization of AI-mediated writing practices and the gradual transformation from other-regulation to self-regulation (Lantolf \u0026amp; Thorne, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). In early encounters with AI tools, most participants exhibited limited agency, characterized by tentative exploration and high reliance on AI-generated content. This initial stage revealed significant variation in attitudes, with some approaching AI with curiosity while others expressed skepticism or technical discomfort. As one paritipant noted, \u0026ldquo;\u003cem\u003eWhen I first used AI, I didn\u0026rsquo;t know how to write prompts, fortunately the teacher provided a guide document\u003c/em\u003e\u0026rdquo; (S10-Interview-2503). During this phase, participants typically deferred to AI authority, accepting suggestions with minimal critical evaluation\u0026mdash;what Vygotsky (\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e1978\u003c/span\u003e) would characterize as other-regulation, with AI functioning as the more capable other directing the writing process.\u003c/p\u003e\u003cp\u003eAs participants gained experience, they developed more strategic approaches characterized by selective adoption of AI suggestions and increased metacognitive awareness. This transition typically emerged after 2\u0026ndash;3 writing tasks, when participants began critically evaluating AI contributions against their own intentions: \u0026ldquo;\u003cem\u003eI select a few feedback opinions to implement. Because I feel I should preserve some of my own elements, if I think mine is better, I keep it\u003c/em\u003e\u0026rdquo; ((S10-Interview-2503)). This stage revealed the development of what Wertsch (\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e) terms mastery\u0026mdash;knowing how to use cultural tools effectively\u0026mdash;as distinct from appropriation, which involves making tools one\u0026rsquo;s own. Participants developed evaluation criteria, typically accepting grammar corrections while being more selective about vocabulary and content recommendations.\u003c/p\u003e\u003cp\u003eIn later stages, many participants (particularly those reporting medium to high enjoyment) demonstrated sophisticated agency through deliberate appropriation of AI affordances for personalized goals. This stage featured intentional prompting strategies, metacognitive reflection, and integration of AI tools into individualized writing processes. As S7 described:\u003c/p\u003e\u003cp\u003e\u003cem\u003eAfter receiving feedback, I first look at which dimension I scored lowest in, and focus on modifying areas where I lost the most points, referring to AI feedback suggestions during revision. Then I send the modified content back to AI to see if the score has improved, and if not, I further question the AI, asking more specifically how I should revise this part.\u003c/em\u003e\u003c/p\u003e\u003cp\u003eThis multi-stage process reveals sophisticated agency through strategic tool use and proactive dialogue with AI. This developing expertise represents what Kaptelinin (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2005\u003c/span\u003e) characterizes as the appropriation of technological affordances to expand one\u0026rsquo;s action capabilities. Individual trajectories varied significantly based on participants\u0026rsquo; prior attitudes toward writing, technological self-efficacy, and perceived writing goals. High-enjoyment participants typically moved more quickly from exploration to appropriation, while others remained longer in the exploratory phase. Some exhibited non-linear trajectories, moving between stages based on task contexts or personal factors. This variation aligns with Rogoff\u0026rsquo;s (2003) view of development as transformation of participation in culturally valued activities, with individual pathways shaped by personal histories and contextual factors.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Cognitive and affective mediations of writing agency construction\u003c/h2\u003e\u003cp\u003eParticipants\u0026rsquo; writing agency construction was mediated by interconnected cognitive and affective processes that shaped their engagement with AI tools. These mediations reflect Vygotsky\u0026rsquo;s (\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e1986\u003c/span\u003e) concept of perezhivanie\u0026mdash;the unity of intellectual and emotional aspects in development\u0026mdash;revealing how agency emerges through the dynamic interplay of thinking, feeling, and action in technology-mediated environments. The cognitive dimension of agency construction manifested through several interrelated processes that evolved as participants gained experience with AI writing tools. Metacognitive awareness emerged as a foundational element, with participants developing increasingly sophisticated understanding of their writing processes and how AI could support them. As S7 reflected:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;I think about what issues AI feedback identified in my previous essays, and in my next writing, I try to avoid making those mistakes. I also consciously set writing goals for myself, such as avoiding repetitive words as much as possible in this essay, or incorporating some of the advanced sentence patterns I\u0026rsquo;ve accumulated.\u0026rdquo;\u003c/em\u003e (S7-interview-2503)\u003c/p\u003e\u003cp\u003eThis illustrates how AI feedback prompted metacognitive regulation through goal-setting and strategy development, creating what Flavell (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e1979\u003c/span\u003e) terms metacognitive experiences that shape future task approaches. As participants engaged with AI, they selectively internalized linguistic resources from these interactions, representing the Vygotskian transition from interpersonal to intrapersonal functioning. Many described deliberately studying AI-suggested vocabulary:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;Yes, I will try to remember those good expressions ChatGPT gives me.