A total of 20 physicians were interviewed (see Table 1). Participants represented a diverse range of medical specialties and hospital types. To capture a broad range of perspectives, physicians from different specialties and hospital types were included (Table 1).

The participants’ contact information was provided through the research consortium, after which the researchers at Saxion contacted them through email. In-person interviews were recorded using a digital audio recorder. Online interviews conducted via Microsoft Teams were recorded verbatim using the platform’s built-in audio-recording functionality. The participants were interviewed from mid-December 2024 to late March 2025, with participation being fully voluntary with no additional incentive. The interviews lasted 45 to 60 minutes and were conducted either in person or virtually.

The interview protocol was based on a pilot interview and consultations with domain experts in medical craftsmanship and the use of AI in hospital settings (see Multimedia Appendix 1 for the protocol). Because this study aimed to explore physicians’ own perceptions and lived experiences, no fixed definition of AI was provided during the interview. Instead, participants were invited to describe what they understood as “AI” in their clinical context and to refer to any technologies they associated with it. This approach ensured that all examples, whether technically accurate or not, reflected what physicians themselves perceive as AI, which is essential for understanding attitudes, expectations, and concerns.

The semistructured interviews were divided into 3 key parts. First, participants were asked about their background and what craftsmanship meant to them in their specific medical profession, including questions about core skills, values, and motivations in their daily work. The second part focused on their current experiences with AI in clinical practice, ranging from familiarity and observed use to perceived benefits and limitations based on their own definition of AI. The final part addressed expectations and concerns about AI’s future role, including its potential influence on craftsmanship.

All interviews were transcribed using the transcription tool Amberscript (Amberscript BV). The transcripts were anonymized before analysis. Data were analyzed using Atlas.ti (ATLAS.ti Scientific Software Development GmbH) following a thematic analysis approach that combined inductive and deductive coding [21]. The exploratory nature of this study required an iterative and reflective process, led by two researchers who were both involved in the project. The researchers involved in the analysis brought different disciplinary perspectives to the study. The primary researcher has a background in human-centered design and qualitative research, while the second researcher has a background in psychology. These perspectives informed the coding process and interpretation of the data and were discussed during analysis to support reflexivity.

Initial coding followed an inductive and iterative process. Codes were developed from recurring patterns in the interview data and were informed by sensitizing concepts derived from earlier stages of the research. To establish analytic credibility and a shared understanding of the data, two researchers independently coded an initial subset of two interview transcripts. Coding decisions and interpretative differences were discussed in detail until consensus was reached, resulting in refinement and clarification of the coding framework.

The primary researcher subsequently applied the finalized coding scheme to the full interview dataset. Throughout this process, analytic decisions were documented and regularly discussed with the other researcher and, when needed, the full research team to support trustworthiness and analytic consistency. During this iterative process of coding and analysis, we continuously assessed whether new codes and themes emerged from the data. After 18 interviews, no new codes or conceptually relevant themes were identified. The two subsequent interviews did not yield additional insights but served to confirm and refine the existing codes and themes, indicating that thematic saturation had been reached.

To synthesize the findings, themes were developed by grouping related codes and identifying overarching patterns in physicians’ perceptions of craftsmanship and the impact of AI. The final themes were mapped onto the aspects of professional and personal craftsmanship.

Two focus groups were conducted, each with 15 participants, during a regional symposium on AI in hospital care. Participation was based on symposium attendance and self-selection into the workshop during the day. Each group included a heterogeneous mix of physicians, physicians in training, hospital staff with responsibilities for digital transformation, hospital policymakers, and AI developers. Because participants did not preregister individually for specific workshops, no additional individual demographic information was collected.

The focus groups were designed as interactive sessions using structured exercises. In the first round, participants worked with 3 personas developed from the interview data (see Multimedia Appendix 2). These personas represented 3 nonclinical stakeholders involved in the design, policy, and implementation of AI in health care: an AI developer working at an external company, a policy advisor for digital health in a teaching hospital, and an implementation coordinator for digital innovations in a general hospital. Each persona faced distinct challenges in aligning AI with clinical workflows and professional values. The personas were constructed by synthesizing recurring frustrations, concerns, and collaboration challenges articulated in the interviews by physicians in relation to nonclinical stakeholders involved in AI development and implementation. To enhance credibility, the personas were discussed with experts (n=6) working in roles comparable to those represented, including AI developers and hospital-based AI implementation experts. These experts were recruited through the researchers’ professional networks and were consulted to better understand practical realities and to assess the recognizability and plausibility of the personas.

