Cognitive Stitching: Informal AI Integration in Everyday Clinical Reasoning

Cognitive Stitching: Informal AI Integration in Everyday Clinical Reasoning | 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 Article Cognitive Stitching: Informal AI Integration in Everyday Clinical Reasoning Zongtian Guo This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9095240/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Artificial intelligence is often discussed in healthcare in terms of algorithmic performance, regulatory approval, or institutional deployment. Such perspectives tend to treat AI as a formally implemented technology integrated into hospital infrastructures. Yet relatively little attention has been paid to how clinicians actually encounter and use AI within the everyday realities of clinical work. Drawing on ethnographic fieldwork conducted in Chinese public hospitals, this article examines how clinicians incorporate AI tools into diagnostic reasoning when institutional digital infrastructures prove incomplete or inflexible. The analysis shows that AI frequently enters clinical practice not through officially deployed hospital systems but through informal practices involving personal mobile devices and widely accessible consumer applications. When institutional platforms fail to provide sufficient interpretive support, clinicians turn to external digital resources to organize information, explore diagnostic possibilities, and manage cognitive uncertainty. To conceptualize these practices, the article introduces the notion of cognitive stitching, referring to the practical process through which clinicians assemble heterogeneous informational resources—including hospital systems, online knowledge sources, professional communication networks, and AI tools—into workable diagnostic interpretations. Scientific community and society/Business and industry Health sciences/Health care Physical sciences/Mathematics and computing Scientific community and society/Scientific community Scientific community and society/Social sciences Figures Figure 1 Introduction Artificial intelligence (AI) has rapidly entered contemporary medical discourse. Across policy reports, clinical journals, and technology media, AI is frequently portrayed as a transformative force capable of reshaping the future of healthcare. Machine learning systems promise earlier diagnoses, more accurate image interpretation, and the ability to analyze vast volumes of biomedical data beyond the limits of human cognition (Topol, 2019 ; Yu et al., 2018 ). In radiology, pathology, and oncology in particular, algorithmic systems are often presented as tools that may fundamentally reconfigure diagnostic work and professional expertise (Rajpurkar et al., 2022 ). Yet alongside these optimistic projections runs an equally visible current of skepticism. Critics warn that algorithmic medicine may introduce new forms of bias, erode professional autonomy, or create opaque decision-making processes that are difficult to audit or contest (Babic et al., 2021 ; Obermeyer & Emanuel, 2016 ). As a result, discussions about AI in healthcare often oscillate between technological optimism and technological anxiety. Despite these debates, much of the existing literature shares a common analytical focus: the evaluation and implementation of AI systems as formal technologies. A large body of research examines the accuracy of diagnostic algorithms, comparing their performance with that of human clinicians (Esteva et al., 2019 ; Guo & Cugurullo, 2023 ; Rajpurkar et al., 2022 ). Other studies explore the institutional integration of AI within healthcare infrastructures, analyzing regulatory frameworks, clinical validation processes, and the organizational conditions required for successful deployment (Reddy et al., 2019 ). While these approaches provide important insights into the capabilities and governance of medical AI, they tend to treat algorithms as discrete technological systems whose impact can be assessed through performance metrics or institutional adoption. What remains less visible in these discussions is the everyday life of AI in clinical practice. Hospitals are complex sociotechnical environments in which multiple technologies coexist: electronic health records, diagnostic imaging platforms, clinical guidelines, communication systems, and an expanding range of algorithmic tools (Bowker & Star, 2000 ; Star & Ruhleder, 2010 ). Within such environments, technologies rarely operate as seamlessly integrated systems (Orlikowski, 2000 ). Instead, they are continuously adapted, bypassed, or supplemented by practitioners responding to the practical demands of patient care. Research in STS and medical sociology has long emphasized that technologies acquire meaning and function through situated practice rather than through design alone (Berg, 1997 ; Suchman, 2007 ; Timmermans & Berg, 2010 ). From this perspective, the key question is not simply whether AI systems work, but how they are actually incorporated into the routines and reasoning processes of clinicians. Recent studies of healthcare technologies have highlighted the prevalence of informal practices and workarounds in digital clinical environments (Laurie, 2017 ). When institutional information systems prove inflexible or poorly aligned with clinical workflows, practitioners often develop pragmatic strategies to accomplish their tasks (Ash et al., 2004 ; Ellingsen & Monteiro, 2003 ; Koppel et al., 2008 ). These practices can include bypassing official software functions, relying on personal devices, or informally sharing knowledge through professional networks. Such forms of technological improvisation are not merely marginal deviations from formal infrastructures. Rather, they constitute an important part of how complex healthcare systems continue to function in practice.This article builds on these insights by examining how clinicians incorporate emerging AI tools into everyday diagnostic reasoning through informal practices. Drawing on ethnographic fieldwork conducted in Chinese public hospitals, the study investigates how clinicians navigate the gap between institutional digital infrastructures and the practical demands of clinical work. Over the past decade, Chinese hospitals have invested heavily in digital health technologies, including hospital information systems, picture archiving and communication systems, and specialized medical AI applications designed for tasks such as image analysis or treatment planning (He et al., 2019 ; Liang et al., 2019 ). Yet clinicians frequently encounter limitations in these institutional platforms, including rigid software interfaces, fragmented data systems, and time-consuming workflows (Chang et al., 2016 ; Deloitte, 2021 ; Gao & Zhang, 2021 ; Liu et al., 2008 ). These frictions become particularly visible in high-pressure contexts such as night shifts or emergency consultations, where clinicians must make rapid decisions with limited cognitive and informational resources. In these moments, AI does not always appear through officially deployed hospital systems. Instead, clinicians often turn to consumer-grade AI tools accessed through their personal smartphones. Applications based on large language models, for example, are sometimes consulted to clarify unfamiliar drug interactions, generate possible diagnostic hypotheses, or organize complex clinical information. These interactions typically occur quietly and informally. Junior doctors describe briefly checking an AI application when confronted with unfamiliar cases, while senior clinicians occasionally circulate screenshots of AI-generated explanations in professional messaging groups to stimulate discussion among colleagues. Although such practices remain outside formal institutional protocols, they increasingly form part of the everyday repertoire through which clinicians manage cognitive uncertainty in demanding clinical environments. To conceptualize these dynamics, this article introduces the notion of cognitive stitching. The concept refers to the practical process through which clinicians assemble multiple technological resources—official hospital systems, online databases, personal devices, and AI applications—to bridge gaps in institutional infrastructures. Rather than functioning as autonomous decision-makers, AI tools in these contexts become one element within a broader patchwork of clinical reasoning. Clinicians selectively draw on these tools to extend their capacity to process information, compare possible interpretations, or structure complex diagnostic thinking. In this sense, AI becomes woven into clinical cognition not through formal integration alone but through everyday acts of technological improvisation. By focusing on these practices, the article addresses a gap in existing research on medical AI. While much scholarship concentrates on algorithmic accuracy or regulatory frameworks, relatively little attention has been paid to how clinicians incorporate AI into the lived realities of clinical work. Understanding these practices is crucial because technologies rarely reshape professional domains through formal implementation alone. Instead, they are gradually integrated through the mundane adaptations of practitioners navigating institutional constraints. The central research question guiding this study is therefore: how do clinicians incorporate AI tools into everyday clinical reasoning when institutional infrastructures prove incomplete or inflexible? By examining this question ethnographically, the article highlights how AI becomes embedded in clinical practice through informal and often improvised arrangements. The article proceeds as follows. The next section outlines the methodological approach and fieldwork through which these dynamics were documented. The empirical analysis then examines how institutional frictions within hospital digital infrastructures create the conditions for cognitive stitching. Through detailed ethnographic accounts, the paper shows how clinicians move between institutional platforms and personal AI tools while managing diagnostic uncertainty in everyday clinical work. The subsequent discussion explores the professional and ethical tensions that emerge when algorithmic assistance becomes informally embedded within clinical reasoning. Finally, the conclusion briefly reflects on these dynamics through a comparative perspective, contrasting informal forms of AI incorporation with more formally regulated models of medical AI adoption. Methods This article draws on ethnographic fieldwork conducted in public hospitals in China between 2024 and 2025. The broader research project explored how emerging digital technologies are reshaping clinical work and professional knowledge in contemporary healthcare settings. Rather than evaluating the technical performance of medical AI systems, the fieldwork focused on how clinicians encounter and incorporate digital tools within the everyday routines of diagnosis, consultation, and decision-making. The research was conducted primarily in large public hospitals located in major Chinese cities. These institutions represent the dominant organizational form of healthcare delivery in China and are characterized by extremely high patient volumes, complex administrative structures, and rapidly expanding digital infrastructures. Over the past decade, Chinese hospitals have invested heavily in information technologies, including hospital information systems, electronic medical records, picture archiving and communication systems, and a growing number of algorithmic applications designed to assist with tasks such as medical imaging interpretation or treatment planning. On paper, these infrastructures suggest a highly digitalized clinical environment in which information flows smoothly between different technological platforms. In practice, however, the everyday experience of clinicians often reveals a more fragmented landscape. Doctors working in tertiary hospitals frequently face intense workloads and time pressure, with outpatient consultations sometimes lasting only a few minutes. Diagnostic departments process large numbers of imaging scans each day, requiring physicians to review hundreds of images within limited time frames (Lipset, 2017 ). Within such environments, digital technologies do not function as seamless infrastructures but as a collection of systems that clinicians must navigate while maintaining the pace of clinical work. These conditions formed the broader context within which the fieldwork was conducted. The empirical material presented in this article is based primarily on qualitative interviews with clinicians working across multiple departments. In total, thirty-four interviews were conducted with medical professionals, including diagnostic radiologists, medical physicists, clinicians from other specialties, and specialist nurses involved in technologically mediated care practices. Participants ranged from junior residents in the early stages of their careers to senior physicians with more than two decades of clinical experience. This diversity allowed the research to capture how clinicians with different levels of professional authority and technological familiarity encounter AI in their everyday work. Interviews typically lasted between twenty and sixty minutes and were conducted in locations such as departmental offices, hospital meeting rooms, or through remote communication when in-person meetings were not possible. Conversations were semi-structured, allowing participants to describe their experiences with digital technologies in their own terms. Rather than focusing exclusively on officially deployed AI systems, discussions often moved toward broader questions about how clinicians search for information, manage diagnostic uncertainty, and cope with moments of cognitive overload during demanding shifts. Participants were frequently encouraged to recall concrete situations in which technological tools played a role in shaping their diagnostic reasoning. Alongside these interviews, the research also incorporated elements of ethnographic observation within clinical environments. Time was spent in workspaces such as radiology reading rooms and departmental offices, where clinicians interact continuously with digital systems while performing diagnostic work. Observing these environments provided insight into how multiple technological interfaces coexist within clinical practice. Hospital workstations displaying imaging data often operate alongside personal devices, online medical databases, and messaging platforms through which clinicians communicate with colleagues. These observations helped situate interview accounts within the material environments in which clinical reasoning takes place. All participants were informed about the purpose of the research and agreed to participate on the condition that their identities and institutional affiliations would remain anonymous. To protect confidentiality, identifying details such as hospital names and personal information have been removed from the material presented in this article. The analysis of the material followed an iterative qualitative process informed by ethnographic approaches to technology and work. Interview transcripts and field notes were first reviewed to identify recurring descriptions of how clinicians rely on digital tools while navigating diagnostic uncertainty. Particular attention was paid to moments in which participants described moving between different technological resources while making clinical decisions. As the analysis progressed, it became increasingly clear that clinicians rarely relied on a single technological system. Instead, they frequently assembled information from multiple platforms, combining institutional infrastructures with external digital resources. To make this analytical process more systematic, the material was coded using an iterative qualitative coding framework informed by ethnographic research on technology in professional practice {Charmaz, 2014 #1627}. In the first stage, interview transcripts and field notes were open coded to identify recurring descriptions of how clinicians interacted with digital systems during diagnostic work. These initial codes captured themes such as information search practices, moments of diagnostic uncertainty, interactions with hospital information systems, and the use of external digital resources. In a second stage of analysis, related codes were grouped into broader analytical categories focusing on the movement between institutional infrastructures and external informational tools. Particular attention was paid to episodes in which clinicians described shifting from hospital workstations to personal devices while attempting to interpret complex cases. Observational field notes were used to contextualize these accounts by documenting how clinicians navigated multiple technological interfaces within clinical workspaces, including imaging systems, electronic health records, and personal mobile devices. Through repeated comparison of interview accounts and observational material, these patterns were gradually synthesized into the analytical concept developed in this article as cognitive stitching. Finding The following section examines how clinicians navigate the gaps and tensions that emerge within hospital digital infrastructures during everyday clinical work. Rather than treating artificial intelligence as a formally implemented technology embedded within institutional systems, the analysis focuses on the practical conditions under which clinicians turn to alternative informational resources while interpreting complex medical cases. In highly digitalized hospital environments, information systems are designed to structure the flow of clinical data, guide diagnostic procedures, and standardize decision-making processes. These infrastructures promise a form of technological order in which patient information is centralized, searchable, and accessible through standardized interfaces. Yet the everyday experience of working within these systems often appears far more complicated. As clinicians repeatedly emphasized during fieldwork, the systems intended to simplify clinical reasoning sometimes generate new forms of friction that shape how doctors search for information and interpret clinical situations. In interviews, clinicians frequently described hospital information systems as simultaneously indispensable and frustrating. On the one hand, these platforms form the backbone of contemporary hospital operations. They store patient histories, display laboratory results, organize imaging data, and provide the structured templates through which clinical documentation must be completed. Without them, it would be nearly impossible to manage the enormous volume of patient information circulating through large tertiary hospitals. At the same time, however, doctors often described these systems as rigid environments that constrained the exploratory dimensions of medical reasoning. While the software functioned well for routine tasks such as retrieving patient records or reviewing imaging results, it was less effective when clinicians attempted to explore unfamiliar diagnostic possibilities or analyze complex relationships between symptoms, medications, and clinical findings. Several participants emphasized that the official platforms were designed primarily for documentation rather than interpretation. The systems organized information according to predefined categories and standardized procedures, reflecting administrative priorities of data management and regulatory compliance. Yet clinical reasoning frequently requires a more flexible mode of investigation in which physicians compare different sources of information, test alternative hypotheses, and interpret ambiguous patterns. When the structured logic of the software collided with these exploratory