\u0026rdquo;\u003c/em\u003e (S10-interview-2503)\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;I try to remember some of the language expressions AI provides, but I haven\u0026rsquo;t specifically organized or memorized the advanced expressions it offers.\u0026rdquo;\u003c/em\u003e (S7-interview-2503)\u003c/p\u003e\u003cp\u003e This strategic knowledge acquisition transformed external AI suggestions into internal linguistic resources that participants could deploy independently. The development of critical evaluation capacities further distinguished more agentic participants, who articulated clear criteria for assessing AI suggestions rather than accepting them uncritically:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;Insisting on independent thinking, maintaining reservations about AI outlines, believing one\u0026rsquo;s own thought process is clearer.\u0026rdquo;\u003c/em\u003e (S6-interview-2503)\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;For example, sometimes it gives me suggestions on how to write a certain sentence. But if I feel that the sentence it provides deviates from what I want to express, I might not adopt it.\u0026rdquo;\u003c/em\u003e (S10-interview-2503)\u003c/p\u003e\u003cp\u003eThis critical stance reflects Wertsch\u0026rsquo;s (\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e) distinction between mastery and appropriation, with participants maintaining distance from aspects that conflicted with their intentions. The most sophisticated cognitive mediation involved personalized tool appropriation, where participants customized AI interactions to serve individual goals. S8 described learning to craft specific prompts to elicit targeted feedback, while S7 developed multi-stage processes alternating between self-revision and AI consultation. These approaches transformed generic AI affordances into individualized resources for writing development, reflecting what Kaptelinin (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2005\u003c/span\u003e) characterizes as the appropriation of technological affordances to expand one\u0026rsquo;s action capabilities.\u003c/p\u003e\u003cp\u003eAffective processes were equally crucial in mediating agency construction, intertwining with cognitive dimensions to shape engagement patterns. Emotional responses to AI feedback varied substantially but followed discernible patterns. Positive emotions\u0026mdash;particularly satisfaction and reduced anxiety\u0026mdash;typically accompanied experiences of increased efficiency and mastery:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;Yes, with AI I feel more relaxed because AI can help in many ways, making the writing process less difficult. The most obvious emotion would be a sense of achievement; there\u0026rsquo;s a notable feeling of accomplishment after improving an essay.\u0026rdquo;\u003c/em\u003e (S2-interview-2503)\u003c/p\u003e\u003cp\u003e\u0026ldquo;\u003cem\u003ePositive emotions dominate. Although initially I felt a bit anxious because low scores made me think my writing level wasn\u0026rsquo;t very high, with more use I discovered that AI can both point out my errors and help improve my writing level, while also saving considerable time. So compared to these advantages of AI, that small anxiety is negligible, and positive emotions are predominant.\u0026rdquo;\u003c/em\u003e(S7-interview-2503)\u003c/p\u003e\u003cp\u003eThese positive emotional experiences created reinforcing cycles of engagement with AI tools. Negative emotions typically emerged when participants encountered discrepancies between AI feedback and their intentions, or when they perceived AI limitations:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;Negative emotions were fewer, mainly manifested as confusion and puzzlement over AI\u0026rsquo;s inaccurate or overly broad suggestions, with anxiety significantly reduced due to AI\u0026rsquo;s involvement.\u0026rdquo;\u003c/em\u003e (S4-interview-2503)\u003c/p\u003e\u003cp\u003eSelf-efficacy beliefs shifted over time, with many describing initial doubts followed by increased confidence as they learned to work effectively with the technology:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;The use of AI enhanced the interviewee\u0026rsquo;s self-confidence and sense of achievement, changed their negative attitude toward writing, and made them more willing to engage in writing.\u0026rdquo;\u003c/em\u003e (S3-interview-2503)\u003c/p\u003e\u003cp\u003eThis trajectory aligns with sociocultural perspectives on self-efficacy as developing through successful tool-mediated activity rather than existing as a stable internal trait (Lantolf \u0026amp; Thorne, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Motivational orientations evolved as participants developed more sophisticated relationships with AI tools. Several reported transitioning from extrinsic motivations (completing assignments efficiently) to more intrinsic orientations focused on writing improvement and self-expression:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;My goal is to express ideas in very clear, precise language. I think in terms of expressing ideas with clear and accurate language, I have made progress.\u0026rdquo;\u003c/em\u003e (S10-interview-2503)\u003c/p\u003e\u003cp\u003eThis shift toward intrinsic motivation aligned with increased perceptions of agency, as participants moved from viewing AI as primarily reducing workload to seeing it as supporting their personal development as writers. The findings suggest that cognitive and affective processes operated in dialectical relationship, mutually reinforcing agency development. Participants who developed sophisticated metacognitive awareness typically also reported positive emotional experiences with AI tools, creating virtuous cycles of engagement. Conversely, those expressing persistent concerns about dependency often maintained more limited cognitive engagement with AI affordances, using tools primarily for surface-level editing rather than deeper writing development. This integrated cognitive-affective dimension underscores Vygotsky\u0026rsquo;s holistic view of development, emphasizing the inseparability of thought and emotion in the construction of agency within technology-mediated learning environments.