Participants were asked to empathize with the perspective of the physician and provide advice to these personas on how they could address their problems, needs, and dilemmas when collaborating with physicians in daily practice. Personas and related techniques such as role-playing are commonly used in focus group and participatory design research as effective instruments to enhance empathy and enable participants to take on perspectives other than their own [22].

For this interactive part, the Mentimeter software (Mentimeter AB) was used to allow each participant to provide input individually. Subsequently, participants worked in groups to reflect on the given advice (visible on a screen) and cocreate practical principles for how AI can best align with physicians’ craftsmanship, values, and daily clinical practice. For this, participants wrote down principles on a worksheet.

All written materials and digital output of the focus groups were anonymized and imported into the collaborative digital workspace Miro (Miro, Inc) and treated as a single analytic dataset.

The primary researcher conducted an inductive thematic clustering of the data. Individual contributions were segmented into analytic units and iteratively grouped into thematic clusters, such as working from clinical practice and end-user needs, reflecting and continuous learning, integration with existing processes and collaborations, and business cases and strategic considerations.

Subsequently, thematic clusters were presented in a cocreation session with 7 members of the project consortium to translate them into concrete design principles for the responsible design and implementation of AI in hospital practice. The design principles were finally chronologically structured into an iterative step-by-step workflow.

When asked to define AI, physicians often described AI as a broad and ambiguous concept. Many participants indicated that they found it difficult to provide a clear definition, referring to AI as a “container concept” or “umbrella term” encompassing a wide range of technologies. Rather than offering formal definitions, physicians frequently described AI in functional terms, such as a system that supports or automates human tasks, or as a technology that performs certain activities faster or more efficiently than humans, as a physician said:

When discussing current use, physicians referred to a wide range of AI-related applications. These included generative and supportive tools, such as large language models (eg, ChatGPT, Copilot, and Perplexity), which were primarily used for information retrieval, summarization, or writing support and were often applied informally or outside direct patient care.

Participants mentioned predictive and decision support applications, including no-show prediction models, discharge and readmission risk prediction tools, and triage support systems. Physicians had heard of these applications or were working with them as a part of pilot projects.

In addition, physicians referred to diagnostic AI applications, particularly in imaging-intensive domains such as radiology and pathology, including tools for image analysis, segmentation, and detection of abnormalities. Administrative and workflow-supporting technologies were also discussed, such as speech-to-text systems and process automation aimed at reducing documentation and registration burdens.

Overall, although awareness of AI applications was high, physicians reported limited active use of AI in routine clinical practice.

Across interviews, physicians depicted craftsmanship as a dynamic and fundamentally human practice in which knowledge, technical and social skills, reflection, and moral awareness come together. Central to this craftsmanship was a strong orientation toward quality: striving to deliver the best possible care within the realities of the health care system.

Figure 2 visualizes how physicians currently perceived the influence of AI on different dimensions of craftsmanship. The perceived influence of AI on their craftsmanship was described as affecting the professional dimensions of craftsmanship (shown by the AI icon), while its perceived influence on personal dimensions was described as more indirect and mediated through changes in professional responsibilities, workflows, and decision-making processes (represented by the dashed arrow).

Physicians unanimously acknowledged the core features of professional craftsmanship within the interviews.

The essence of personal craftsmanship lies in the professional’s intrinsic drive to deliver the highest possible quality of work or service. Physicians echoed this notion, describing craftsmanship as the willingness for optimal quality of care. Quality was not limited to technical performance but also encompassed moral and relational dimensions: doing the right thing for the patient, the team or the system, and society. Three levels of quality orientation were identified: patient-level (individualized, value-based care), system-level (working within guidelines and resource limits), and societal-level (responsibility for equity and sustainability). As one surgeon explained:

Overall, physicians anticipated that AI could indirectly support quality by reducing repetitive or administrative work, thereby creating more time for patient contact, reflection, and complex clinical reasoning. Some also noted that AI-supported data integration could contribute to more individualized decision-making at the patient level.

At the same time, physicians expressed concerns that excessive reliance on AI might undermine essential skills and, in turn, compromise care quality. Across interviews, participants emphasized that defining and safeguarding quality ultimately remains a human responsibility, requiring sensitivity to context and ongoing judgment. Maintaining this orientation toward quality was therefore closely linked to continuous education and AI literacy, ensuring that technological support does not replace, but informs, professional decision-making.