needs, clinicians often experienced the system less as an aid to reasoning than as a technological constraint. One physician described repeatedly encountering automated error messages while attempting to search for drug interaction information through the hospital’s internal database. The system was designed to identify known interactions between medications, but its query structure imposed strict limits on how searches could be performed. Sometimes you try to look up a combination of drugs and the system simply blocks the request,” the physician explained. “It just says ‘Forbidden’ and stops there. The doctor described the experience as particularly frustrating when working under time pressure. You’re trying to understand the situation quickly, but the system refuses to process the query. At that moment you feel like the technology is working against you. Although such restrictions were intended to prevent incorrect inputs or unauthorized searches, clinicians often experienced them as interruptions in the reasoning process. Instead of clarifying uncertainty, the system sometimes halted the search at precisely the moment when additional information was needed. Another physician described similar limitations when analyzing complex medication combinations. While the official database could easily identify standard interactions between two drugs, it struggled to process cases involving multiple medications with overlapping pharmacological effects. The system gives you a few standard warnings,” the doctor explained, “but when the case becomes complicated, it doesn’t really help you understand the relationships. The physician described situations in which the system produced fragments of information without offering a coherent explanation. “It shows you several alerts, but then you still have to figure out what they actually mean together.” In these situations, the system functioned less as an interpretive resource than as a collection of partial signals that required further reasoning. These experiences were not isolated incidents but recurring features of everyday clinical work. In large public hospitals, clinicians are expected to process large volumes of patient data while operating under considerable time pressure (Shanafelt et al., 2016 ; Wachter, 2017 ). Diagnostic departments often handle hundreds of imaging cases each day, while outpatient clinics may see dozens of patients during a single shift. Within such environments, digital infrastructures are intended to support rapid access to information and streamline clinical workflows. Yet the same infrastructures may become sources of friction when their structured logic fails to accommodate the complexity of real clinical situations. These tensions became particularly visible during night shifts. Several younger doctors described the experience of working alone in hospital departments after senior physicians had left for the day. During these shifts, clinicians were responsible for evaluating new cases while simultaneously monitoring ongoing patients and completing documentation tasks. Although institutional systems remained accessible, the absence of immediate consultation with colleagues intensified the pressure to resolve diagnostic uncertainties independently. One resident recalled a particularly challenging shift involving a patient whose symptoms did not correspond neatly with established diagnostic categories. It was already very late,” the doctor explained. “You’ve been reviewing images for hours, and suddenly you encounter a case that doesn’t fit any familiar pattern. The physician attempted to search the hospital system for guidance but found the available information insufficient. The system gave some basic references, but they didn’t really address the specific situation. The resident described the cognitive fatigue that often accompanies prolonged diagnostic work. Sometimes your brain just stops. You read the information again and again, but it still doesn’t connect. Moments like these illustrate how institutional infrastructures can generate what might be described as cognitive gaps. Hospital information systems are designed to standardize knowledge through predefined categories and procedural logic. Clinical reasoning, however, frequently requires the ability to move beyond these categories, combining fragments of information in order to interpret complex cases. When institutional platforms fail to support this exploratory process, clinicians must search for alternative ways of assembling the knowledge they need. It is within these moments of infrastructural limitation that a different pattern of technological practice begins to appear. Rather than abandoning digital tools altogether, clinicians expand the range of informational resources they draw upon. The search for answers gradually extends beyond the institutional workstation toward a broader digital landscape that includes external databases, online medical references, and increasingly, resources accessible through personal mobile devices. Understanding this shift requires recognizing that hospital technologies rarely function as unified infrastructures. Instead, clinicians operate within a heterogeneous technological environment composed of multiple overlapping systems. Institutional platforms structure official documentation and imaging analysis, while informal channels of information seeking operate alongside them. It is through the movement between these different environments that clinicians begin to assemble workable interpretations of complex clinical situations.The material layout of clinical workspaces often reflects this layered technological environment. In radiology reading rooms where part of the fieldwork was conducted, clinicians worked at stations equipped with several large monitors displaying CT and MRI images. The central screens were connected to the hospital’s imaging system, allowing doctors to scroll through hundreds of image slices while simultaneously reviewing patient records. Yet beside the keyboard, many clinicians kept their personal smartphones within easy reach, illustrating the coexistence of institutional and informal informational resources within everyday clinical work (Fig. 1 ). During difficult cases, it was common to see a doctor lean back from the workstation, briefly pick up their phone, and search for additional information while continuing to examine patient images on the screen. These movements between devices were usually quick and almost routine, lasting only a few seconds before clinicians returned their attention to the diagnostic workstation. Several clinicians described this movement between institutional and personal technologies as an almost habitual response to uncertainty. You always start with the hospital system,” one physician explained. “But if the information there isn’t enough, you naturally start looking elsewhere. For many participants, this “elsewhere” included external digital resources that could provide additional perspectives on unfamiliar clinical situations. Personal smartphones offered immediate access to medical reference websites, research articles, and communication platforms through which clinicians could consult colleagues. One physician described this shift as a practical adaptation to the limitations of institutional infrastructures. The hospital system is good for retrieving patient data,” the doctor explained. “But when you need to think through a complicated problem, you often need other sources. Through these everyday practices of information seeking, clinicians gradually extended the informational environment in which clinical reasoning takes place. As clinicians expanded their search for information beyond institutional infrastructures, artificial intelligence applications gradually began to appear within these informal pathways of information seeking. Although most hospitals involved in the study had experimented with specialized medical AI systems—particularly in imaging analysis—clinicians rarely described relying on these tools as part of their everyday reasoning process. Instead, the AI tools that appeared most frequently in interviews were consumer-oriented applications based on large language models, accessed through personal mobile devices. Participants emphasized that these tools were not used as formal diagnostic systems. Rather, they functioned as flexible informational resources that could help clinicians organize unfamiliar information, generate potential explanations, or structure complex reasoning processes. In this sense, AI entered clinical practice not as an officially sanctioned component of hospital infrastructures but as an informal supplement to existing knowledge resources. Several clinicians described using applications such as DeepSeek or Doubao during moments of diagnostic uncertainty. These interactions typically occurred when institutional systems failed to provide sufficient interpretive support for understanding complex cases. Under such circumstances, doctors occasionally turned to AI interfaces as a way of reorganizing fragmented pieces of information. One resident described encountering a complicated clinical situation during a late-night shift in which a patient presented with a combination of symptoms that did not correspond clearly to any familiar diagnosis. You read the laboratory results and imaging findings again and again,” the resident recalled. “But they still don’t connect in a clear way. After several unsuccessful attempts to search the hospital system for additional information, the physician opened an AI application on their phone and entered a short description of the case. “I wrote a few sentences describing the symptoms and the medications the patient was taking,” the doctor explained. The system generated a response outlining several possible diagnostic interpretations. It didn’t give a definitive answer,” the resident emphasized. “But it helped organize the information. Sometimes you just need something that helps you see the structure of the problem. For clinicians working under significant cognitive pressure, such interactions could provide a way of reorganizing information that initially appeared confusing or disconnected. Rather than replacing clinical judgment, AI responses were typically treated as provisional suggestions that required further verification through established medical knowledge. Another physician described using an AI application when encountering unfamiliar drug interactions that the hospital’s internal database could not easily interpret. While the official system could identify known interactions between two medications, it struggled to analyze more complex combinations involving multiple pharmacological effects. Sometimes the patient is taking five or six medications,” the doctor explained. “The hospital system only tells you about simple interactions. In such cases, the physician occasionally entered the list of medications into an AI interface to obtain a preliminary overview of possible relationships. The AI might suggest several possibilities,” the doctor said. “Then you check the guidelines or research papers to see whether those suggestions make sense. The physician emphasized that the AI response functioned primarily as an organizational tool rather than a source of authoritative knowledge. It helps you think,” the doctor explained. “But the final interpretation still depends on your own judgment. These accounts suggest that clinicians often used AI applications as cognitive aids rather than diagnostic authorities. The systems were consulted briefly, often for only a few seconds, before clinicians returned to their primary workstations to continue analyzing patient data. In this sense, AI tools became one informational element within a broader set of resources that clinicians used while navigating diagnostic uncertainty. Participants frequently emphasized that such practices were pragmatic responses to the cognitive demands of contemporary medical work. Clinical reasoning requires physicians to synthesize diverse forms of information, including imaging data, laboratory results, patient histories, treatment guidelines, and emerging research findings. When institutional systems failed to integrate these elements effectively, clinicians often relied on additional informational resources to bridge the resulting gaps. From this perspective, the use of AI applications represented a continuation of long-standing practices of information seeking within medical work. Doctors have traditionally relied on a wide range of external resources—textbooks, online databases, research articles, and consultations with colleagues—when confronting unfamiliar cases(Timmermans & Berg, 2010 ). AI tools appeared as a new addition to this informational landscape, offering a different way of organizing and exploring medical knowledge. The flexibility of language-model interfaces made them particularly suited to this role. Unlike structured hospital databases, which required users to navigate predefined search categories, AI systems allowed clinicians to formulate questions in natural language. This conversational format made it easier to explore complex or ambiguous clinical situations that did not fit neatly within standardized query structures. One clinician described the contrast between the two types of systems. Hospital software is very structured,” the doctor explained. “You have to search in a specific way. AI interfaces, by contrast, allowed doctors to formulate questions more freely. With AI, you can describe the situation in your own words,” the physician said. “It feels more like discussing a problem than searching a database. For clinicians accustomed to navigating rigid institutional systems, this conversational flexibility could provide a useful complement to existing information infrastructures. What began as an individual strategy for managing uncertainty, however, often extended beyond private consultation. As clinicians shared experiences with colleagues, information obtained through digital searches—including AI-generated responses—sometimes circulated within broader professional networks. Messaging platforms played an important role in this process. Many departments maintained group chats through which clinicians exchanged diagnostic opinions, discussed unusual cases, and shared references to relevant medical literature. These digital communication channels functioned as informal spaces for collaborative reasoning, operating alongside the formal documentation processes required by hospital systems. One participant described a situation in which a senior physician shared a screenshot of an AI-generated explanation within a departmental messaging group. The discussion began when several doctors debated the interpretation of an unusual imaging pattern. “Different people had different ideas,” the participant recalled. After several interpretations had been proposed, the senior physician posted an AI-generated response that outlined a possible explanation for the observed pattern. The professor sent the screenshot and asked everyone what they thought about it. The AI output did not end the discussion. Instead, it became one element within an ongoing conversation among clinicians. Some doctors thought the explanation was reasonable,” the participant explained. “Others said the AI was simplifying the case too much. Rather than functioning as an authoritative diagnosis, the AI-generated text became a conversational artifact around which clinicians debated alternative interpretations. Through this process, the informational output of the AI system was evaluated, critiqued, and contextualized within the collective knowledge of the group. These exchanges illustrate how information obtained through digital tools may circulate through professional networks before becoming integrated into clinical reasoning. Diagnostic interpretation in hospital environments rarely occurs in isolation. Instead, clinicians frequently rely on discussions with colleagues when confronting complex cases. Messaging platforms provide a convenient medium for these conversations, allowing doctors to share images, references, and opinions in real time. AI-generated responses occasionally entered these discussions as additional informational resources. Yet their significance depended entirely on how clinicians interpreted them within the context of professional expertise. The algorithmic output did not replace medical judgment; rather, it became one element within a broader process of collaborative reasoning. Participants also noted that these practices often reflected generational differences in how clinicians engaged with digital technologies. Younger doctors tended to experiment more frequently with AI applications, particularly during demanding training periods when rapid access to explanatory information could help manage cognitive overload. For residents still developing diagnostic confidence, AI tools sometimes served as a way of organizing unfamiliar knowledge. When you are still learning,” one resident explained, “you sometimes need help seeing the structure of a problem. Senior physicians, by contrast, were generally more cautious in their use of AI tools. Many described relying primarily on established medical references and professional experience when interpreting cases. Yet even among senior clinicians, AI-generated information occasionally entered professional discussions through messaging groups or informal conversations. Over time, informational resources discovered by younger doctors could circulate upward through these interactions. Sometimes a junior doctor finds something interesting and shares it with the group,” one participant explained. “Then the senior doctors comment on it or add their own interpretation. Through these exchanges, what begins as an individual strategy for resolving uncertainty may gradually become incorporated into collective discussions about clinical reasoning. These observations suggest that clinicians increasingly operate within a complex informational environment composed of multiple technological and social resources. Institutional platforms provide access to patient records and imaging data, while personal devices offer entry points to external knowledge sources. Professional communication networks allow clinicians to exchange interpretations and evaluate new information collaboratively. Diagnostic reasoning unfolds through the movement between these different environments rather than through reliance on any single system. The practices described above illustrate how clinicians assemble fragments of knowledge drawn from multiple sources in order to interpret complex cases. Rather than depending on a single authoritative platform, doctors combine insights from institutional systems, external digital tools, and professional discussions. Through this ongoing process of searching, comparing, and interpreting information, clinicians gradually construct workable understandings of clinical situations even when institutional infrastructures prove incomplete or inflexible. At the same time, these practices exist alongside persistent structures of professional responsibility. Regardless of how information is assembled during the reasoning process, the final clinical decision must still be attributed to the physician who signs the medical report. Doctors may consult colleagues, databases, or digital tools while interpreting a case, yet the institutional framework of accountability remains firmly individualized.The implications of this tension between distributed informational practices and individualized responsibility are explored in the following discussion section Discussion The findings presented in this study illuminate how clinicians incorporate AI tools into everyday clinical reasoning in contexts where institutional digital infrastructures prove incomplete or inflexible. Rather than