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Task design influences on agency manifestation\u003c/h2\u003e\u003cp\u003eTask design significantly influenced how participants exercised and developed agency in AI-supported writing. This influence reflects activity theoretical perspectives on how structural elements of learning environments constrain and enable particular forms of participation (Engestr\u0026ouml;m, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). Through systematic analysis of participant experiences, three key aspects of task design emerged as particularly consequential for agency development: the degree of structure in AI-assisted activities, the positioning of AI within the writing process, and the nature of evaluation mechanisms. The degree of structure in tasks created varied affordances for agency expression across the learning trajectory. Highly structured activities with specific AI interaction protocols provided necessary scaffolding for initial agency development but sometimes limited more advanced expressions of agency. As one participant explained:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;I gradually used AI software functions under the teacher\u0026rsquo;s guidance, such as grammar checking, sentence diversification, and so on.\"\u003c/em\u003e (S4-interview-2503)\u003c/p\u003e\u003cp\u003eThis structured approach helped novice users develop fundamental competencies with AI tools through guided practice. In contrast, less structured tasks that invited exploratory AI use revealed greater variation in agency manifestations. Some participants flourished with this autonomy, developing sophisticated and personalized interaction strategies:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;During writing class, I first ask it to write an outline, then write a complete essay based on the outline, and afterward have it polish my work, checking if there\u0026rsquo;s anything that needs revision.\u0026rdquo;\u003c/em\u003e (S10-interview-2503)\u003c/p\u003e\u003cp\u003eHowever, not all students thrived with minimal structure. Those with less technological self-efficacy sometimes struggled with unstructured assignments:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;Lack of technical proficiency, users\u0026rsquo; unfamiliarity with AI use, leads to confusion and difficulties during the usage process.\u0026rdquo;\u003c/em\u003e (S5-interview-2503)\u003c/p\u003e\u003cp\u003eThese differential responses suggest that optimal task structuring for agency development may follow what Collins et al. (1989) described as a scaffold-and-fade approach, with initial structure gradually giving way to greater autonomy as students develop expertise in AI interaction. The positioning of AI within the writing process\u0026mdash;whether as planning aid, drafting assistant, or revision tool\u0026mdash;significantly affected how participants manifested agency in their writing practices. Students generally reported stronger expressions of agency when tasks positioned AI as a planning resource or revision aid rather than as a primary text generator. This preserved students\u0026rsquo; sense of authorial control while leveraging AI\u0026rsquo;s analytical capabilities:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;I first think about the logical flow of the essay myself, then send the essay topic requirements to AI to see what logical structure it generates, supplementing my existing logical framework with AI\u0026rsquo;s suggestions. The logical structure provided by AI is just auxiliary.\u0026rdquo;\u003c/em\u003e (S7-interview-2503)\u003c/p\u003e\u003cp\u003eThis approach reflects what Wertsch (\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1998\u003c/span\u003e) termed \u0026ldquo;authoritative stance,\" with participants maintaining primary authorship while using AI as a resource. In contrast, participants who began by generating full AI drafts often reported weaker agency expressions initially:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;AI will give me some materials and such that I can directly use.\"\u003c/em\u003e (S2-interview-2503)\u003c/p\u003e\u003cp\u003eThis reliance on AI-generated content positioned students more as editors than authors. However, many participants developed increasingly agentic approaches over time even when using AI for full draft generation, learning to critically evaluate and substantially revise AI content as their expertise grew. Task designs also varied in their evaluation mechanisms, with significant consequences for agency development and metacognitive awareness. AI scoring and detailed feedback provided valuable information for self-regulation but sometimes created dependency when positioned as the authoritative evaluation source:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;I don\u0026rsquo;t think AI scoring is very good, though I might be misjudging. But my intuition is that the essay, honestly, was one that AI helped me revise, and then it gave it a low score, though I thought the essay was actually quite good. But I don\u0026rsquo;t really understand AI.\u0026rdquo;\u003c/em\u003e (S1-interview-2503)\u003c/p\u003e\u003cp\u003eThis quote reveals the confusion that can arise when AI simultaneously serves as both writing assistant and evaluator, creating contradictions that undermine students\u0026rsquo; developing sense of agency. In contrast, tasks that combined AI feedback with teacher assessment or peer discussion typically supported stronger agency development by creating opportunities for critical evaluation of AI contributions:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;Feedback handling: Critically accepting AI feedback (accepting all grammar suggestions, screening vocabulary); addressing AI deficiencies through teacher communication.\u0026rdquo;\u003c/em\u003e (S6-interview-2503)\u003c/p\u003e\u003cp\u003eThis multi-source feedback approach encouraged students to develop their own evaluative criteria rather than deferring entirely to AI assessment. Tasks that explicitly prompted reflection on the human-AI writing process\u0026mdash;rather than focusing exclusively on the final product\u0026mdash;further supported metacognitive awareness and agency development:\u003c/p\u003e\u003cp\u003e\u003cem\u003e\u0026ldquo;I consciously set writing goals for myself, such as avoiding repetitive words as much as possible in this essay, or incorporating some advanced sentence patterns I\u0026rsquo;ve accumulated.