There is a vast amount of literature on collaborative writing in second language teaching and learning, much of it inspired by Storch (2002 - 2015). Although the topic of collaborative writing has been researched extensively, few studies have addressed the individual learners from an agentic perspective (Bitchener & Storch, 2016; Li & Zhu, 2017a; Yu & Lee, 2016). None to the best of my knowledge investigated learners’ student agency using Bandura’s (1989, 2001, 2006) four human agentic characteristics. Moreover, while some researchers (Blin & Appel, 2011; Yu & Lee, 2015) have attempted to explain the complexity of collaborative writing using Engeström’s (1987, 1999) activity theory framework, few examined the role learners’ human agency plays in their group activity of collaborative writing. Therefore, the present study attempts to investigate adult English language learners’ practices and perceptions of collaborative writing from an agentic perspective. Finally, while much collaborative writing research has been informed by sociocultural theory, the present study has adopted a sociocognitive approach (Atkinson, 2002, 2010, 2014) taking a learner’s mind, body and world as an inseparable, but adaptive unit. Research has shown that collaborative writing can offer a number of benefits that are not found in other approaches to teaching writing. These benefits are made possible because interactions with other learners during the process of writing can provide additional learning opportunities through peer discussions, peer feedforward and peer feedback. In this way, learners are mutually able to scaffold one another’s learning and writing development. Past studies have also revealed the interactions of learners in a group can play a crucial role in the effectiveness of peer scaffolding. While the majority of studies have investigated the issue by applying Storch’s (2002) dyadic interaction model based on the concepts of equality and mutuality, few have examined triadic interactions in such depth. The present study aims to better understand how learners interact in triads when completing collaborative writing tasks. Moreover, learners have generally been analysed as a collective unit for the understanding of patterns of interactions. While this may help with identifying why certain pairs/groups are more successful than others, it does not explain why learners behave differently. Therefore, this study attempts to contribute to this area by explaining collaborative writing from an agentic perspective and how the individual learners can be an active change agent in their own learning activity. Collaborative writing tasks are often implemented either in a conventional classroom or on an online platform, each of which has advantages and disadvantages. However, the two platforms are rarely blended in the same study where learners are required to interact on both platforms to jointly complete one or more pieces of writing. The design of the present study has adopted a blended learning platform for the implementation of its collaborative writing tasks. Finally, as a teacher, researching this topic in my own classroom has not only helped me to achieve a better understanding my own beliefs and practices regarding the teaching of writing to adult English language learners, but it has also helped me to generate a personal theory of learning which may be applied in wider contexts. The present study was an action research project conducted from May to October 2016 in the context of a university language centre in New Zealand. It adopted an interpretive approach, believing each individual learner will develop a unique experience, perception and interpretation of learning through a blended collaborative approach to writing. The study examined 21 adult English language learners in their 20s from five different countries. Data were collected through a combination of pre- and post-course essays, pre- and post-course narrative frames, written drafts of group assignments, audio recordings of class discussions, text-based online communication and focus group sessions. All data were subjected to a process of grounded analysis. This multi-method approach has provided a detailed picture of both the participants’ perceptions and practices. Firstly, this was achieved by assessing participants’ pre- and post-course essays for the effectiveness of the blended collaborative approach. Secondly, participants’ interactions within their triads were transcribed and analysed for evidence of language learning and their developing relationships with their group members. Thirdly, participants’ reported perceptions and experiences of triadic collaborative writing were analysed and triangulated with their observable practices. In brief, findings revealed that the effectiveness of the triadic collaborative approach to writing in a blended learning environment appeared to be largely associated with a triad’s patterns of interactions. In addition, differences in learners’ collaborative behaviour which contributed to their patterns of interactions in triads were connected with the extent to which the learners practised their agentic potential by adapting and aligning their actions in and on reflections with their intentionality and forethought, which are the four human agentic characteristics examined in the present study. Finally, action research was a powerful tool for the teacher-researcher’s own professional development at both a pedagogical and theoretical level.