entering clinical practice primarily through formally deployed hospital systems, AI tools frequently appear through informal and situational practices that emerge within the everyday realities of clinical work (Blease et al., 2019 ; Shortliffe & Sepúlveda, 2018 ). In the empirical cases described above, clinicians encountered institutional platforms that structured documentation and information retrieval but often failed to support the interpretive flexibility required for complex diagnostic reasoning. When these infrastructures proved insufficient, clinicians expanded their search for information beyond the institutional workstation, turning toward personal devices, external digital resources, and increasingly AI-powered applications (Coiera, 2015 ; Sinsky et al., 2016 ). In these moments, AI did not function as a fully integrated clinical technology but rather as one informational element within a broader assemblage of resources through which clinicians attempted to resolve uncertainty. These findings suggest that the incorporation of AI into clinical reasoning unfolds through a process that differs significantly from the scenarios often described in discussions of medical AI adoption (Jiang et al., 2017 ). Much existing research examines AI primarily as a formally implemented technological system, focusing on algorithmic accuracy, regulatory approval, or institutional deployment (Rajpurkar et al., 2022 ; Reddy et al., 2019 ). From this perspective, AI appears as a discrete technology that enters clinical environments through structured processes of validation and integration. Yet the empirical material presented in this article reveals a different pathway of technological incorporation. Rather than waiting for formal institutional integration, clinicians frequently engage with AI tools through pragmatic experimentation, drawing on consumer applications available through personal devices to supplement existing infrastructures (Amann et al., 2020 ; Cabitza et al., 2017 ). In doing so, they integrate AI into diagnostic reasoning not through official systems but through everyday practices of information seeking and interpretation. A growing body of research on medical AI has focused on the challenges of integrating algorithmic systems into clinical workflows through formal institutional processes (Greenhalgh et al., 2017 ; Sendak et al., 2020 ). Studies in health informatics and medical AI governance often examine questions of validation, regulation, and clinical implementation, emphasizing how AI technologies must undergo extensive evaluation before being incorporated into hospital infrastructures (Reddy et al., 2019 ; Topol, 2019 ). Within this literature, the central analytical problem typically concerns how healthcare institutions can successfully integrate algorithmic systems into established clinical routines. By contrast, the empirical cases examined in this study suggest that the incorporation of AI into clinical reasoning may also occur through pathways that lie largely outside formal institutional implementation. Rather than waiting for AI systems to be officially deployed within hospital infrastructures, clinicians sometimes experiment with readily available digital tools in order to address immediate cognitive demands arising within everyday medical work. To conceptualize this process, the article introduced the notion of cognitive stitching. The concept captures the practical work through which clinicians assemble heterogeneous informational resources in order to sustain diagnostic reasoning under conditions of infrastructural limitation. As the empirical accounts illustrate, clinicians rarely rely on a single technological system when confronting complex cases. Instead, they move between multiple informational environments: hospital information systems provide structured patient data; imaging platforms offer visual evidence; online databases supply reference knowledge; messaging networks facilitate collegial consultation; and AI applications generate provisional explanations or analytical summaries. Diagnostic reasoning emerges through the continual movement between these sources, as clinicians selectively combine fragments of information into coherent interpretations of patient conditions. Understanding this process requires situating AI within the broader ecology of clinical knowledge production rather than treating it as an isolated technological intervention (Abbott, 2014 ; Larson, 2003 ). Clinical reasoning has long been understood as a process that unfolds across multiple informational resources, including diagnostic technologies, professional consultation, and institutional knowledge systems. The findings presented here suggest that clinicians actively assemble these heterogeneous resources through a practical process that this article conceptualizes as cognitive stitching. Medical knowledge is rarely generated solely through the internal cognition of individual practitioners (Good, 1994 ; Mol, 2002 ). Instead, it emerges through the interplay of diagnostic technologies, professional collaboration, and standardized knowledge systems that collectively shape how clinicians interpret clinical information. In contemporary healthcare environments, the number of informational resources available to clinicians has expanded dramatically as digital technologies multiply the tools through which medical knowledge can be accessed and interpreted. Under such conditions, clinicians must actively coordinate these heterogeneous resources in order to sustain diagnostic reasoning—a process captured by the concept of cognitive stitching. From this perspective, the practices described in this study can be understood as part of a broader pattern of technological adaptation within clinical work. Research in science and technology studies has repeatedly shown that technologies rarely determine professional practice through their design alone. Instead, their meaning and function emerge through situated interactions within specific organizational environments (Bruni & Teli, 2007 ; Douglas, 2012 ; Suchman, 2007 ). Technologies may be designed with particular uses in mind, but their practical roles are continually reshaped by the everyday activities of users navigating institutional constraints. Berg’s studies of medical information systems similarly demonstrated how digital infrastructures become embedded within clinical work through processes of local negotiation and adjustment (Berg, 1997 ). The empirical findings presented here extend these insights to the emerging domain of AI-assisted medicine. Rather than entering clinical practice as autonomous diagnostic systems, AI tools become integrated into reasoning processes through the situated improvisations of clinicians responding to gaps within existing infrastructures. The importance of these improvisational practices becomes particularly visible when examining the moments of infrastructural friction described by participants. Institutional hospital systems were designed to structure the flow of clinical information, enforce standardized documentation procedures, and regulate diagnostic workflows. Yet clinicians frequently encountered situations in which these systems proved unable to accommodate the complexity or ambiguity of real clinical cases. Rigid search interfaces, fragmented data architectures, and restrictive query structures often limited the ability of practitioners to explore alternative interpretations of clinical information. In such contexts, the technological infrastructures intended to support clinical reasoning could themselves become sources of cognitive constraint. Rather than abandoning digital tools altogether, clinicians responded by expanding the range of resources they used in the reasoning process. Personal devices, external databases, and AI applications became supplementary tools through which doctors attempted to overcome the limitations of institutional systems. These practices resonate strongly with earlier research on workarounds in healthcare information systems. Studies of hospital technologies have documented how clinicians frequently develop informal strategies to circumvent rigid digital infrastructures that fail to align with the practical demands of clinical work (Ash et al., 2004 ; Ellingsen & Monteiro, 2003 ; Koppel et al., 2008 ). Such workarounds are often interpreted as deviations from intended system use. However, they may also be understood as essential mechanisms through which complex sociotechnical systems remain operational in practice. Rather than simply representing failures of technological design, these adaptations reveal how practitioners actively reshape infrastructures to fit the realities of everyday work. The incorporation of AI tools described in this article can be interpreted as a continuation of this broader pattern of infrastructural improvisation. When institutional systems prove incomplete, clinicians extend their reasoning capacities by incorporating additional digital resources into the diagnostic process. Earlier research on clinical decision-support systems has similarly emphasized how digital tools may assist clinicians by providing structured recommendations within institutional workflows. However, such systems are typically designed as formally integrated components of hospital infrastructures, operating through standardized interfaces connected to electronic health records. The practices observed in this study differ in an important respect. The AI tools used by clinicians were not embedded within official hospital platforms but accessed through personal devices and external applications. As a result, their incorporation into clinical reasoning occurred through informal and situational practices rather than through institutional design. This distinction highlights the importance of examining how emerging digital technologies become integrated into professional work not only through formal implementation but also through the everyday improvisations of practitioners navigating infrastructural constraints. Within this expanding informational landscape, AI tools acquire a particular role. Unlike traditional databases or static reference materials, large language model applications provide conversational interfaces capable of generating structured explanations in response to complex queries. For clinicians confronting unfamiliar clinical situations, such tools offer a rapid way to reorganize fragmented information into provisional interpretive frameworks. Importantly, participants in this study did not describe AI outputs as authoritative diagnoses. Instead, they treated them as tentative suggestions that required further verification through established professional practices. Algorithmic responses were compared with clinical guidelines, medical literature, or the experiential knowledge of colleagues before being incorporated into final diagnostic judgments. In this sense, AI functions less as a replacement for professional expertise than as a cognitive aid embedded within existing interpretive routines. This observation also diverges from a strand of literature that frames medical AI primarily in terms of automation or the potential displacement of clinical expertise. Public and scholarly debates frequently center on whether machine learning systems might eventually outperform physicians in diagnostic tasks (Esteva et al., 2019 ; Topol, 2019 ). While such discussions highlight important technological possibilities, they often assume that AI systems will enter clinical environments as relatively self-contained diagnostic tools. The empirical findings presented here instead suggest that AI is more likely to become embedded within existing interpretive practices in incremental and often informal ways. Rather than replacing clinical reasoning, AI applications become integrated into the broader ecology of knowledge resources through which clinicians organize, interpret, and evaluate medical information(El-Najdawi & Stylianou, 1993 ; Lebovitz et al., 2022 ). This mode of incorporation has important implications for how we understand the role of AI in medical work. Public discussions of medical AI often frame the technology in terms of potential substitution—whether algorithms will eventually replace human clinicians in diagnostic tasks. The empirical findings presented here suggest a different dynamic. Rather than displacing clinical reasoning, AI tools are woven into existing processes through which clinicians interpret complex information. They become additional informational elements that clinicians incorporate into the ongoing process of cognitive stitching, through which fragments of medical knowledge drawn from different sources are combined into workable diagnostic interpretations. The result is not a simple transfer of cognitive authority from human practitioners to algorithmic systems, but a transformation in how clinicians perform cognitive stitching, assembling informational fragments from institutional systems, professional networks, and digital tools into coherent clinical interpretations. At the same time, the emergence of cognitive stitching reveals an important tension within the institutional organization of clinical practice. While diagnostic reasoning increasingly involves distributed informational resources, the structures governing professional accountability remain firmly individualized. Medical decisions continue to be formally attributed to the physician who signs the clinical report, regardless of the technological resources involved in generating that decision. This arrangement reflects the longstanding professional structure of medicine, in which clinical authority and responsibility are closely tied to the individual practitioner. Yet the practices described in this study indicate that the cognitive processes underlying those decisions may involve a far more complex network of informational actors. The practices described above also reveal an important institutional tension surrounding cognitive stitching. While clinicians assemble diagnostic understanding by drawing on multiple informational resources—including institutional systems, digital tools, and professional networks—the formal structures of medical responsibility remain firmly individualized. Technologies often redistribute tasks, information flows, and cognitive labor without fundamentally altering the institutional structures through which responsibility is assigned (Suchman, 2007 ). In clinical environments, AI systems may assist with data interpretation, hypothesis generation, or information retrieval, but they do not assume accountability for diagnostic outcomes. Instead, clinicians remain responsible for evaluating algorithmic suggestions and integrating them into professional judgment. As a result, the incorporation of AI into clinical reasoning does not eliminate the central role of human expertise but rather reshapes the informational landscape within which that expertise operates. Importantly, participants in this study did not generally describe this situation as an ethical crisis. For most clinicians, consulting AI tools was understood as a pragmatic extension of existing practices of information seeking. Doctors have long relied on a wide range of external resources—textbooks, online databases, professional networks—when confronting unfamiliar cases. From this perspective, AI appears as a new addition to an already diverse informational ecosystem. What distinguishes AI from earlier resources is not simply its content but the speed and flexibility with which it can generate explanatory structures. By quickly organizing dispersed information into coherent narratives, AI applications provide clinicians with a new way of navigating moments of cognitive uncertainty. Yet the informal nature of these practices also contributes to a degree of institutional ambiguity. AI tools accessed through personal devices often remain outside formal hospital governance structures. Their use is rarely documented within official clinical systems, and their role in shaping diagnostic reasoning remains largely invisible within institutional records. Consequently, AI becomes incorporated into clinical cognition without necessarily being recognized as part of the official technological infrastructure of healthcare institutions. This gap between practical use and institutional recognition illustrates how technological change in professional settings frequently occurs through incremental adaptations rather than formal implementation. Seen in this light, the incorporation of AI into clinical reasoning should be understood less as a technological revolution and more as a gradual transformation in how clinicians perform cognitive stitching, integrating new digital tools into existing practices of medical interpretation. Clinicians continually negotiate the boundaries between institutional systems, professional knowledge, and emerging digital tools in order to sustain the cognitive work of medicine. AI applications become meaningful within this process not because of their technical capabilities alone but because of how they are integrated into the everyday practices through which clinicians interpret clinical information. By examining these practices ethnographically, this study contributes to broader discussions about the social life of AI in professional contexts. Rather than treating AI as an external force reshaping medical practice from the outside, the analysis highlights the central role of practitioners in actively incorporating technologies into their work. The concept of cognitive stitching offers one way of understanding how clinicians assemble distributed informational resources in order to bridge the gaps left by institutional infrastructures. Through these small but consequential acts of coordination, emerging technologies become woven into the ongoing routines of clinical reasoning. Understanding these processes is essential for anticipating how AI will continue to evolve within healthcare environments. The future impact of AI in medicine will not depend solely on advances in algorithmic performance or regulatory frameworks. It will also depend on how clinicians integrate these technologies into the situated practices of everyday clinical work. By focusing on these practices, the analysis presented here shifts attention away from abstract debates about technological transformation and toward the practical realities through which AI becomes embedded within the sociotechnical fabric of contemporary medicine Conclusion This article has examined how clinicians incorporate AI tools into everyday clinical reasoning in environments where institutional digital infrastructures remain incomplete or inflexible. Rather than focusing on the formal deployment of medical AI systems, the analysis has traced the situated practices through which clinicians navigate gaps within hospital information systems. Ethnographic accounts from Chinese public hospitals show that doctors frequently rely on a heterogeneous set of informational resources—including institutional platforms, personal mobile devices, professional communication networks, and increasingly AI applications—in order to interpret complex clinical cases. Through these practices, clinicians assemble fragmented knowledge into workable diagnostic interpretations, a process described in this article as cognitive stitching. By foregrounding these practices, the study contributes to a growing body of research emphasizing that technologies rarely enter professional environments as fully stabilized systems. Instead, they are incorporated through situated forms of adaptation shaped by the practical conditions of work (Leonardi, 2011 ; Suchman, 2007 ). In the clinical settings examined here, AI tools do not replace medical expertise nor function as autonomous diagnostic authorities. Rather, they become embedded within the cognitive routines through