\u0026rdquo;\u003c/em\u003e (S7-interview-2503)\u003c/p\u003e\u003cp\u003eThe findings on task design align with Deci \u0026amp; Ryan\u0026rsquo;s (2000) self-determination theory, suggesting that tasks supporting autonomy, competence, and relatedness foster more agentic engagement with AI writing tools. The optimal approach appears to be tasks that provide sufficient structure to support competence while allowing space for autonomy and encouraging reflective awareness of the human-AI relationship. Such designs can scaffold the transition from other-regulation to self-regulation in technology-mediated writing environments, supporting what Vygotsky termed the zone of proximal development\u0026mdash;the gap between what learners can accomplish independently and what they can achieve with appropriate support. This socio-culturally informed approach to task design recognizes that agency is not simply an individual attribute but emerges through the dynamic interaction between learners and the structured environments in which they operate, including the technological tools that mediate their writing activities.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e4.4 Negotiated agency in Human-AI Partnerships\u003c/h2\u003e\u003cp\u003eThe study found that participants\u0026rsquo; views and practices towards AI were both complex and dynamic, directly shaping their agency manifestations. Most commonly, learners perceived AI as a powerful tool that could efficiently address much of their writing needs:\u003c/p\u003e\u003cp\u003e\u0026ldquo;\u003cem\u003eUsing AI definitely has more benefits than drawbacks; it can solve at least 60\u0026ndash;70% of the problems at hand, and I can find solutions for the remaining issues myself\u0026rdquo;\u003c/em\u003e (S1-interview-2503)\u0026mdash;and this instrumental approach often corresponded with a clear sense of authorial control and selective adoption of AI suggestions.\u003c/p\u003e\u003cp\u003eSome, particularly those who engaged more deeply with AI, developed a more collaborative stance, describing dialogical processes where they negotiated meaning and refined ideas in partnership with AI: \u0026ldquo;\u003cem\u003eAI\u0026rsquo;s feedback speed...I find the feedback very detailed\u003c/em\u003e\u0026rdquo;(S10-interview-2503). Others, especially in early stages, deferred to AI as an authority, accepting revisions with little critique: \u0026ldquo;\u003cem\u003eMost of the time, I directly let AI revise for me...\u003c/em\u003e\u0026rdquo; (S1-interview-2503). Notably, many participants\u0026rsquo; perspectives evolved over time, blending these roles as they became more familiar with AI\u0026rsquo;s strengths and limits.\u003c/p\u003e\u003cp\u003eNegotiation of contribution boundaries emerged as a central theme. Learners increasingly distinguished between aspects of writing to keep personal and those to delegate to AI, such as reserving content and logic for themselves but using AI for language enhancement and revision. As one participant noted, \u0026ldquo;\u003cem\u003eIndependently conceptualizing logical thought patterns with AI as auxiliary support, monitoring writing progress through feedback and goal setting\u0026rdquo;\u003c/em\u003e (S7-interview-2503), while another balanced independent ideation with AI-provided scoring and revision support (S8). This selective, often task-dependent appropriation reflected growing evaluative criteria and confidence\u0026mdash; \u0026ldquo;\u003cem\u003eNo, I select a few feedback suggestions to implement. If I think mine is better, I keep it and don\u0026rsquo;t change it\u003c/em\u003e\u0026rdquo; (S10-interview-2503).\u003c/p\u003e\u003cp\u003eIdentity positioning was also fluid. Some learners insisted on maintaining independence and clear boundaries\u0026mdash; \u0026ldquo;\u003cem\u003eInsisting on independent conception, maintaining reservations about AI outli\u003c/em\u003enes\u0026rdquo; (S6-interview-2503)\u0026mdash;while others viewed themselves as part of a productive partnership, e.g., \u0026ldquo;\u003cem\u003eAI can help me correct grammar errors and polish my completed essays...\u003c/em\u003e\u0026rdquo; (S7-interview-2503). Feelings of dependency, particularly early on, were voiced: \u0026ldquo;\u003cem\u003eFeeling highly dependent...may create user dependency, affecting independent thinking ability\u003c/em\u003e\u0026rdquo; (S5-interview-2503). However, most evolved toward more autonomous or collaborative identities as experience grew.\u003c/p\u003e\u003cp\u003eOverall, agency in human-AI writing partnerships was the outcome of ongoing negotiation\u0026mdash;participants continuously redefined the role of AI, the boundaries of its contribution, and their own identity as writers. As participants\u0026rsquo; experience with AI deepened, their agency shifted from passive acceptance toward strategic, reflective, and evaluative engagement, revealing agency as an emergent, contextual, and relational process shaped by both personal intention and the affordances of AI technology.\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Discussion","content":"\u003cp\u003eThe findings illuminate how EFL learnersconstruct, negotiate, and develop writing agency when engaging with GenAI tools in academic writing contexts. In this section, we interpret these findings through sociocultural theoretical lenses, discussing their implications for understanding learning dynamics in AI-mediated writing environments.