The digitalization of society has recently given rise to a new wave of interest in artificial intelligence (machine learning) and its organizational consequences for work and professional knowledge (Faraj, Pachidi et al. 2018). Unlike traditional technology, modern artificial intelligence (AI) is self-learning in the sense that it can autonomously adjust its output (predictions, categorizations, etc.) based on the data that is fed to it. This makes the results and consequences of AI inherently dynamic and changing and sometimes difficult to predict and explain not only by professionals using the systems but even by its creators and developers. AI has therefore been described as a black box and as being opaque to its users (Burrell 2016). This raises the question how sensemaking and understanding of modern day AI is coordinated and practically accomplished in situ by developers and practitioners. The understanding of technology is always situated in particular social practices and cultural contexts (Winograd, Flores et al. 1986, Engeström and Middleton 1996, Coeckelbergh and Funk 2018). Developers and users are often participating in disparate communities of practice with little overlap and mutual knowledge. Technologies that are eventually adopted and enacted by users and their professional communities – as intended by designers or by appropriating functions (Pachidi, Berends et al. 2021) – sometimes end up disrupting and re-shaping epistemic practices; perceptions, relations and meanings constituting the world for that group (Sergeeva, Faraj et al. 2020, Anthony 2021). This is particularly a possibility in the case of digital technologies that are integrated with artificial intelligence, such as deep neural networks trained on massive amounts of big data (Lebovitz, Lifshitz-Assaf et al. 2022). AI has the potential to radically transform many traditional professions (such as medicine and law) by changing old or opening new epistemic domains hitherto obscured from view for the practitioners, thereby expanding existing realms of knowing. This could imply a shift in workplace learning and potentially in professional jurisdictions (Abbott 1988); With more power and authority given (or taken) by computer scientists in domains where learning is traditionally built on apprenticeship and experience, AI is likely to spur new practices and divisions of expert labor. Now, the organizational process of implementing and making sense of AI and the transition into new modes of representing and conceiving traditional epistemic objects is not determined by the technology or formal institutional structures (Orlikowski 2000). Understanding is rather likely to be an interactional process of knowledge-sharing, communication and negotiation between developers and users taking place in situ and over extended periods of time. It is, however, currently unclear how developers of AI systems coordinate and communicate with professionals regarding the principles, possibilities, functions, risks and limitations of systems designed for particular domains and uses. Researchers in fields such as information systems and management, have become increasingly interested in how AI tools are actually developed and implemented, and with what consequences for work, knowledge and organizing. A growing number of studies have for instance studied AI in organization and work ethnographically; including Waardenburg, Huysman et al. (2022) study on knowledge brokerage in predictive policing, Lebovitz, Lifshitz-Assaf et al. (2022) investigation of algorithmic opacity in medical diagnoses, van den Broek, Sergeeva et al. (2021) examination of AI in hiring processes, and many more (Barrett, Oborn et al. 2012, Schwennesen 2019, Grønsund and Aanestad 2020).  Aim of this study In this study we analyze the social interaction between developers and professionals collaborating in a project aimed to develop and implement AI in shipping. We base the investigation on a workplace study approach which is a theoretical-methodological framework for analyzing how technologies feature in organization and work (Heath and Luff 2000, Luff, Hindmarsh et al. 2000). Our research question is: How are explanations and understandings of AI in the context of a particular work practice produced and coordinated (practically accomplished) between developers and users?