which clinicians organize and interpret medical information. When institutional infrastructures prove insufficient for resolving complex cases, clinicians extend their reasoning by drawing on additional informational resources available through the wider digital environment. At the same time, the incorporation of AI into clinical reasoning unfolds within institutional structures that continue to attribute responsibility primarily to individual physicians. As the analysis has shown, clinicians may consult AI tools, online databases, or colleagues while reasoning through difficult cases, yet the final diagnostic decision remains formally attached to the doctor who signs the medical record. This produces a subtle but significant tension between the distributed character of contemporary clinical cognition and the individualized framework of professional accountability. While informational work becomes increasingly mediated by digital tools, responsibility remains firmly anchored in traditional institutional arrangements. Although the empirical material analyzed in this article is drawn from Chinese hospitals, it is useful to briefly reflect on how these dynamics appear differently in other healthcare contexts. Drawing on observations from a separate Norwegian research project, the following comparison is offered not as a formal comparative analysis but as a discussion point highlighting how institutional conditions may shape the incorporation of AI in different ways. In Norway, the integration of medical AI tends to occur through more formalized regulatory pathways, including institutional approval processes and certified clinical software infrastructures. Algorithmic tools are incorporated into hospital systems through clearly defined governance frameworks, making their role within clinical decision-making more visible and institutionally recognized. By contrast, the practices described in Chinese hospitals illustrate a more informal pathway of technological incorporation, in which clinicians draw on widely accessible AI tools through personal devices and professional networks. These contrasting configurations should not be understood simply as differences in technological development but as variations in the sociotechnical organization of medical practice. The Norwegian case demonstrates how AI can become integrated through formal institutional channels, while the Chinese context reveals how clinicians adapt technologies pragmatically when institutional infrastructures lag behind technological possibilities. In both situations, however, the incorporation of AI ultimately depends on the everyday practices of clinicians who must interpret, evaluate, and contextualize algorithmic information within the realities of patient care This brief comparison should therefore be understood as an interpretive reflection rather than a systematic cross-national analysis, pointing to how different institutional environments may shape the ways clinicians incorporate AI into everyday reasoning. Recognizing these practices helps shift discussions of medical AI away from deterministic narratives about technological transformation. Rather than viewing AI as a system that either replaces or augments clinicians in a straightforward manner, this study suggests that its integration is mediated by the situated practices through which professionals manage uncertainty, interpret digital outputs, and coordinate multiple informational resources. AI becomes meaningful not as an abstract technological capability but as part of the everyday work through which clinicians sustain diagnostic reasoning in complex institutional environments. The concept of cognitive stitching offers a way to understand how clinicians sustain diagnostic reasoning in environments where institutional infrastructures remain incomplete. Rather than relying on a single authoritative technological system, practitioners assemble fragments of knowledge drawn from multiple informational environments. Through this ongoing process of stitching together institutional data, professional experience, and algorithmic suggestions, clinicians maintain the interpretive work through which medical decisions are produced. Understanding these practices highlights that the future of AI in healthcare will depend not only on algorithmic capabilities but also on how clinicians incorporate these technologies into the everyday practices through which medical knowledge is constructed. Declarations CONFLICT OF INTEREST STATEMENT The author declares no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. ETHICS STATEMENT All procedures performed in this study involving human participants were conducted in a hospital setting and were in accordance with the ethical standards of the relevant institutional research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Ethical approval for this research was obtained from the relevant institutional ethics review board. Informed consent was obtained from all participants prior to their participation. Informed Consent Before the interviews, participants were provided with an information sheet outlining the purpose of the study and their rights. They were informed that interviews may be audio-recorded, anonymized, and used for research purposes, and that participation was voluntary. Informed consent was obtained from all participants prior to participation, and all data were handled confidentially. DATA Availability Statement The data underlying this study are not publicly available due to ethical and privacy restrictions. The research was conducted in a hospital setting and received approval from the Ethics Review Board of the relevant institutional research committee, in compliance with the data management regulations of the European Research Council (ERC) and the University of Amsterdam. Due to the nature of the study, which involves qualitative interviews and observational materials collected in a clinical environment, even de-identified data may carry a risk of participant re-identification. As such, sharing raw or processed data through publicly accessible repositories is not permitted under the approved ethical protocols. All data are securely stored in the University of Amsterdam’s protected research environment for a period of ten years in accordance with institutional guidelines. De-identified data may be made available from the author upon reasonable request and subject to appropriate ethical approvals and data access agreements. References Abbott A (2014) The system of professions: An essay on the division of expert labor. University of Chicago Press Amann J, Blasimme A, Vayena E, Frey D, Madai VI, Consortium PQ (2020) Explainability for artificial intelligence in healthcare: a multidisciplinary perspective. BMC Med Inf Decis Mak 20(1):310 Ash JS, Berg M, Coiera E (2004) Some unintended consequences of information technology in health care: the nature of patient care information system-related errors. J Am Med Inform Assoc 11(2):104–112 Babic B, Gerke S, Evgeniou T, Cohen IG (2021) Beware explanations from AI in health care. Science 373(6552):284–286 Berg M (1997) Of forms, containers, and the electronic medical record: some tools for a sociology of the formal. Sci Technol Hum Values 22(4):403–433 Blease C, Kaptchuk TJ, Bernstein MH, Mandl KD, Halamka JD, DesRoches CM (2019) Artificial intelligence and the future of primary care: exploratory qualitative study of UK general practitioners’ views. J Med Internet Res 21(3):e12802 Bowker GC, Star SL (2000) Sorting things out: Classification and its consequences. MIT Press Bruni A, Teli M (2007) Reassembling the social—An introduction to actor network theory. Manage Learn 38(1):121–125 Cabitza F, Rasoini R, Gensini GF (2017) Unintended consequences of machine learning in medicine. JAMA 318(6):517–518 Chang CK, Chiari L, Cao Y, Jin H, Mokhtari M, Aloulou H (2016) Inclusive Smart Cities and Digital Health: 14th International Conference on Smart Homes and Health Telematics, ICOST 2016, Wuhan, China, May 25–27, 2016. Proceedings (Vol. 9677). Springer Coiera E (2015) Guide to health informatics. CRC Deloitte (2021) Internet Hospitals in China: The new step into digital healthcare Douglas DG (2012) The social construction of technological systems, anniversary edition: New directions in the sociology and history of technology. MIT Press El-Najdawi M, Stylianou AC (1993) Expert support systems: integrating AI technologies. Commun ACM 36(12):55–ff Ellingsen G, Monteiro E (2003) A patchwork planet integration and cooperation in hospitals. Comput Supported Coop Work (CSCW) 12(1):71–95 Esteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, Cui C, Corrado G, Thrun S, Dean J (2019) A guide to deep learning in healthcare. Nat Med 25(1):24–29 Gao J, Zhang P (2021) China's Public Health Policies in Response to COVID-19: From an Authoritarian Perspective. Front Public Health 9:756677. https://doi.org/10.3389/fpubh.2021.756677 Good BJ (1994) Medicine, rationality and experience: an anthropological perspective. Cambridge University Press Greenhalgh T, Wherton J, Papoutsi C, Lynch J, Hughes G, Hinder S, Fahy N, Procter R, Shaw S (2017) Beyond adoption: a new framework for theorizing and evaluating nonadoption, abandonment, and challenges to the scale-up, spread, and sustainability of health and care technologies. J Med Internet Res 19(11):e8775 Guo Z, Cugurullo F (2023) AI doctors or AI for doctors? Augmenting urban healthcare services through artificial intelligence. Artificial Intelligence and the City. Routledge, pp 307–321 He J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K (2019) The practical implementation of artificial intelligence technologies in medicine. Nat Med 25(1):30–36 Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, present and future. Stroke and vascular neurology , 2 (4) Koppel R, Wetterneck T, Telles JL, Karsh B-T (2008) Workarounds to barcode medication administration systems: their occurrences, causes, and threats to patient safety. J Am Med Inform Assoc 15(4):408–423 Larson MS (2003) Professionalism: The third logic. Perspect Biol Med 46(3):458–462 Laurie G (2017) Liminality and the limits of law in health research regulation: what are we missing in the spaces in-between? Med Law Rev 25(1):47–72 Lebovitz S, Lifshitz-Assaf H, Levina N (2022) To engage or not to engage with AI for critical judgments: How professionals deal with opacity when using AI for medical diagnosis. Organ Sci 33(1):126–148 Leonardi PM (2011) When flexible routines meet flexible technologies: Affordance, constraint, and the imbrication of human and material agencies1. MIS Q 35(1):147–167 Liang H, Tsui BY, Ni H, Valentim CC, Baxter SL, Liu G, Cai W, Kermany DS, Sun X, Chen J (2019) Evaluation and accurate diagnoses of pediatric diseases using artificial intelligence. Nat Med 25(3):433–438 Lipset S (2017) Social organization of medical work. Routledge Liu Y, Rao K, Wu J, Gakidou E (2008) China's health system performance. Lancet 372(9653):1914–1923 Mol A (2002) The body multiple: Ontology in medical practice. Duke University Press Obermeyer Z, Emanuel EJ (2016) Predicting the future—big data, machine learning, and clinical medicine. N Engl J Med 375(13):1216 Orlikowski WJ (2000) Using technology and constituting structures: A practice lens for studying technology in organizations. Organ Sci 11(4):404–428 Rajpurkar P, Chen E, Banerjee O, Topol EJ (2022) AI in health and medicine. Nat Med 28(1):31–38 Reddy S, Fox J, Purohit MP (2019) Artificial intelligence-enabled healthcare delivery. J R Soc Med 112(1):22–28 Sendak MP, D’Arcy J, Kashyap S, Gao M, Nichols M, Corey K, Ratliff W, Balu S (2020) A path for translation of machine learning products into healthcare delivery. EMJ Innov 10:19–00172 Shanafelt TD, Dyrbye LN, Sinsky C, Hasan O, Satele D, Sloan J, West CP (2016) Relationship between clerical burden and characteristics of the electronic environment with physician burnout and professional satisfaction. Mayo clinic proceedings Shortliffe EH, Sepúlveda MJ (2018) Clinical decision support in the era of artificial intelligence. JAMA 320(21):2199–2200 Sinsky C, Colligan L, Li L, Prgomet M, Reynolds S, Goeders L, Westbrook J, Tutty M, Blike G (2016) Allocation of physician time in ambulatory practice: a time and motion study in 4 specialties. Ann Intern Med 165(11):753–760 Star SL, Ruhleder K (2010) Steps toward an Ecology of Infrastructure. Revue d'anthropologie des connaissances 41(1):114–161 Suchman LA (2007) Human-machine reconfigurations: Plans and situated actions. Cambridge University Press Timmermans S, Berg M (2010) The gold standard: the challenge of evidence-based medicine. Temple University Topol E (2019) Deep medicine: how artificial intelligence can make healthcare human again . Hachette UK Wachter R (2017) Digital Doctor. Hope, Hype and Harm at Dawn. McGraw-Hill, New York Yu KH, Beam AL, Kohane IS (2018) Artificial intelligence in healthcare. Nat Biomed Eng 2(10):719–731. https://doi.org/10.1038/s41551-018-0305-z Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 23 Apr, 2026 Reviewers agreed at journal 22 Apr, 2026 Reviewers invited by journal 21 Apr, 2026 Editor assigned by journal 14 Apr, 2026 Submission checks completed at journal 06 Apr, 2026 First submitted to journal 06 Apr, 2026 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-9095240","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":628305382,"identity":"6d841837-fcda-4e86-a96e-c42b262e928c","order_by":0,"name":"Zongtian Guo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzklEQVRIiWNgGAWjYDADNgYGxgdgFjuROiSAWpgNwExmYrWALJIgSot8RPLhDx931NbxsZ89Vs3bZsdgTkiL4Y20BMOZZ45LsPHkpd3mbUtmsGwmpGVGjkEyb9sxoF9yzG7znGFmMDhMhJbDYC38b8yKec7UE9YiL5Fj2MzbViPBJpFjxsxTcZiwFgOeZ8mMM9sOSLZJvDGWnFNxnIewLe2gEGur45fvzzH88MagWs7geAMBWw6AKYTJPATsANoCMbKOoMJRMApGwSgYwQAA4TA6VS5aDBEAAAAASUVORK5CYII=","orcid":"","institution":"University of Amsterdam","correspondingAuthor":true,"prefix":"","firstName":"Zongtian","middleName":"","lastName":"Guo","suffix":""}],"badges":[],"createdAt":"2026-03-11 13:53:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9095240/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9095240/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108139684,"identity":"bdb4e5fb-535e-426d-8141-85c71ae43f30","added_by":"auto","created_at":"2026-04-29 18:39:48","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":364531,"visible":true,"origin":"","legend":"\u003cp\u003eRadiology reading room in a tertiary hospital. Clinicians work with institutional imaging systems.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9095240/v1/5e7854a9d3fc36e5927b92a2.png"},{"id":108182931,"identity":"2f4f814d-f177-4197-ae31-87a1ff41b282","added_by":"auto","created_at":"2026-04-30 08:59:41","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":641893,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9095240/v1/503361c5-c96e-46d8-a458-fdf266b5e0f3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Cognitive Stitching: Informal AI Integration in Everyday Clinical Reasoning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eArtificial intelligence (AI) has rapidly entered contemporary medical discourse. Across policy reports, clinical journals, and technology media, AI is frequently portrayed as a transformative force capable of reshaping the future of healthcare. Machine learning systems promise earlier diagnoses, more accurate image interpretation, and the ability to analyze vast volumes of biomedical data beyond the limits of human cognition (Topol, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Yu et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In radiology, pathology, and oncology in particular, algorithmic systems are often presented as tools that may fundamentally reconfigure diagnostic work and professional expertise (Rajpurkar et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Yet alongside these optimistic projections runs an equally visible current of skepticism. Critics warn that algorithmic medicine may introduce new forms of bias, erode professional autonomy, or create opaque decision-making processes that are difficult to audit or contest (Babic et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Obermeyer \u0026amp; Emanuel, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). As a result, discussions about AI in healthcare often oscillate between technological optimism and technological anxiety.\u003c/p\u003e \u003cp\u003eDespite these debates, much of the existing literature shares a common analytical focus: the evaluation and implementation of AI systems as formal technologies. A large body of research examines the accuracy of diagnostic algorithms, comparing their performance with that of human clinicians (Esteva et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Guo \u0026amp; Cugurullo, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Rajpurkar et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Other studies explore the institutional integration of AI within healthcare infrastructures, analyzing regulatory frameworks, clinical validation processes, and the organizational conditions required for successful deployment (Reddy et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). While these approaches provide important insights into the capabilities and governance of medical AI, they tend to treat algorithms as discrete technological systems whose impact can be assessed through performance metrics or institutional adoption. What remains less visible in these discussions is the everyday life of AI in clinical practice. Hospitals are complex sociotechnical environments in which multiple technologies coexist: electronic health records, diagnostic imaging platforms, clinical guidelines, communication systems, and an expanding range of algorithmic tools (Bowker \u0026amp; Star, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2000\u003c/span\u003e; Star \u0026amp; Ruhleder, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). Within such environments, technologies rarely operate as seamlessly integrated systems (Orlikowski, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2000\u003c/span\u003e). Instead, they are continuously adapted, bypassed, or supplemented by practitioners responding to the practical demands of patient care. Research in STS and medical sociology has long emphasized that technologies acquire meaning and function through situated practice rather than through design alone (Berg, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1997\u003c/span\u003e; Suchman, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Timmermans \u0026amp; Berg, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). From this perspective, the key question is not simply whether AI systems work, but how they are actually incorporated into the routines and reasoning processes of clinicians.\u003c/p\u003e \u003cp\u003eRecent studies of healthcare technologies have highlighted the prevalence of informal practices and workarounds in digital clinical environments (Laurie, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). When institutional information systems prove inflexible or poorly aligned with clinical workflows, practitioners often develop pragmatic strategies to accomplish their tasks (Ash et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Ellingsen \u0026amp; Monteiro, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Koppel et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). These practices can include bypassing official software functions, relying on personal devices, or informally sharing knowledge through professional networks. Such forms of technological improvisation are not merely marginal deviations from formal infrastructures. Rather, they constitute an important part of how complex healthcare systems continue to function in practice.This article builds on these insights by examining how clinicians incorporate emerging AI tools into everyday diagnostic reasoning through informal practices. Drawing on ethnographic fieldwork conducted in Chinese public hospitals, the study investigates how clinicians navigate the gap between institutional digital infrastructures and the practical demands of clinical work.