\u003c/p\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003e5.1 The evolution of writing agency in prolonged GenAI engagement (Addressing RQ1)\u003c/h2\u003e\u003cp\u003eOur findings demonstrate that EFL students\u0026rsquo; writing agency in GenAI-supported contexts is a dynamic and developmental phenomenon, not a static trait. Learners usually progressed through a three-phase trajectory: 1) initial exploration, marked by reliance and experimentation; 2) strategic adaptation, characterized by selective and thoughtful tool use; and 3) deliberate appropriation, where GenAI becomes an integrated, personalized cognitive tool. This pattern aligns with Vygotsky\u0026rsquo;s (\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e1978\u003c/span\u003e) principle that higher-order functions emerge through socially mediated activity and the gradual internalization of external tools (Lantolf \u0026amp; Thorne, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2006\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eHowever, the process is highly individualized. Students\u0026rsquo; movement across these phases was neither unidirectional nor uniform, being shaped by factors such as prior writing experience, technological self-efficacy, and institutional expectations. This observation reflects Valsiner\u0026rsquo; s (1997) \u0026ldquo;bounded indeterminacy\u0026rdquo; and echoes recent findings by Lee \u0026amp; Wang (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) and Aydin \u0026amp; Aky\u0026uuml;z (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), who noted that agency development in AI-rich environments is marked by personalized, recursive shifts rather than linear progressions. The data support the argument by Chen et al. (2024) that flexible, adaptive pedagogies are necessary to cater to such diversified development. Moreover, the evolution of agency was not merely the result of tool exposure but also continual negotiation and praxis within authentic academic activities. This challenges the binary view of agency as either present or absent and instead reveals it as a situated, evolving achievement that occurs as learners appropriate GenAI\u0026rsquo;s mediational means into their own writing practices (Sun et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Trust et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e5.2 The interplay of cognitive, affective, and contextual factors (Addressing RQ2)\u003c/h2\u003e\u003cp\u003eThe construction and distribution of agency in EFL students\u0026rsquo; AI-supported writing developed through the intertwined influences of cognitive, affective, and contextual factors. Cognitively, learners who exhibited strong metacognitive awareness (such as evaluating GenAI feedback, transferring learning across contexts, and strategic task planning) were able to exert greater agency, moving from passive recipients of AI output to critical co-authors. Recent studies (Lin \u0026amp; Lee, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zou et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) confirm that metacognition is a key predictor of effective and agentive AI tool usage in writing. Affective experiences like confidence, anxiety, and technological motivation proved equally salient. Positive emotional responses\u0026mdash;often linked to mastery moments or successful negotiation with GenAI\u0026mdash;facilitated reflective engagement, while ongoing anxiety or fear of over-reliance restricted agentive growth. These findings are consistent with Saghafian et al. (\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) and Kim \u0026amp; Reeves (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), who highlight the mediating effects of affect on technology adoption and creative autonomy in academic literacy contexts. The contextual and relational dimension was no less critical. The design of writing assignments, positioning of AI within the process, and the quality of scaffolding either expanded or limited students\u0026rsquo; opportunities for agentive action. Tasks that began with tightly structured AI use but faded support as learners gained expertise created productive \u0026ldquo;zones of proximal development\u0026rdquo; (Kohnke et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Reich et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Furthermore, when GenAI was positioned as a revision aid or dialogic collaborator rather than a text generator, learners demonstrated stronger sense of authorship and self-regulation\u0026mdash;reflecting Bakhtin\u0026rsquo;s (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1981\u003c/span\u003e) ideal of internally persuasive discourse and recent distributed cognition frameworks (de Bot \u0026amp; Zhang, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zheng et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn sum, agency in GenAI-Supported writing emerges through a dialectical movement: learners continuously navigate and negotiate personal goals, social roles, technological affordances, and educational constraints. This confirms the argument by Brynjolfsson et al. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) that distributed agency in AI-supported contexts is always contingent, relational, and open for pedagogical optimization.\u003c/p\u003e\u003c/div\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eThis study reveals that the development of writing agency among EFL learners in GenAI-assisted academic contexts is a dynamic, non-linear process, shaped by sustained engagement with technological tools, task design, and sociocultural relationships. Agency emerges not as a fixed trait but through evolving trajectories characterized by initial dependency, strategic adaptation, and, for some, the deliberate appropriation of AI affordances. Both cognitive and affective mediators\u0026mdash;such as metacognitive awareness, critical evaluation, and emotional responses\u0026mdash;interact dialectically, enabling learners to negotiate meaningful partnerships with AI while maintaining autonomy and personal authorship.\u003c/p\u003e\u003cp\u003eThese findings underscore the necessity of flexible pedagogical scaffolding and thoughtfully designed tasks that facilitate reflective, dialogic interaction with AI. Effective writing instruction in AI-assisted environments should recognize the relational and distributed nature of agency, allowing for diverse learner pathways and critical engagement with technology. As GenAI continues to evolve, educators are encouraged to foster technological literacy, encourage reflective tool use, and design adaptive supports that nurture empowered and adaptive agency in academic writing.