Within personal craftsmanship, knowledge and skills refer to how individual physicians practically develop, embody, and apply expertise in situated clinical practice, rather than to the formal and collectively defined knowledge base that legitimizes the profession. Participants described these skills as experiential, tacit, and context-dependent, developed through repeated exposure, hands-on practice, and reflection rather than through formal instruction alone. As one surgeon remarked:

AI was perceived to influence this dimension indirectly. On the one hand, it introduces new personal competencies, such as maintaining critical oversight of algorithmic outputs and integrating AI-generated information into clinical reasoning. On the other hand, physicians cautioned that reduced engagement with certain tasks due to AI support may limit opportunities to exercise and refine experiential skills, potentially altering how personal expertise is developed and sustained over time.

Within personal craftsmanship, development and reflection do not refer to the structured, collective arrangement of learning (experience and continuous development at the professional level), but to the individual’s reflective disposition: the willingness to critically examine one’s own actions, remain open to feedback, and adjust behavior accordingly. Participants framed this as a mindset that underpins integrity and everyday judgment, distinct from formal training structures. As one participant explained:

When discussing AI, physicians generally viewed it as a potential aid to reflection. AI could support post hoc learning, for instance, by generating analyses or prompting comparison between clinical decisions and algorithmic outputs.

Overall, physicians considered reflection and self-development as enduring core elements of craftsmanship qualities that may be technologically supported but must remain firmly anchored in human responsibility and judgment.

Within personal craftsmanship, individual autonomy does not refer to the formal decision-making space granted to the profession but to the individual physician’s capacity to inhabit that space: exercising independent judgment, recognizing uncertainty, and taking responsibility for decisions in concrete situations. Physicians described individual autonomy as the ability to act at one’s own discretion based on training, knowledge, and experience. Autonomy was seen as knowing both the limits of one’s own expertise and when to seek support:

Several participants stressed that context-sensitive judgment cannot be handed over to AI. While they acknowledged that AI might simulate aspects of communication or empathy, they were clear that tailoring care to the unique situation of an individual patient remains a human task.

Autonomy was therefore also used to draw a clear boundary for AI: algorithms may support protocol-based decisions but lack the capacity to judge when protocols no longer fit the situation.

Physicians described involvement as something that comes automatically with craftsmanship, an intrinsic sense of engagement with patients, colleagues, and the profession itself. This involvement was often framed in relation to taking or having time for being present, listening to patients, and staying up to date with knowledge to provide the best possible care. Several physicians linked this involvement to a sense of shared responsibility within their peer group.

However, participants also acknowledged that this deep level of engagement often comes at a personal price. The demanding nature of the profession was seen as both a source of pride and strain, as an emergency physician said:

Within this context, AI was perceived as a potential support for gaining more time for professional commitment during working hours. Physicians hoped that AI could reduce repetitive documentation tasks and administrative overload, allowing more space for genuine human interaction with patients. While many physicians viewed AI as a potential enabler of renewed patient engagement, some also warned that poorly designed systems could unintentionally distance them from patients, for instance, when screen interactions or automated reports take precedence over direct dialogue.

Physicians described ethical orientation as a fundamental and nonnegotiable component of craftsmanship, characterized by acting in the interest of patients, colleagues, and society rather than oneself. While Brinkman conceptualizes ethical orientation as an inner moral disposition, physicians in this study framed it more as ethically responsible action in practice, an expectation that applies universally within the profession. As one rheumatologist noted:

In the context of AI, physicians expected ethical judgment to become increasingly important as more cognitive and procedural tasks shift toward automated systems. Participants raised concerns about responsibility and accountability when decisions are influenced by AI, as well as uncertainty regarding the transparency and integrity of training data.

Several participants emphasized that ethical reflection should be explicitly incorporated into AI literacy and medical education.

In Brinkman’s framework, calling refers to an inner drive or vocation that motivates professionals to commit themselves to doing good work. It concerns intrinsic motivation rather than formal requirements or organizational conditions. Across interviews, calling was rarely articulated as a distinct element of craftsmanship. Physicians typically described their motivation in pragmatic terms, interest in the field, educational and career opportunities, and treated it as a taken-for-granted part of being a doctor rather than something they reflect on explicitly.

Unlike other dimensions, calling was not linked to AI by participants.