\u003c/p\u003e \u003cp\u003eOver the past decade, Chinese hospitals have invested heavily in digital health technologies, including hospital information systems, picture archiving and communication systems, and specialized medical AI applications designed for tasks such as image analysis or treatment planning (He et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Liang et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Yet clinicians frequently encounter limitations in these institutional platforms, including rigid software interfaces, fragmented data systems, and time-consuming workflows (Chang et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Deloitte, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Gao \u0026amp; Zhang, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Liu et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). These frictions become particularly visible in high-pressure contexts such as night shifts or emergency consultations, where clinicians must make rapid decisions with limited cognitive and informational resources. In these moments, AI does not always appear through officially deployed hospital systems. Instead, clinicians often turn to consumer-grade AI tools accessed through their personal smartphones. Applications based on large language models, for example, are sometimes consulted to clarify unfamiliar drug interactions, generate possible diagnostic hypotheses, or organize complex clinical information. These interactions typically occur quietly and informally. Junior doctors describe briefly checking an AI application when confronted with unfamiliar cases, while senior clinicians occasionally circulate screenshots of AI-generated explanations in professional messaging groups to stimulate discussion among colleagues. Although such practices remain outside formal institutional protocols, they increasingly form part of the everyday repertoire through which clinicians manage cognitive uncertainty in demanding clinical environments.\u003c/p\u003e \u003cp\u003eTo conceptualize these dynamics, this article introduces the notion of cognitive stitching. The concept refers to the practical process through which clinicians assemble multiple technological resources\u0026mdash;official hospital systems, online databases, personal devices, and AI applications\u0026mdash;to bridge gaps in institutional infrastructures. Rather than functioning as autonomous decision-makers, AI tools in these contexts become one element within a broader patchwork of clinical reasoning. Clinicians selectively draw on these tools to extend their capacity to process information, compare possible interpretations, or structure complex diagnostic thinking. In this sense, AI becomes woven into clinical cognition not through formal integration alone but through everyday acts of technological improvisation. By focusing on these practices, the article addresses a gap in existing research on medical AI. While much scholarship concentrates on algorithmic accuracy or regulatory frameworks, relatively little attention has been paid to how clinicians incorporate AI into the lived realities of clinical work. Understanding these practices is crucial because technologies rarely reshape professional domains through formal implementation alone. Instead, they are gradually integrated through the mundane adaptations of practitioners navigating institutional constraints.\u003c/p\u003e \u003cp\u003eThe central research question guiding this study is therefore: \u003cb\u003ehow do clinicians incorporate AI tools into everyday clinical reasoning when institutional infrastructures prove incomplete or inflexible?\u003c/b\u003e By examining this question ethnographically, the article highlights how AI becomes embedded in clinical practice through informal and often improvised arrangements.\u003c/p\u003e \u003cp\u003eThe article proceeds as follows. The next section outlines the methodological approach and fieldwork through which these dynamics were documented. The empirical analysis then examines how institutional frictions within hospital digital infrastructures create the conditions for cognitive stitching. Through detailed ethnographic accounts, the paper shows how clinicians move between institutional platforms and personal AI tools while managing diagnostic uncertainty in everyday clinical work. The subsequent discussion explores the professional and ethical tensions that emerge when algorithmic assistance becomes informally embedded within clinical reasoning. Finally, the conclusion briefly reflects on these dynamics through a comparative perspective, contrasting informal forms of AI incorporation with more formally regulated models of medical AI adoption.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThis article draws on ethnographic fieldwork conducted in public hospitals in China between 2024 and 2025. The broader research project explored how emerging digital technologies are reshaping clinical work and professional knowledge in contemporary healthcare settings. Rather than evaluating the technical performance of medical AI systems, the fieldwork focused on how clinicians encounter and incorporate digital tools within the everyday routines of diagnosis, consultation, and decision-making.\u003c/p\u003e \u003cp\u003eThe research was conducted primarily in large public hospitals located in major Chinese cities. These institutions represent the dominant organizational form of healthcare delivery in China and are characterized by extremely high patient volumes, complex administrative structures, and rapidly expanding digital infrastructures. Over the past decade, Chinese hospitals have invested heavily in information technologies, including hospital information systems, electronic medical records, picture archiving and communication systems, and a growing number of algorithmic applications designed to assist with tasks such as medical imaging interpretation or treatment planning. On paper, these infrastructures suggest a highly digitalized clinical environment in which information flows smoothly between different technological platforms.\u003c/p\u003e \u003cp\u003eIn practice, however, the everyday experience of clinicians often reveals a more fragmented landscape. Doctors working in tertiary hospitals frequently face intense workloads and time pressure, with outpatient consultations sometimes lasting only a few minutes. Diagnostic departments process large numbers of imaging scans each day, requiring physicians to review hundreds of images within limited time frames (Lipset, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Within such environments, digital technologies do not function as seamless infrastructures but as a collection of systems that clinicians must navigate while maintaining the pace of clinical work. These conditions formed the broader context within which the fieldwork was conducted.\u003c/p\u003e \u003cp\u003eThe empirical material presented in this article is based primarily on qualitative interviews with clinicians working across multiple departments. In total, thirty-four interviews were conducted with medical professionals, including diagnostic radiologists, medical physicists, clinicians from other specialties, and specialist nurses involved in technologically mediated care practices. Participants ranged from junior residents in the early stages of their careers to senior physicians with more than two decades of clinical experience. This diversity allowed the research to capture how clinicians with different levels of professional authority and technological familiarity encounter AI in their everyday work. Interviews typically lasted between twenty and sixty minutes and were conducted in locations such as departmental offices, hospital meeting rooms, or through remote communication when in-person meetings were not possible. Conversations were semi-structured, allowing participants to describe their experiences with digital technologies in their own terms. Rather than focusing exclusively on officially deployed AI systems, discussions often moved toward broader questions about how clinicians search for information, manage diagnostic uncertainty, and cope with moments of cognitive overload during demanding shifts. Participants were frequently encouraged to recall concrete situations in which technological tools played a role in shaping their diagnostic reasoning.\u003c/p\u003e \u003cp\u003eAlongside these interviews, the research also incorporated elements of ethnographic observation within clinical environments. Time was spent in workspaces such as radiology reading rooms and departmental offices, where clinicians interact continuously with digital systems while performing diagnostic work. Observing these environments provided insight into how multiple technological interfaces coexist within clinical practice. Hospital workstations displaying imaging data often operate alongside personal devices, online medical databases, and messaging platforms through which clinicians communicate with colleagues. These observations helped situate interview accounts within the material environments in which clinical reasoning takes place.\u003c/p\u003e \u003cp\u003eAll participants were informed about the purpose of the research and agreed to participate on the condition that their identities and institutional affiliations would remain anonymous. To protect confidentiality, identifying details such as hospital names and personal information have been removed from the material presented in this article.\u003c/p\u003e \u003cp\u003eThe analysis of the material followed an iterative qualitative process informed by ethnographic approaches to technology and work. Interview transcripts and field notes were first reviewed to identify recurring descriptions of how clinicians rely on digital tools while navigating diagnostic uncertainty. Particular attention was paid to moments in which participants described moving between different technological resources while making clinical decisions. As the analysis progressed, it became increasingly clear that clinicians rarely relied on a single technological system. Instead, they frequently assembled information from multiple platforms, combining institutional infrastructures with external digital resources. To make this analytical process more systematic, the material was coded using an iterative qualitative coding framework informed by ethnographic research on technology in professional practice {Charmaz, 2014 #1627}. In the first stage, interview transcripts and field notes were open coded to identify recurring descriptions of how clinicians interacted with digital systems during diagnostic work. These initial codes captured themes such as information search practices, moments of diagnostic uncertainty, interactions with hospital information systems, and the use of external digital resources. In a second stage of analysis, related codes were grouped into broader analytical categories focusing on the movement between institutional infrastructures and external informational tools. Particular attention was paid to episodes in which clinicians described shifting from hospital workstations to personal devices while attempting to interpret complex cases. Observational field notes were used to contextualize these accounts by documenting how clinicians navigated multiple technological interfaces within clinical workspaces, including imaging systems, electronic health records, and personal mobile devices. Through repeated comparison of interview accounts and observational material, these patterns were gradually synthesized into the analytical concept developed in this article as cognitive stitching.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eFinding\u003c/h2\u003e \u003cp\u003eThe following section examines how clinicians navigate the gaps and tensions that emerge within hospital digital infrastructures during everyday clinical work. Rather than treating artificial intelligence as a formally implemented technology embedded within institutional systems, the analysis focuses on the practical conditions under which clinicians turn to alternative informational resources while interpreting complex medical cases. In highly digitalized hospital environments, information systems are designed to structure the flow of clinical data, guide diagnostic procedures, and standardize decision-making processes. These infrastructures promise a form of technological order in which patient information is centralized, searchable, and accessible through standardized interfaces. Yet the everyday experience of working within these systems often appears far more complicated. As clinicians repeatedly emphasized during fieldwork, the systems intended to simplify clinical reasoning sometimes generate new forms of friction that shape how doctors search for information and interpret clinical situations.\u003c/p\u003e \u003cp\u003eIn interviews, clinicians frequently described hospital information systems as simultaneously indispensable and frustrating. On the one hand, these platforms form the backbone of contemporary hospital operations. They store patient histories, display laboratory results, organize imaging data, and provide the structured templates through which clinical documentation must be completed. Without them, it would be nearly impossible to manage the enormous volume of patient information circulating through large tertiary hospitals. At the same time, however, doctors often described these systems as rigid environments that constrained the exploratory dimensions of medical reasoning. While the software functioned well for routine tasks such as retrieving patient records or reviewing imaging results, it was less effective when clinicians attempted to explore unfamiliar diagnostic possibilities or analyze complex relationships between symptoms, medications, and clinical findings. Several participants emphasized that the official platforms were designed primarily for documentation rather than interpretation. The systems organized information according to predefined categories and standardized procedures, reflecting administrative priorities of data management and regulatory compliance. Yet clinical reasoning frequently requires a more flexible mode of investigation in which physicians compare different sources of information, test alternative hypotheses, and interpret ambiguous patterns. When the structured logic of the software collided with these exploratory needs, clinicians often experienced the system less as an aid to reasoning than as a technological constraint.\u003c/p\u003e \u003cp\u003eOne physician described repeatedly encountering automated error messages while attempting to search for drug interaction information through the hospital\u0026rsquo;s internal database. The system was designed to identify known interactions between medications, but its query structure imposed strict limits on how searches could be performed.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSometimes you try to look up a combination of drugs and the system simply blocks the request,\u0026rdquo; the physician explained. \u0026ldquo;It just says \u0026lsquo;Forbidden\u0026rsquo; and stops there.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe doctor described the experience as particularly frustrating when working under time pressure.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eYou\u0026rsquo;re trying to understand the situation quickly, but the system refuses to process the query. At that moment you feel like the technology is working against you.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eAlthough such restrictions were intended to prevent incorrect inputs or unauthorized searches, clinicians often experienced them as interruptions in the reasoning process. Instead of clarifying uncertainty, the system sometimes halted the search at precisely the moment when additional information was needed. Another physician described similar limitations when analyzing complex medication combinations. While the official database could easily identify standard interactions between two drugs, it struggled to process cases involving multiple medications with overlapping pharmacological effects.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe system gives you a few standard warnings,\u0026rdquo; the doctor explained, \u0026ldquo;but when the case becomes complicated, it doesn\u0026rsquo;t really help you understand the relationships.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe physician described situations in which the system produced fragments of information without offering a coherent explanation.\u003c/p\u003e \u003cp\u003e\u0026ldquo;It shows you several alerts, but then you still have to figure out what they actually mean together.\u0026rdquo;\u003c/p\u003e \u003cp\u003eIn these situations, the system functioned less as an interpretive resource than as a collection of partial signals that required further reasoning. These experiences were not isolated incidents but recurring features of everyday clinical work. In large public hospitals, clinicians are expected to process large volumes of patient data while operating under considerable time pressure (Shanafelt et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Wachter, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Diagnostic departments often handle hundreds of imaging cases each day, while outpatient clinics may see dozens of patients during a single shift. Within such environments, digital infrastructures are intended to support rapid access to information and streamline clinical workflows. Yet the same infrastructures may become sources of friction when their structured logic fails to accommodate the complexity of real clinical situations.\u003c/p\u003e \u003cp\u003eThese tensions became particularly visible during night shifts. Several younger doctors described the experience of working alone in hospital departments after senior physicians had left for the day. During these shifts, clinicians were responsible for evaluating new cases while simultaneously monitoring ongoing patients and completing documentation tasks. Although institutional systems remained accessible, the absence of immediate consultation with colleagues intensified the pressure to resolve diagnostic uncertainties independently. One resident recalled a particularly challenging shift involving a patient whose symptoms did not correspond neatly with established diagnostic categories.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eIt was already very late,\u0026rdquo; the doctor explained. \u0026ldquo;You\u0026rsquo;ve been reviewing images for hours, and suddenly you encounter a case that doesn\u0026rsquo;t fit any familiar pattern.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe physician attempted to search the hospital system for guidance but found the available information insufficient.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe system gave some basic references, but they didn\u0026rsquo;t really address the specific situation.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe resident described the cognitive fatigue that often accompanies prolonged diagnostic work.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSometimes your brain just stops. You read the information again and again, but it still doesn\u0026rsquo;t connect.