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author(s) received no financial support for the research.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Information\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003ch3\u003eAuthors and Affiliations\u003c/h3\u003e\n\u003cp\u003eSchool of Foreign Languages, Beijing University of Technology, Beijing, People’s Republic of China,\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLei Zhang \u0026amp; Chunli Jiang\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eContributions\u003c/p\u003e\n\u003cp\u003eLei Zhang: conceptualization, investigation, and draft the original manuscript writing; Chunli Jiang: reviewing the manuscript. All authors reviewed the findings and approved the final version of the manuscript.\u003c/p\u003e\n\u003ch3\u003eCorresponding author\u003c/h3\u003e\n\u003cp\u003eCorrespondence to Chunlijiang\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Declarations\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003ch3\u003eEthics approval and consent to participate\u003c/h3\u003e\n\u003cp\u003eThis study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the University Ethics Committee for Research Involving Humans at Beijing University of Technology. All participants received detailed information about the study and voluntarily provided written informed consent prior to participation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConsent to participant\u003c/p\u003e\n\u003cp\u003eInformed consent was obtained from every participant provided.\u003c/p\u003e\n\u003cp\u003eHuman ethics and consent to participate declarations\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003eConsent to publication\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003eConflict interest\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe authors have no potential conflicts of interest to disclose.\u003c/p\u003e\n\u003cp\u003eCompeting interest\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAhearn, L. M. (2001). Language and agency. \u003cem\u003eAnnual Review of Anthropology\u003c/em\u003e, 30(1), 109-137.\u003c/li\u003e\n\u003cli\u003eArcher, M. S. (2010). \u003cem\u003eRoutine, reflexivity, and realism: Three lectures on structure and agency\u003c/em\u003e. Oxford University Press.\u003c/li\u003e\n\u003cli\u003eAydin, S. \u0026amp; Aky\u0026uuml;z, S. (2022). Learner agency in digital language education: A systematic review. \u003cem\u003eReCALL\u003c/em\u003e, 34(1), 23-44.\u003c/li\u003e\n\u003cli\u003eBaker, W. (2014). \u0026ldquo;It\u0026rsquo;s not their job to share content\u0026rdquo;: A case study of the role of teaching assistants in the diffusion of knowledge and the enhancement of research-teaching linkages in undergraduate courses. \u003cem\u003eHigher Education,\u003c/em\u003e 67(5), 581-594.\u003c/li\u003e\n\u003cli\u003eBakhtin, M. M. (1981). The dialogic imagination: Four essays (M. Holquist, Ed.; C. Emerson \u0026amp; M. Holquist, Trans.). University of Texas Press.\u003c/li\u003e\n\u003cli\u003eBandura, A. (2001). Social cognitive theory: An agentic perspective. \u003cem\u003eAnnual Review of Psychology\u003c/em\u003e, 52(1), 1\u0026ndash;26.\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2006). Using thematic analysis in psychology. \u003cem\u003eQualitative Research in Psychology\u003c/em\u003e, 3(2), 77-101.\u003c/li\u003e\n\u003cli\u003eBrynjolfsson, E., Ikeda, S., \u0026amp; Liang, J. (2023). Augmenting human learning and agency with AI: Emerging evidence and considerations. \u003cem\u003eHarvard Business Review\u003c/em\u003e, 101(3), 74-79.\u003c/li\u003e\n\u003cli\u003eChapelle, C. A., \u0026amp; Sauro, S. (Eds.). (2017). \u003cem\u003eThe handbook of technology and second language teaching and learning\u003c/em\u003e. John Wiley \u0026amp; Sons.\u003c/li\u003e\n\u003cli\u003eChen, W., \u0026amp; Xie, H. (2024). Digital literacy and agency in AI-enhanced academic writing: Evidence from ESL classrooms. \u003cem\u003eLanguage Learning \u0026amp; Technology\u003c/em\u003e, 28(1), 33-48.\u003c/li\u003e\n\u003cli\u003eClark, E., Ross, A. S., Tan, C., Ji, Y., \u0026amp; Smith, N. A. (2018). \u003cem\u003eCreative writing with a machine in the loop: Case studies on slogans and stories\u003c/em\u003e. In 23rd International Conference on Intelligent User Interfaces (pp. 329-340).\u003c/li\u003e\n\u003cli\u003eDattathrani, S., \u0026amp; De\u0026rsquo;, A. (2023). Technomediated agency: Reconceptualizing authorship in human-AI collaborative writing.\u003cem\u003e Computers and Composition\u003c/em\u003e, 68, 102814.\u003c/li\u003e\n\u003cli\u003eDe Bot, K., \u0026amp; Zhang, L. (2022). Distributed cognition in digital academic writing: Human\u0026ndash;machine partnerships in practice. \u003cem\u003eLanguage Teaching Research\u003c/em\u003e, 26(4), 524-538.\u003c/li\u003e\n\u003cli\u003eDeters, P., Gao, X., Miller, E. R., \u0026amp; Vitanova, G. (Eds.). (2015). \u003cem\u003eTheorizing and analyzing agency in second language learning: Interdisciplinary approaches\u003c/em\u003e. Multilingual Matters.\u003c/li\u003e\n\u003cli\u003eDing, H. (2008). Negotiating voice and agency in L2 academic writing: Tensions between discourse conformity and self-expression. \u003cem\u003eJournal of Second Language Writing\u003c/em\u003e, 17(3), 101-120.\u003c/li\u003e\n\u003cli\u003eDuff, P. A. (2012). Identity, agency, and second language acquisition. Routledge.\u003c/li\u003e\n\u003cli\u003eEmirbayer, M., \u0026amp; Mische, A. (1998). What is agency? \u003cem\u003eAmerican Journal of Sociology\u003c/em\u003e, 103(4), 962-1022.\u003c/li\u003e\n\u003cli\u003eEngestr\u0026ouml;m, Y. (2001). Expansive learning at work: Toward an activity theoretical reconceptualization. \u003cem\u003eJournal of Education and Work\u003c/em\u003e, 14(1), 133-156.\u003c/li\u003e\n\u003cli\u003eFeng, L., Wong, L. H., \u0026amp; Chen, W. (2022). Effects and moderating factors of automated writing evaluation on L2 learners\u0026rsquo; writing self-efficacy and behaviors. \u003cem\u003eComputers \u0026amp; Education\u003c/em\u003e, 176, 104356.