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eMoments like these illustrate how institutional infrastructures can generate what might be described as cognitive gaps. Hospital information systems are designed to standardize knowledge through predefined categories and procedural logic. Clinical reasoning, however, frequently requires the ability to move beyond these categories, combining fragments of information in order to interpret complex cases. When institutional platforms fail to support this exploratory process, clinicians must search for alternative ways of assembling the knowledge they need. It is within these moments of infrastructural limitation that a different pattern of technological practice begins to appear. Rather than abandoning digital tools altogether, clinicians expand the range of informational resources they draw upon. The search for answers gradually extends beyond the institutional workstation toward a broader digital landscape that includes external databases, online medical references, and increasingly, resources accessible through personal mobile devices.\u003c/p\u003e \u003cp\u003eUnderstanding this shift requires recognizing that hospital technologies rarely function as unified infrastructures. Instead, clinicians operate within a heterogeneous technological environment composed of multiple overlapping systems. Institutional platforms structure official documentation and imaging analysis, while informal channels of information seeking operate alongside them. It is through the movement between these different environments that clinicians begin to assemble workable interpretations of complex clinical situations.The material layout of clinical workspaces often reflects this layered technological environment. In radiology reading rooms where part of the fieldwork was conducted, clinicians worked at stations equipped with several large monitors displaying CT and MRI images. The central screens were connected to the hospital\u0026rsquo;s imaging system, allowing doctors to scroll through hundreds of image slices while simultaneously reviewing patient records. Yet beside the keyboard, many clinicians kept their personal smartphones within easy reach, illustrating the coexistence of institutional and informal informational resources within everyday clinical work (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eDuring difficult cases, it was common to see a doctor lean back from the workstation, briefly pick up their phone, and search for additional information while continuing to examine patient images on the screen. These movements between devices were usually quick and almost routine, lasting only a few seconds before clinicians returned their attention to the diagnostic workstation.\u003c/p\u003e \u003cp\u003eSeveral clinicians described this movement between institutional and personal technologies as an almost habitual response to uncertainty.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eYou always start with the hospital system,\u0026rdquo; one physician explained. \u0026ldquo;But if the information there isn\u0026rsquo;t enough, you naturally start looking elsewhere.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eFor many participants, this \u0026ldquo;elsewhere\u0026rdquo; included external digital resources that could provide additional perspectives on unfamiliar clinical situations. Personal smartphones offered immediate access to medical reference websites, research articles, and communication platforms through which clinicians could consult colleagues. One physician described this shift as a practical adaptation to the limitations of institutional infrastructures.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe hospital system is good for retrieving patient data,\u0026rdquo; the doctor explained. \u0026ldquo;But when you need to think through a complicated problem, you often need other sources.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThrough these everyday practices of information seeking, clinicians gradually extended the informational environment in which clinical reasoning takes place. As clinicians expanded their search for information beyond institutional infrastructures, artificial intelligence applications gradually began to appear within these informal pathways of information seeking. Although most hospitals involved in the study had experimented with specialized medical AI systems\u0026mdash;particularly in imaging analysis\u0026mdash;clinicians rarely described relying on these tools as part of their everyday reasoning process. Instead, the AI tools that appeared most frequently in interviews were consumer-oriented applications based on large language models, accessed through personal mobile devices. Participants emphasized that these tools were not used as formal diagnostic systems. Rather, they functioned as flexible informational resources that could help clinicians organize unfamiliar information, generate potential explanations, or structure complex reasoning processes. In this sense, AI entered clinical practice not as an officially sanctioned component of hospital infrastructures but as an informal supplement to existing knowledge resources.\u003c/p\u003e \u003cp\u003eSeveral clinicians described using applications such as DeepSeek or Doubao during moments of diagnostic uncertainty. These interactions typically occurred when institutional systems failed to provide sufficient interpretive support for understanding complex cases. Under such circumstances, doctors occasionally turned to AI interfaces as a way of reorganizing fragmented pieces of information. One resident described encountering a complicated clinical situation during a late-night shift in which a patient presented with a combination of symptoms that did not correspond clearly to any familiar diagnosis.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eYou read the laboratory results and imaging findings again and again,\u0026rdquo; the resident recalled. \u0026ldquo;But they still don\u0026rsquo;t connect in a clear way.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eAfter several unsuccessful attempts to search the hospital system for additional information, the physician opened an AI application on their phone and entered a short description of the case.\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;I wrote a few sentences describing the symptoms and the medications the patient was taking,\u0026rdquo; the doctor explained.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe system generated a response outlining several possible diagnostic interpretations.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eIt didn\u0026rsquo;t give a definitive answer,\u0026rdquo; the resident emphasized. \u0026ldquo;But it helped organize the information. Sometimes you just need something that helps you see the structure of the problem.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eFor clinicians working under significant cognitive pressure, such interactions could provide a way of reorganizing information that initially appeared confusing or disconnected. Rather than replacing clinical judgment, AI responses were typically treated as provisional suggestions that required further verification through established medical knowledge. Another physician described using an AI application when encountering unfamiliar drug interactions that the hospital\u0026rsquo;s internal database could not easily interpret. While the official system could identify known interactions between two medications, it struggled to analyze more complex combinations involving multiple pharmacological effects.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSometimes the patient is taking five or six medications,\u0026rdquo; the doctor explained. \u0026ldquo;The hospital system only tells you about simple interactions.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIn such cases, the physician occasionally entered the list of medications into an AI interface to obtain a preliminary overview of possible relationships.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe AI might suggest several possibilities,\u0026rdquo; the doctor said. \u0026ldquo;Then you check the guidelines or research papers to see whether those suggestions make sense.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe physician emphasized that the AI response functioned primarily as an organizational tool rather than a source of authoritative knowledge.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eIt helps you think,\u0026rdquo; the doctor explained. \u0026ldquo;But the final interpretation still depends on your own judgment.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThese accounts suggest that clinicians often used AI applications as cognitive aids rather than diagnostic authorities. The systems were consulted briefly, often for only a few seconds, before clinicians returned to their primary workstations to continue analyzing patient data. In this sense, AI tools became one informational element within a broader set of resources that clinicians used while navigating diagnostic uncertainty. Participants frequently emphasized that such practices were pragmatic responses to the cognitive demands of contemporary medical work. Clinical reasoning requires physicians to synthesize diverse forms of information, including imaging data, laboratory results, patient histories, treatment guidelines, and emerging research findings. When institutional systems failed to integrate these elements effectively, clinicians often relied on additional informational resources to bridge the resulting gaps. From this perspective, the use of AI applications represented a continuation of long-standing practices of information seeking within medical work. Doctors have traditionally relied on a wide range of external resources\u0026mdash;textbooks, online databases, research articles, and consultations with colleagues\u0026mdash;when confronting unfamiliar cases(Timmermans \u0026amp; Berg, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). AI tools appeared as a new addition to this informational landscape, offering a different way of organizing and exploring medical knowledge. The flexibility of language-model interfaces made them particularly suited to this role. Unlike structured hospital databases, which required users to navigate predefined search categories, AI systems allowed clinicians to formulate questions in natural language. This conversational format made it easier to explore complex or ambiguous clinical situations that did not fit neatly within standardized query structures. One clinician described the contrast between the two types of systems.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eHospital software is very structured,\u0026rdquo; the doctor explained. \u0026ldquo;You have to search in a specific way.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eAI interfaces, by contrast, allowed doctors to formulate questions more freely.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eWith AI, you can describe the situation in your own words,\u0026rdquo; the physician said. \u0026ldquo;It feels more like discussing a problem than searching a database.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eFor clinicians accustomed to navigating rigid institutional systems, this conversational flexibility could provide a useful complement to existing information infrastructures. What began as an individual strategy for managing uncertainty, however, often extended beyond private consultation. As clinicians shared experiences with colleagues, information obtained through digital searches\u0026mdash;including AI-generated responses\u0026mdash;sometimes circulated within broader professional networks. Messaging platforms played an important role in this process. Many departments maintained group chats through which clinicians exchanged diagnostic opinions, discussed unusual cases, and shared references to relevant medical literature. These digital communication channels functioned as informal spaces for collaborative reasoning, operating alongside the formal documentation processes required by hospital systems. One participant described a situation in which a senior physician shared a screenshot of an AI-generated explanation within a departmental messaging group. The discussion began when several doctors debated the interpretation of an unusual imaging pattern.\u003c/p\u003e \u003cp\u003e \u003cem\u003e\u0026ldquo;Different people had different ideas,\u0026rdquo; the participant recalled.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eAfter several interpretations had been proposed, the senior physician posted an AI-generated response that outlined a possible explanation for the observed pattern.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe professor sent the screenshot and asked everyone what they thought about it.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe AI output did not end the discussion. Instead, it became one element within an ongoing conversation among clinicians.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSome doctors thought the explanation was reasonable,\u0026rdquo; the participant explained. \u0026ldquo;Others said the AI was simplifying the case too much.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eRather than functioning as an authoritative diagnosis, the AI-generated text became a conversational artifact around which clinicians debated alternative interpretations. Through this process, the informational output of the AI system was evaluated, critiqued, and contextualized within the collective knowledge of the group. These exchanges illustrate how information obtained through digital tools may circulate through professional networks before becoming integrated into clinical reasoning. Diagnostic interpretation in hospital environments rarely occurs in isolation. Instead, clinicians frequently rely on discussions with colleagues when confronting complex cases. Messaging platforms provide a convenient medium for these conversations, allowing doctors to share images, references, and opinions in real time. AI-generated responses occasionally entered these discussions as additional informational resources. Yet their significance depended entirely on how clinicians interpreted them within the context of professional expertise. The algorithmic output did not replace medical judgment; rather, it became one element within a broader process of collaborative reasoning. Participants also noted that these practices often reflected generational differences in how clinicians engaged with digital technologies. Younger doctors tended to experiment more frequently with AI applications, particularly during demanding training periods when rapid access to explanatory information could help manage cognitive overload. For residents still developing diagnostic confidence, AI tools sometimes served as a way of organizing unfamiliar knowledge.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eWhen you are still learning,\u0026rdquo; one resident explained, \u0026ldquo;you sometimes need help seeing the structure of a problem.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eSenior physicians, by contrast, were generally more cautious in their use of AI tools. Many described relying primarily on established medical references and professional experience when interpreting cases. Yet even among senior clinicians, AI-generated information occasionally entered professional discussions through messaging groups or informal conversations. Over time, informational resources discovered by younger doctors could circulate upward through these interactions.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSometimes a junior doctor finds something interesting and shares it with the group,\u0026rdquo; one participant explained. \u0026ldquo;Then the senior doctors comment on it or add their own interpretation.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThrough these exchanges, what begins as an individual strategy for resolving uncertainty may gradually become incorporated into collective discussions about clinical reasoning.\u003c/p\u003e \u003cp\u003eThese observations suggest that clinicians increasingly operate within a complex informational environment composed of multiple technological and social resources. Institutional platforms provide access to patient records and imaging data, while personal devices offer entry points to external knowledge sources. Professional communication networks allow clinicians to exchange interpretations and evaluate new information collaboratively. Diagnostic reasoning unfolds through the movement between these different environments rather than through reliance on any single system.\u003c/p\u003e \u003cp\u003eThe practices described above illustrate how clinicians assemble fragments of knowledge drawn from multiple sources in order to interpret complex cases. Rather than depending on a single authoritative platform, doctors combine insights from institutional systems, external digital tools, and professional discussions. Through this ongoing process of searching, comparing, and interpreting information, clinicians gradually construct workable understandings of clinical situations even when institutional infrastructures prove incomplete or inflexible. At the same time, these practices exist alongside persistent structures of professional responsibility. Regardless of how information is assembled during the reasoning process, the final clinical decision must still be attributed to the physician who signs the medical report. Doctors may consult colleagues, databases, or digital tools while interpreting a case, yet the institutional framework of accountability remains firmly individualized.The implications of this tension between distributed informational practices and individualized responsibility are explored in the following discussion section\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe findings presented in this study illuminate how clinicians incorporate AI tools into everyday clinical reasoning in contexts where institutional digital infrastructures prove incomplete or inflexible. Rather than entering clinical practice primarily through formally deployed hospital systems, AI tools frequently appear through informal and situational practices that emerge within the everyday realities of clinical work (Blease et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Shortliffe \u0026amp; Sep\u0026uacute;lveda, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In the empirical cases described above, clinicians encountered institutional platforms that structured documentation and information retrieval but often failed to support the interpretive flexibility required for complex diagnostic reasoning. When these infrastructures proved insufficient, clinicians expanded their search for information beyond the institutional workstation, turning toward personal devices, external digital resources, and increasingly AI-powered applications (Coiera, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Sinsky et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). In these moments, AI did not function as a fully integrated clinical technology but rather as one informational element within a broader assemblage of resources through which clinicians attempted to resolve uncertainty. These findings suggest that the incorporation of AI into clinical reasoning unfolds through a process that differs significantly from the scenarios often described in discussions of medical AI adoption (Jiang et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Much existing research examines AI primarily as a formally implemented technological system, focusing on algorithmic accuracy, regulatory approval, or institutional deployment (Rajpurkar et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Reddy et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). From this perspective, AI appears as a discrete technology that enters clinical environments through structured processes of validation and integration. Yet the empirical material presented in this article reveals a different pathway of technological incorporation. Rather than waiting for formal institutional integration, clinicians frequently engage with AI tools through pragmatic experimentation, drawing on consumer applications available through personal devices to supplement existing infrastructures (Amann et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Cabitza et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). In doing so, they integrate AI into diagnostic reasoning not through official systems but through everyday practices of information seeking and interpretation.\u003c/p\u003e \u003cp\u003eA growing body of research on medical AI has focused on the challenges of integrating algorithmic systems into clinical workflows through formal institutional processes (Greenhalgh et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Sendak et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Studies in health informatics and medical AI governance often examine questions of validation, regulation, and clinical implementation, emphasizing how AI technologies must undergo extensive evaluation before being incorporated into hospital infrastructures (Reddy et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Topol, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Within this literature, the central analytical problem typically concerns how healthcare institutions can successfully integrate algorithmic systems into established clinical routines. By contrast, the empirical cases examined in this study suggest that the incorporation of AI into clinical reasoning may also occur through pathways that lie largely outside formal institutional implementation. Rather than waiting for AI systems to be officially deployed within hospital infrastructures, clinicians sometimes experiment with readily available digital tools in order to address immediate cognitive demands arising within everyday medical work.\u003c/p\u003e \u003cp\u003eTo conceptualize this process, the article introduced the notion of cognitive stitching. The concept captures the practical work through which clinicians assemble heterogeneous informational resources in order to sustain diagnostic reasoning under conditions of infrastructural limitation. As the empirical accounts illustrate, clinicians rarely rely on a single technological system when confronting complex cases. Instead, they move between multiple informational environments: hospital information systems provide structured patient data; imaging platforms offer visual evidence; online databases supply reference knowledge; messaging networks facilitate collegial consultation; and AI applications generate provisional explanations or analytical summaries. Diagnostic reasoning emerges through the continual movement between these sources, as clinicians selectively combine fragments of information into coherent interpretations of patient conditions. Understanding this process requires situating AI within the broader ecology of clinical knowledge production rather than treating it as an isolated technological intervention (Abbott, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Larson, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). Clinical reasoning has long been understood as a process that unfolds across multiple informational resources, including diagnostic technologies, professional consultation, and institutional knowledge systems. The findings presented here suggest that clinicians actively assemble these heterogeneous resources through a practical process that this article conceptualizes as cognitive stitching. Medical knowledge is rarely generated solely through the internal cognition of individual practitioners (Good, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e1994\u003c/span\u003e; Mol, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). Instead, it emerges through the interplay of diagnostic technologies, professional collaboration, and standardized knowledge systems that collectively shape how clinicians interpret clinical information. In contemporary healthcare environments, the number of informational resources available to clinicians has expanded dramatically as digital technologies multiply the tools through which medical knowledge can be accessed and interpreted. Under such conditions, clinicians must actively coordinate these heterogeneous resources in order to sustain diagnostic reasoning\u0026mdash;a process captured by the concept of cognitive stitching. From this perspective, the practices described in this study can be understood as part of a broader pattern of technological adaptation within clinical work. Research in science and technology studies has repeatedly shown that technologies rarely determine professional practice through their design alone. Instead, their meaning and function emerge through situated interactions within specific organizational environments (Bruni \u0026amp; Teli, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Douglas, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Suchman, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). Technologies may be designed with particular uses in mind, but their practical roles are continually reshaped by the everyday activities of users navigating institutional constraints. Berg\u0026rsquo;s studies of medical information systems similarly demonstrated how digital infrastructures become embedded within clinical work through processes of local negotiation and adjustment (Berg, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). The empirical findings presented here extend these insights to the emerging domain of AI-assisted medicine. Rather than entering clinical practice as autonomous diagnostic systems, AI tools become integrated into reasoning processes through the situated improvisations of clinicians responding to gaps within existing infrastructures.\u003c/p\u003e \u003cp\u003e The importance of these improvisational practices becomes particularly visible when examining the moments of infrastructural friction described by participants. Institutional hospital systems were designed to structure the flow of clinical information, enforce standardized documentation procedures, and regulate diagnostic workflows. Yet clinicians frequently encountered situations in which these systems proved unable to accommodate the complexity or ambiguity of real clinical cases. Rigid search interfaces, fragmented data architectures, and restrictive query structures often limited the ability of practitioners to explore alternative interpretations of clinical information. In such contexts, the technological infrastructures intended to support clinical reasoning could themselves become sources of cognitive constraint. Rather than abandoning digital tools altogether, clinicians responded by expanding the range of resources they used in the reasoning process. Personal devices, external databases, and AI applications became supplementary tools through which doctors attempted to overcome the limitations of institutional systems. These practices resonate strongly with earlier research on workarounds in healthcare information systems. Studies of hospital technologies have documented how clinicians frequently develop informal strategies to circumvent rigid digital infrastructures that fail to align with the practical demands of clinical work (Ash et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Ellingsen \u0026amp; Monteiro, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2003\u003c/span\u003e; Koppel et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). Such workarounds are often interpreted as deviations from intended system use. However, they may also be understood as essential mechanisms through which complex sociotechnical systems remain operational in practice. Rather than simply representing failures of technological design, these adaptations reveal how practitioners actively reshape infrastructures to fit the realities of everyday work. The incorporation of AI tools described in this article can be interpreted as a continuation of this broader pattern of infrastructural improvisation. When institutional systems prove incomplete, clinicians extend their reasoning capacities by incorporating additional digital resources into the diagnostic process.\u003c/p\u003e \u003cp\u003eEarlier research on clinical decision-support systems has similarly emphasized how digital tools may assist clinicians by providing structured recommendations within institutional workflows. However, such systems are typically designed as formally integrated components of hospital infrastructures, operating through standardized interfaces connected to electronic health records. The practices observed in this study differ in an important respect. The AI tools used by clinicians were not embedded within official hospital platforms but accessed through personal devices and external applications. As a result, their incorporation into clinical reasoning occurred through informal and situational practices rather than through institutional design. This distinction highlights the importance of examining how emerging digital technologies become integrated into professional work not only through formal implementation but also through the everyday improvisations of practitioners navigating infrastructural constraints. Within this expanding informational landscape, AI tools acquire a particular role. Unlike traditional databases or static reference materials, large language model applications provide conversational interfaces capable of generating structured explanations in response to complex queries. For clinicians confronting unfamiliar clinical situations, such tools offer a rapid way to reorganize fragmented information into provisional interpretive frameworks. Importantly, participants in this study did not describe AI outputs as authoritative diagnoses. Instead, they treated them as tentative suggestions that required further verification through established professional practices. Algorithmic responses were compared with clinical guidelines, medical literature, or the experiential knowledge of colleagues before being incorporated into final diagnostic judgments. In this sense, AI functions less as a replacement for professional expertise than as a cognitive aid embedded within existing interpretive routines.\u003c/p\u003e \u003cp\u003eThis observation also diverges from a strand of literature that frames medical AI primarily in terms of automation or the potential displacement of clinical expertise. Public and scholarly debates frequently center on whether machine learning systems might eventually outperform physicians in diagnostic tasks (Esteva et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Topol, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). While such discussions highlight important technological possibilities, they often assume that AI systems will enter clinical environments as relatively self-contained diagnostic tools. The empirical findings presented here instead suggest that AI is more likely to become embedded within existing interpretive practices in incremental and often informal ways. Rather than replacing clinical reasoning, AI applications become integrated into the broader ecology of knowledge resources through which clinicians organize, interpret, and evaluate medical information(El-Najdawi \u0026amp; Stylianou, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e1993\u003c/span\u003e; Lebovitz et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This mode of incorporation has important implications for how we understand the role of AI in medical work. Public discussions of medical AI often frame the technology in terms of potential substitution\u0026mdash;whether algorithms will eventually replace human clinicians in diagnostic tasks. The empirical findings presented here suggest a different dynamic. Rather than displacing clinical reasoning, AI tools are woven into existing processes through which clinicians interpret complex information. They become additional informational elements that clinicians incorporate into the ongoing process of cognitive stitching, through which fragments of medical knowledge drawn from different sources are combined into workable diagnostic interpretations. The result is not a simple transfer of cognitive authority from human practitioners to algorithmic systems, but a transformation in how clinicians perform cognitive stitching, assembling informational fragments from institutional systems, professional networks, and digital tools into coherent clinical interpretations. At the same time, the emergence of cognitive stitching reveals an important tension within the institutional organization of clinical practice. While diagnostic reasoning increasingly involves distributed informational resources, the structures governing professional accountability remain firmly individualized. Medical decisions continue to be formally attributed to the physician who signs the clinical report, regardless of the technological resources involved in generating that decision. This arrangement reflects the longstanding professional structure of medicine, in which clinical authority and responsibility are closely tied to the individual practitioner. Yet the practices described in this study indicate that the cognitive processes underlying those decisions may involve a far more complex network of informational actors.\u003c/p\u003e \u003cp\u003eThe practices described above also reveal an important institutional tension surrounding cognitive stitching. While clinicians assemble diagnostic understanding by drawing on multiple informational resources\u0026mdash;including institutional systems, digital tools, and professional networks\u0026mdash;the formal structures of medical responsibility remain firmly individualized. Technologies often redistribute tasks, information flows, and cognitive labor without fundamentally altering the institutional structures through which responsibility is assigned (Suchman, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). In clinical environments, AI systems may assist with data interpretation, hypothesis generation, or information retrieval, but they do not assume accountability for diagnostic outcomes. Instead, clinicians remain responsible for evaluating algorithmic suggestions and integrating them into professional judgment. As a result, the incorporation of AI into clinical reasoning does not eliminate the central role of human expertise but rather reshapes the informational landscape within which that expertise operates.\u003c/p\u003e \u003cp\u003eImportantly, participants in this study did not generally describe this situation as an ethical crisis. For most clinicians, consulting AI tools was understood as a pragmatic extension of existing practices of information seeking. Doctors have long relied on a wide range of external resources\u0026mdash;textbooks, online databases, professional networks\u0026mdash;when confronting unfamiliar cases. From this perspective, AI appears as a new addition to an already diverse informational ecosystem. What distinguishes AI from earlier resources is not simply its content but the speed and flexibility with which it can generate explanatory structures. By quickly organizing dispersed information into coherent narratives, AI applications provide clinicians with a new way of navigating moments of cognitive uncertainty. Yet the informal nature of these practices also contributes to a degree of institutional ambiguity. AI tools accessed through personal devices often remain outside formal hospital governance structures. Their use is rarely documented within official clinical systems, and their role in shaping diagnostic reasoning remains largely invisible within institutional records. Consequently, AI becomes incorporated into clinical cognition without necessarily being recognized as part of the official technological infrastructure of healthcare institutions. This gap between practical use and institutional recognition illustrates how technological change in professional settings frequently occurs through incremental adaptations rather than formal implementation. Seen in this light, the incorporation of AI into clinical reasoning should be understood less as a technological revolution and more as a gradual transformation in how clinicians perform cognitive stitching, integrating new digital tools into existing practices of medical interpretation. Clinicians continually negotiate the boundaries between institutional systems, professional knowledge, and emerging digital tools in order to sustain the cognitive work of medicine. AI applications become meaningful within this process not because of their technical capabilities alone but because of how they are integrated into the everyday practices through which clinicians interpret clinical information.\u003c/p\u003e \u003cp\u003eBy examining these practices ethnographically, this study contributes to broader discussions about the social life of AI in professional contexts. Rather than treating AI as an external force reshaping medical practice from the outside, the analysis highlights the central role of practitioners in actively incorporating technologies into their work. The concept of cognitive stitching offers one way of understanding how clinicians assemble distributed informational resources in order to bridge the gaps left by institutional infrastructures. Through these small but consequential acts of coordination, emerging technologies become woven into the ongoing routines of clinical reasoning. Understanding these processes is essential for anticipating how AI will continue to evolve within healthcare environments. The future impact of AI in medicine will not depend solely on advances in algorithmic performance or regulatory frameworks. It will also depend on how clinicians integrate these technologies into the situated practices of everyday clinical work. By focusing on these practices, the analysis presented here shifts attention away from abstract debates about technological transformation and toward the practical realities through which AI becomes embedded within the sociotechnical fabric of contemporary medicine\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis article has examined how clinicians incorporate AI tools into everyday clinical reasoning in environments where institutional digital infrastructures remain incomplete or inflexible. Rather than focusing on the formal deployment of medical AI systems, the analysis has traced the situated practices through which clinicians navigate gaps within hospital information systems. Ethnographic accounts from Chinese public hospitals show that doctors frequently rely on a heterogeneous set of informational resources\u0026mdash;including institutional platforms, personal mobile devices, professional communication networks, and increasingly AI applications\u0026mdash;in order to interpret complex clinical cases. Through these practices, clinicians assemble fragmented knowledge into workable diagnostic interpretations, a process described in this article as cognitive stitching. By foregrounding these practices, the study contributes to a growing body of research emphasizing that technologies rarely enter professional environments as fully stabilized systems. Instead, they are incorporated through situated forms of adaptation shaped by the practical conditions of work (Leonardi, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Suchman, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). In the clinical settings examined here, AI tools do not replace medical expertise nor function as autonomous diagnostic authorities. Rather, they become embedded within the cognitive routines through which clinicians organize and interpret medical information. When institutional infrastructures prove insufficient for resolving complex cases, clinicians extend their reasoning by drawing on additional informational resources available through the wider digital environment.