\u003c/li\u003e\n\u003cli\u003eFlavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive\u0026ndash;developmental inquiry. \u003cem\u003eAmerican Psychologist\u003c/em\u003e, 34(10), 906-911.\u003c/li\u003e\n\u003cli\u003eJin, L., \u0026amp; Deifell, E. (2013). Foreign language learners\u0026rsquo; use and perception of online dictionaries: A survey study. \u003cem\u003eMERLOT Journal of Online Learning and Teaching\u003c/em\u003e, 9(4), 515-533.\u003c/li\u003e\n\u003cli\u003eGiddens, A. (1984). \u003cem\u003eThe constitution of society: Outline of the theory of structuration\u003c/em\u003e. Polity Press.\u003c/li\u003e\n\u003cli\u003eKaptelinin, V. (2005). The object of activity: Making sense of the sense-maker. Mind, Culture, and Activity, 12(1), 4-18.\u003c/li\u003e\n\u003cli\u003eKaptelinin, V., \u0026amp; Nardi, B. A. (2006). \u003cem\u003eActing with technology: Activity theory and interaction design\u003c/em\u003e. MIT Press.\u003c/li\u003e\n\u003cli\u003eKim, Y., \u0026amp; Reeves, T. (2023). Affective agency in digital literacy practices. Language Learning \u0026amp; Technology, 27(2), 121-138.\u003c/li\u003e\n\u003cli\u003eKohnke, L., Zou, D., \u0026amp; Zhang, R. (2022). Scaffolded feedback and student agency in AI-supported EFL writing. ReCALL, 34(2), 165-182.\u003c/li\u003e\n\u003cli\u003eLantolf, J. P., \u0026amp; Thorne, S. L. (2006). \u003cem\u003eSociocultural theory and the genesis of second language development\u003c/em\u003e. Oxford University Press.\u003c/li\u003e\n\u003cli\u003eLarsen-Freeman, D. (2019). On language learner agency: A complex dynamic systems theory perspective. \u003cem\u003eModern Language Journal\u003c/em\u003e, 103(S1), 61-78.\u003c/li\u003e\n\u003cli\u003eLave, J., \u0026amp; Wenger, E. (1991). \u003cem\u003eSituated learning: Legitimate peripheral participation\u003c/em\u003e. Cambridge University Press.\u003c/li\u003e\n\u003cli\u003eLee, E., Kim, H., \u0026amp; Reeves, T. C. (2023). Student motivation and affect in technology-rich language writing.\u003cem\u003e Computer Assisted Language Learning\u003c/em\u003e, 36(2), 131-149.\u003c/li\u003e\n\u003cli\u003eLee, J., \u0026amp; Wang, K. (2024). Student agency in AI-integrated EFL writing: A longitudinal study. \u003cem\u003eComputers and Composition\u003c/em\u003e, 72, 102760.\u003c/li\u003e\n\u003cli\u003eLi, M., \u0026amp; Zhu, W. (2013). Patterns of computer-mediated interaction in small writing groups using wikis. \u003cem\u003eComputer Assisted Language Learning\u003c/em\u003e, 26(1), 61-82.\u003c/li\u003e\n\u003cli\u003eLin, M., \u0026amp; Lee, J. (2022). Metacognition in AI-supported writing tasks: A review. \u003cem\u003eTESOL Quarterly\u003c/em\u003e, 56(3), 828-845.\u003c/li\u003e\n\u003cli\u003eLingard, L. (2023). Reconfiguring agency in the age of artificial intelligence: Dialogic possibilities for language classrooms. \u003cem\u003eTESOL Quarterly\u003c/em\u003e, 57(3), 901\u0026ndash;924.\u003c/li\u003e\n\u003cli\u003eLo, C. K., Chan, W. H., Ng, T. K., \u0026amp; Wong, K. Y. (2024). Generative AI in education: Balancing personalized learning and learner autonomy erosion. \u003cem\u003eComputers \u0026amp; Education\u003c/em\u003e, 215, 105025.\u003c/li\u003e\n\u003cli\u003eLuria, A. R. (1976). \u003cem\u003eCognitive development: Its cultural and social foundations\u003c/em\u003e. Harvard University Press.\u003c/li\u003e\n\u003cli\u003eMa, Y., \u0026amp; Chen, M (2024). AI-empowered applications effects on EFL learners\u0026rsquo; engagement in the classroom and academic procrastination. BMC Psychol 12, 739. \u003c/li\u003e\n\u003cli\u003eMolenaar, I. (2022). \u003cem\u003eScaffolding agency in AI-enhanced learning environments: A dynamic systems approach\u003c/em\u003e. Paper presented at the International Conference on Artificial Intelligence in Education, Durham, United Kingdom.\u003c/li\u003e\n\u003cli\u003eNiekerk, J. F., Schmidt, T. L., van der Merwe, A. R., \u0026amp; Chen, X. (2025). Generative AI and learner agency in digital learning ecologies: A sociotechnical perspective. Educational Technology Research and Development, Advance online publication.\u003c/li\u003e\n\u003cli\u003ePriestley, M., Biesta, G., \u0026amp; Robinson, S. (2015). \u003cem\u003eTeacher agency: An ecological approach\u003c/em\u003e. In M. Priestley \u0026amp; G. Biesta (Eds.), Reinventing the curriculum: New trends in curriculum policy and practice (pp. 1\u0026ndash;20). Bloomsbury Academic.\u003c/li\u003e\n\u003cli\u003ePrior, P. (2006). \u003cem\u003eA sociocultural theory of writing\u003c/em\u003e. In C. A. MacArthur, S. Graham, \u0026amp; J. Fitzgerald (Eds.), Handbook of writing research (pp. 54-66). The Guilford Press.\u003c/li\u003e\n\u003cli\u003eRainio, A. P., \u0026amp; Hilpp\u0026ouml;, J. (2017). The dialectics of agency in educational ethnography. \u003cem\u003eEthnography and Education\u003c/em\u003e, 12(1), 78-94.\u003c/li\u003e\n\u003cli\u003eReich, J., Ito, M., \u0026amp; Richards, J. (2023). Task design for engagement in AI-mediated instruction. \u003cem\u003eComputers \u0026amp; Education\u003c/em\u003e, 193, 104698.\u003c/li\u003e\n\u003cli\u003eRoe, A. B., \u0026amp; Perkins, C. D. (2024). A\u003cem\u003egency in educational innovation: Negotiating identity and sociotechnical change\u003c/em\u003e. In T. J. Li \u0026amp; G. S. Park (Eds.), Advances in language education research (pp. 145\u0026ndash;167). Springer. \u003c/li\u003e\n\u003cli\u003eRose, J., \u0026amp; Jones, M. (2005). The architecture of opportunity: Socio-technical systems and the distribution of agency in digital environments. MIT Press.\u003c/li\u003e\n\u003cli\u003eSaghafian, H., Hayati, K., \u0026amp; Eskandari, Z. (2022). Learners\u0026rsquo; emotions, motivation, and self-efficacy in AI-integrated language learning: A mixed-methods study. \u003cem\u003eLanguage Teaching Research\u003c/em\u003e, 26(6), 944\u0026ndash;962.\u003c/li\u003e\n\u003cli\u003eSzabo, T. L., \u0026amp; Szoke, I. (2024). \u003cem\u003eTheorizing machine agency in language learning environments: A posthumanist lens\u003c/em\u003e. Technology, Pedagogy and Education, Advance online publication. \u003c/li\u003e\n\u003cli\u003eSun, Y., Yin, Y., Wang, F., \u0026amp; Chen, N.S. (2023). The role of affective factors in AI-based language learning. \u003cem\u003eComputers \u0026amp; Education: Artificial Intelligence\u003c/em\u003e, 4, 100175.