\u003c/p\u003e \u003cp\u003eAt the same time, the incorporation of AI into clinical reasoning unfolds within institutional structures that continue to attribute responsibility primarily to individual physicians. As the analysis has shown, clinicians may consult AI tools, online databases, or colleagues while reasoning through difficult cases, yet the final diagnostic decision remains formally attached to the doctor who signs the medical record. This produces a subtle but significant tension between the distributed character of contemporary clinical cognition and the individualized framework of professional accountability. While informational work becomes increasingly mediated by digital tools, responsibility remains firmly anchored in traditional institutional arrangements.\u003c/p\u003e \u003cp\u003eAlthough the empirical material analyzed in this article is drawn from Chinese hospitals, it is useful to briefly reflect on how these dynamics appear differently in other healthcare contexts. Drawing on observations from a separate Norwegian research project, the following comparison is offered not as a formal comparative analysis but as a discussion point highlighting how institutional conditions may shape the incorporation of AI in different ways. In Norway, the integration of medical AI tends to occur through more formalized regulatory pathways, including institutional approval processes and certified clinical software infrastructures. Algorithmic tools are incorporated into hospital systems through clearly defined governance frameworks, making their role within clinical decision-making more visible and institutionally recognized. By contrast, the practices described in Chinese hospitals illustrate a more informal pathway of technological incorporation, in which clinicians draw on widely accessible AI tools through personal devices and professional networks. These contrasting configurations should not be understood simply as differences in technological development but as variations in the sociotechnical organization of medical practice. The Norwegian case demonstrates how AI can become integrated through formal institutional channels, while the Chinese context reveals how clinicians adapt technologies pragmatically when institutional infrastructures lag behind technological possibilities. In both situations, however, the incorporation of AI ultimately depends on the everyday practices of clinicians who must interpret, evaluate, and contextualize algorithmic information within the realities of patient care This brief comparison should therefore be understood as an interpretive reflection rather than a systematic cross-national analysis, pointing to how different institutional environments may shape the ways clinicians incorporate AI into everyday reasoning.\u003c/p\u003e \u003cp\u003eRecognizing these practices helps shift discussions of medical AI away from deterministic narratives about technological transformation. Rather than viewing AI as a system that either replaces or augments clinicians in a straightforward manner, this study suggests that its integration is mediated by the situated practices through which professionals manage uncertainty, interpret digital outputs, and coordinate multiple informational resources. AI becomes meaningful not as an abstract technological capability but as part of the everyday work through which clinicians sustain diagnostic reasoning in complex institutional environments.\u003c/p\u003e \u003cp\u003eThe concept of cognitive stitching offers a way to understand how clinicians sustain diagnostic reasoning in environments where institutional infrastructures remain incomplete. Rather than relying on a single authoritative technological system, practitioners assemble fragments of knowledge drawn from multiple informational environments. Through this ongoing process of stitching together institutional data, professional experience, and algorithmic suggestions, clinicians maintain the interpretive work through which medical decisions are produced. Understanding these practices highlights that the future of AI in healthcare will depend not only on algorithmic capabilities but also on how clinicians incorporate these technologies into the everyday practices through which medical knowledge is constructed.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCONFLICT OF INTEREST STATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author declares no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eETHICS STATEMENT\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll procedures performed in this study involving human participants were conducted in a hospital setting and were in accordance with the ethical standards of the relevant institutional research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.\u003c/p\u003e\n\u003cp\u003eEthical approval for this research was obtained from the relevant institutional ethics review board.\u003c/p\u003e\n\u003cp\u003eInformed consent was obtained from all participants prior to their participation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed Consent\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBefore the interviews, participants were provided with an information sheet outlining the purpose of the study and their rights. They were informed that interviews may be audio-recorded, anonymized, and used for research purposes, and that participation was voluntary. Informed consent was obtained from all participants prior to participation, and all data were handled confidentially.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDATA Availability Statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data underlying this study are not publicly available due to ethical and privacy restrictions. The research was conducted in a hospital setting and received approval from the Ethics Review Board of the relevant institutional research committee, in compliance with the data management regulations of the European Research Council (ERC) and the University of Amsterdam.\u003c/p\u003e\n\u003cp\u003eDue to the nature of the study, which involves qualitative interviews and observational materials collected in a clinical environment, even de-identified data may carry a risk of participant re-identification. As such, sharing raw or processed data through publicly accessible repositories is not permitted under the approved ethical protocols.\u003c/p\u003e\n\u003cp\u003eAll data are securely stored in the University of Amsterdam\u0026rsquo;s protected research environment for a period of ten years in accordance with institutional guidelines. De-identified data may be made available from the author upon reasonable request and subject to appropriate ethical approvals and data access agreements.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAbbott A (2014) The system of professions: An essay on the division of expert labor. University of Chicago Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmann J, Blasimme A, Vayena E, Frey D, Madai VI, Consortium PQ (2020) Explainability for artificial intelligence in healthcare: a multidisciplinary perspective. BMC Med Inf Decis Mak 20(1):310\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAsh JS, Berg M, Coiera E (2004) Some unintended consequences of information technology in health care: the nature of patient care information system-related errors. J Am Med Inform Assoc 11(2):104\u0026ndash;112\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBabic B, Gerke S, Evgeniou T, Cohen IG (2021) Beware explanations from AI in health care. Science 373(6552):284\u0026ndash;286\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerg M (1997) Of forms, containers, and the electronic medical record: some tools for a sociology of the formal. Sci Technol Hum Values 22(4):403\u0026ndash;433\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlease C, Kaptchuk TJ, Bernstein MH, Mandl KD, Halamka JD, DesRoches CM (2019) Artificial intelligence and the future of primary care: exploratory qualitative study of UK general practitioners\u0026rsquo; views. J Med Internet Res 21(3):e12802\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBowker GC, Star SL (2000) Sorting things out: Classification and its consequences. MIT Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBruni A, Teli M (2007) Reassembling the social\u0026mdash;An introduction to actor network theory. Manage Learn 38(1):121\u0026ndash;125\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCabitza F, Rasoini R, Gensini GF (2017) Unintended consequences of machine learning in medicine. JAMA 318(6):517\u0026ndash;518\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChang CK, Chiari L, Cao Y, Jin H, Mokhtari M, Aloulou H (2016) \u003cem\u003eInclusive Smart Cities and Digital Health: 14th International Conference on Smart Homes and Health Telematics, ICOST 2016, Wuhan, China, May 25\u0026ndash;27, 2016. Proceedings\u003c/em\u003e (Vol. 9677). Springer\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoiera E (2015) Guide to health informatics. CRC\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeloitte (2021) \u003cem\u003eInternet Hospitals in China: The new step into digital healthcare\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDouglas DG (2012) The social construction of technological systems, anniversary edition: New directions in the sociology and history of technology. MIT Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEl-Najdawi M, Stylianou AC (1993) Expert support systems: integrating AI technologies. Commun ACM 36(12):55\u0026ndash;ff\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEllingsen G, Monteiro E (2003) A patchwork planet integration and cooperation in hospitals. Comput Supported Coop Work (CSCW) 12(1):71\u0026ndash;95\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEsteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, Cui C, Corrado G, Thrun S, Dean J (2019) A guide to deep learning in healthcare. Nat Med 25(1):24\u0026ndash;29\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao J, Zhang P (2021) China's Public Health Policies in Response to COVID-19: From an Authoritarian Perspective. Front Public Health 9:756677. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpubh.2021.756677\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2021.756677\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGood BJ (1994) Medicine, rationality and experience: an anthropological perspective. Cambridge University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGreenhalgh T, Wherton J, Papoutsi C, Lynch J, Hughes G, Hinder S, Fahy N, Procter R, Shaw S (2017) Beyond adoption: a new framework for theorizing and evaluating nonadoption, abandonment, and challenges to the scale-up, spread, and sustainability of health and care technologies. J Med Internet Res 19(11):e8775\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo Z, Cugurullo F (2023) AI doctors or AI for doctors? Augmenting urban healthcare services through artificial intelligence. Artificial Intelligence and the City. Routledge, pp 307\u0026ndash;321\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K (2019) The practical implementation of artificial intelligence technologies in medicine. Nat Med 25(1):30\u0026ndash;36\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, present and future. \u003cem\u003eStroke and vascular neurology\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(4)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoppel R, Wetterneck T, Telles JL, Karsh B-T (2008) Workarounds to barcode medication administration systems: their occurrences, causes, and threats to patient safety. J Am Med Inform Assoc 15(4):408\u0026ndash;423\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLarson MS (2003) Professionalism: The third logic. Perspect Biol Med 46(3):458\u0026ndash;462\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLaurie G (2017) Liminality and the limits of law in health research regulation: what are we missing in the spaces in-between? Med Law Rev 25(1):47\u0026ndash;72\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLebovitz S, Lifshitz-Assaf H, Levina N (2022) To engage or not to engage with AI for critical judgments: How professionals deal with opacity when using AI for medical diagnosis. Organ Sci 33(1):126\u0026ndash;148\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeonardi PM (2011) When flexible routines meet flexible technologies: Affordance, constraint, and the imbrication of human and material agencies1. MIS Q 35(1):147\u0026ndash;167\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang H, Tsui BY, Ni H, Valentim CC, Baxter SL, Liu G, Cai W, Kermany DS, Sun X, Chen J (2019) Evaluation and accurate diagnoses of pediatric diseases using artificial intelligence. Nat Med 25(3):433\u0026ndash;438\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipset S (2017) Social organization of medical work. Routledge\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y, Rao K, Wu J, Gakidou E (2008) China's health system performance. Lancet 372(9653):1914\u0026ndash;1923\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMol A (2002) The body multiple: Ontology in medical practice. Duke University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eObermeyer Z, Emanuel EJ (2016) Predicting the future\u0026mdash;big data, machine learning, and clinical medicine. N Engl J Med 375(13):1216\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOrlikowski WJ (2000) Using technology and constituting structures: A practice lens for studying technology in organizations. Organ Sci 11(4):404\u0026ndash;428\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajpurkar P, Chen E, Banerjee O, Topol EJ (2022) AI in health and medicine. Nat Med 28(1):31\u0026ndash;38\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReddy S, Fox J, Purohit MP (2019) Artificial intelligence-enabled healthcare delivery. J R Soc Med 112(1):22\u0026ndash;28\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSendak MP, D\u0026rsquo;Arcy J, Kashyap S, Gao M, Nichols M, Corey K, Ratliff W, Balu S (2020) A path for translation of machine learning products into healthcare delivery. EMJ Innov 10:19\u0026ndash;00172\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShanafelt TD, Dyrbye LN, Sinsky C, Hasan O, Satele D, Sloan J, West CP (2016) Relationship between clerical burden and characteristics of the electronic environment with physician burnout and professional satisfaction. Mayo clinic proceedings\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShortliffe EH, Sep\u0026uacute;lveda MJ (2018) Clinical decision support in the era of artificial intelligence. JAMA 320(21):2199\u0026ndash;2200\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSinsky C, Colligan L, Li L, Prgomet M, Reynolds S, Goeders L, Westbrook J, Tutty M, Blike G (2016) Allocation of physician time in ambulatory practice: a time and motion study in 4 specialties. Ann Intern Med 165(11):753\u0026ndash;760\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStar SL, Ruhleder K (2010) Steps toward an Ecology of Infrastructure. Revue d'anthropologie des connaissances 41(1):114\u0026ndash;161\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuchman LA (2007) Human-machine reconfigurations: Plans and situated actions. Cambridge University Press\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTimmermans S, Berg M (2010) The gold standard: the challenge of evidence-based medicine. Temple University\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTopol E (2019) \u003cem\u003eDeep medicine: how artificial intelligence can make healthcare human again\u003c/em\u003e. Hachette UK\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWachter R (2017) Digital Doctor. Hope, Hype and Harm at Dawn. McGraw-Hill, New York\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu KH, Beam AL, Kohane IS (2018) Artificial intelligence in healthcare. Nat Biomed Eng 2(10):719\u0026ndash;731. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41551-018-0305-z\u003c/span\u003e\u003cspan address=\"10.1038/s41551-018-0305-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9095240/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9095240/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial intelligence is often discussed in healthcare in terms of algorithmic performance, regulatory approval, or institutional deployment. Such perspectives tend to treat AI as a formally implemented technology integrated into hospital infrastructures. Yet relatively little attention has been paid to how clinicians actually encounter and use AI within the everyday realities of clinical work. Drawing on ethnographic fieldwork conducted in Chinese public hospitals, this article examines how clinicians incorporate AI tools into diagnostic reasoning when institutional digital infrastructures prove incomplete or inflexible.\u003c/p\u003e \u003cp\u003eThe analysis shows that AI frequently enters clinical practice not through officially deployed hospital systems but through informal practices involving personal mobile devices and widely accessible consumer applications. When institutional platforms fail to provide sufficient interpretive support, clinicians turn to external digital resources to organize information, explore diagnostic possibilities, and manage cognitive uncertainty. To conceptualize these practices, the article introduces the notion of cognitive stitching, referring to the practical process through which clinicians assemble heterogeneous informational resources\u0026mdash;including hospital systems, online knowledge sources, professional communication networks, and AI tools\u0026mdash;into workable diagnostic interpretations.\u003c/p\u003e","manuscriptTitle":"Cognitive Stitching: Informal AI Integration in Everyday Clinical Reasoning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-29 18:39:28","doi":"10.21203/rs.3.rs-9095240/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"31985972698892949123669193533919170551","date":"2026-04-23T10:16:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"245756168530217788336246087304861642259","date":"2026-04-22T11:09:43+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-21T10:05:34+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-14T13:33:15+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-06T12:35:59+00:00","index":"","fulltext":""},{"type":"submitted","content":"Humanities and Social Sciences Communications","date":"2026-04-06T11:42:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humanities-and-social-sciences-communications","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"palcomms","sideBox":"Learn more about [Humanities \u0026 Social Sciences Communications](http://www.nature.com/palcomms/)","snPcode":"41599","submissionUrl":"https://submission.springernature.com/new-submission/41599/3","title":"Humanities and Social Sciences Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"634f44a0-a967-405f-9169-08f26388f357","owner":[],"postedDate":"April 29th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":66873834,"name":"Scientific community and society/Business and industry"},{"id":66873836,"name":"Health sciences/Health care"},{"id":66873838,"name":"Physical sciences/Mathematics and computing"},{"id":66873839,"name":"Scientific community and society/Scientific community"},{"id":66873841,"name":"Scientific community and society/Social sciences"}],"tags":[],"updatedAt":"2026-04-29T18:39:28+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-29 18:39:28","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9095240","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9095240","identity":"rs-9095240","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}