\u003c/li\u003e\n\u003cli\u003eTao, J., \u0026amp; Gao, X. (2021). Language teacher agency in emergency remote teaching: A longitudinal narrative inquiry.\u003cem\u003e System\u003c/em\u003e, 103, 102660.\u003c/li\u003e\n\u003cli\u003eTrust, T., et al. (2023). Promoting reflective digital practices in AI writing contexts. \u003cem\u003eBritish Journal of Educational Technology\u003c/em\u003e, 54(1), 42-62.\u003c/li\u003e\n\u003cli\u003eValsiner, J. (1997). Culture and the development of children\u0026rsquo;s action: A theory of human development (2nd ed.). John Wiley \u0026amp; Sons.\u003c/li\u003e\n\u003cli\u003evan Lier, L. (2008). Agency in the classroom. In J. P. Lantolf \u0026amp; M. E. Poehner (Eds.), Sociocultural theory and the teaching of second languages (pp. 163-186). Equinox.\u003c/li\u003e\n\u003cli\u003eVygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (M. Cole, V. John-Steiner, S. Scribner, \u0026amp; E. Souberman, Eds.). Harvard University Press.\u003c/li\u003e\n\u003cli\u003eVygotsky, L. S. (1986). \u003cem\u003eThought and language\u003c/em\u003e (A. Kozulin, Trans.). MIT Press.\u003c/li\u003e\n\u003cli\u003eWarschauer, M. (2005).\u003cem\u003e Sociocultural perspectives on CALL\u003c/em\u003e. In J. Egbert \u0026amp; G. M. Petrie (Eds.), CALL research perspectives (pp. 41-51). Lawrence Erlbaum Associates.\u003c/li\u003e\n\u003cli\u003eWertsch, J. V. (1998).\u003cem\u003e Mind as action\u003c/em\u003e. Oxford University Press.\u003c/li\u003e\n\u003cli\u003eYi, Y., \u0026amp; Angay-Crowder, T. (2016). Multimodal pedagogies for teacher education in TESOL. \u003cem\u003eTESOL Quarterly\u003c/em\u003e, 50(4), 988-998.\u003c/li\u003e\n\u003cli\u003eYin, R. K. (2018). \u003cem\u003eCase study research and applications: Design and methods\u003c/em\u003e (6th ed.). SAGE Publications.\u003c/li\u003e\n\u003cli\u003eZhang, Y., \u0026amp; Tur, G. (2024). Agency at risk? Generative AI and the paradox of self-directed learning. \u003cem\u003eAustralasian Journal of Educational Technology\u003c/em\u003e, 40(3), 1-17. \u003c/li\u003e\n\u003cli\u003eZheng, B., Warschauer, M., \u0026amp; Farkas, G. (2022). Engaging with AI: Agency and accountability in learning with emerging technologies. \u003cem\u003eBritish Journal of Educational Technology\u003c/em\u003e, 53(4), 786-802.\u003c/li\u003e\n\u003cli\u003eZheng, D., \u0026amp; Newgarden, K. (2012). Rethinking language learning: Virtual worlds as a catalyst for change. \u003cem\u003eInternational Journal of Learning and Media\u003c/em\u003e, 3(2), 13-36.\u003c/li\u003e\n\u003cli\u003eZou, D., Lin, J., \u0026amp; Sun, J. (2021). Metacognition and AI literacy in EFL writing. \u003cem\u003eReCALL\u003c/em\u003e, 33(1), 113-130.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"EFL writing, generative artificial intelligence, learner agency, sociocultural theory, human-AI collaboration, academic writing","lastPublishedDoi":"10.21203/rs.3.rs-7972095/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7972095/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study investigates how EFL learners construct, negotiate, and develop writing agency when engaging with generative artificial intelligence (GenAI) tools in second language writing courses, conceptualizing these technologies as cultural mediational means within an integrated sociocultural and positive psychological theoretical framework. Eleven Chinese EFL university learners were observed over a 16-week period as they completed multiple academic writing tasks using GenAI assistance. Data collection employed a qualitative multi-method approach including screen recordings of writing sessions, stimulated recall interviews, reflective journals, and sequential drafts with AI interaction logs. Learners\u0026rsquo; agency trajectories evolved through three developmental phases: initial exploration (characterized by novelty, dependency, and emergent self-efficacy), strategic adaptation (marked by selective tool use and growing writing-specific psychological capital), and deliberate appropriation (reflecting personalized integration and flourishing techno-authorial identity). Four patterns of agency manifestation emerged: transitional agency (shifting dependencies with emerging metacognitive awareness and increased resilience), distributed agency (strategic negotiation of textual authority fostering growth mindsets), reflective agency (critical evaluation of AI contributions coupled with self-determination), and transformative agency (reconceptualization of writer identities within human-AI collaboration leading to authentic engagement and eudaimonic well-being). Cognitive processes, affective dimensions, positive psychological resources, and task design characteristics significantly mediated how learners constructed agency through interactions with technological affordances. This study extends theoretical understanding of agency in GenAI-assisted educational contexts by illuminating the dynamic, developmental nature of writing agency and identifying specific patterns of agency manifestation in human-AI collaborative writing, while bridging sociocultural perspectives with positive psychological constructs that support optimal functioning.\u003c/p\u003e","manuscriptTitle":"‘AI-Learner Partnership:’: Psychological Mechanisms and Developmental Trajectories of EFL Learners’ Writing Agency in GenAI-Assisted Courses","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-18 08:47:45","doi":"10.21203/rs.3.rs-7972095/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bb51ec1b-c443-4105-90c0-371412b71366","owner":[],"postedDate":"November 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-12-09T11:24:02+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-18 08:47:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7972095","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7972095","identity":"rs-7972095","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}