The focus group discussions expanded on the interview findings by exploring how AI can be responsibly designed and embedded in medical practice to strengthen craftsmanship. The following design principles were derived: consider a business case and strategic rationale; start from real clinical needs; let professional groups take the lead; design for contextual diversity; use user research for validation; design AI as supportive, not intrusive; safeguard autonomy and trust; cocreate with end users; learn across contexts; and use AI as a mirror for craftsmanship. To support practical use, the design principles were synthesized into an iterative flowchart (Figure 3). The sections below describe the design principles and their position within the flowchart.

The first step concerns identifying a real and clinically relevant problem that requires attention in daily work. Participants emphasized that AI design should always begin with a problem that physicians themselves recognize, rather than with a technology-driven initiative. Relevant design principles included:

This step involves examining whether the identified problem is genuine, widespread, and sufficiently significant to merit an AI solution. Participants stressed the importance of validating the problem across different roles, specialties, and hospital types. Relevant design principles included:

Once the problem is validated, stakeholders can explore whether AI is indeed the most suitable solution, or whether alternative approaches may be more appropriate. This stage requires a multidisciplinary workgroup to design and shape the solution. Relevant design principles included:

This step involves introducing the solution into practice, evaluating its fit, and reflecting on its effect and workflow. This step is iterative, requiring pilots, adjustments, and continuous learning. The relevant design principle is as follows:

This step concerns evaluating how the AI application influences the personal and professional dimensions of craftsmanship, both positive and negative. Participants emphasized that AI adoption should not only be assessed on performance or workflow efficiency, but also on how it shapes core clinical skills, judgment, and professional identity. The relevant design principle is as follows:

Physicians defined craftsmanship as a dynamic practice. A practice that requires knowledge, technical and social skills, and (moral) reflection. The division between professional and personal craftsmanship, as defined by Brinkman [19], was recognized: our health system has established a professional context to work, to which the physician should relate.

All elements of professional craftsmanship were acknowledged as important by the physicians. Yet, they consistently wanted to add a component: collaboration, referring to the ability to engage in multidisciplinary collaborations with other medical professions. Collaboration remains underarticulated in classical craftsmanship accounts. Sennett [16], Freidson [17], and Evetts [18] primarily conceptualize craftsmanship as an individual pursuit, offering little elaboration on collaborative or team-based dimensions. Scholarship on medical professionalism, a term closely related to craftsmanship, however, does address collaboration. It states that good care arises from more than individual competence alone, but from the interplay between professional judgment, values, and collaborative practice [23-25].

For the components of personal craftsmanship, physicians recognized themselves well in all but one: calling. Calling, understood as an intrinsic or vocational drive, was not mentioned or recognized as belonging to craftsmanship. Although calling is often defined in the literature as an intrinsic motivation that provides meaning to work [26,27], physicians in this study appeared to distinguish it from craftsmanship, which they primarily associated with skills, knowledge, and ethical standards. Calling was instead framed as a personal motivation or a sense of vocation that may develop over time, rather than as an inherent feature of their craftsmanship. There is literature suggesting that physicians often initially experience medicine as a career path and only later, during or after training, come to experience their work as a calling [26]. In this sense, calling appears to function more as a background driver of professional engagement rather than as a constitutive element of craftsmanship itself.

Taken together, these findings suggest that while the framework offers a useful analytical lens, its dimensions do not fully align with how physicians conceptualize their craftsmanship, as collaboration emerges as a central component and “calling” appears to function more as a background motivation than as a defining element, pointing to the need for contextual reinterpretation.

Overall, the craftsmanship framework served as a good reference point to discuss the influence of AI on medical practice. Most physicians referred to AI in broad, functional terms, frequently describing it as an “umbrella” or “container” concept encompassing a set of supportive and generative tools rather than a single, clearly defined technology. Given examples clustered into large language supportive tools, decision support applications, diagnostic applications, and administrative tools. Although no scientific literature is available on the current use of AI in Dutch hospitals, our results resemble a recent AI monitor, showcasing both generative AI and established clinical tools [28].

Where “dynamic” is already in physicians’ definitions of craftsmanship, AI shows the importance of the term. AI seems to directly influence the professional conditions under which physicians work: AI becomes a given in health care, requiring structural and organizational changes. Through its implementation, it might change the knowledge base needed to work and interact with AI, the ratio between cognitive and manual tasks, the autonomy that should be exercised, ways and needs for continuous development, and how professionals work together.

Through the reshaping of roles, physicians must relate to this new professional context, thereby indirectly influencing their personal craftsmanship. We encountered needs for redefining quality standards as AI becomes part of high-quality care delivery. Further, physicians should relate their knowledge and skills to AI: learning about AI might become part of the development of a professional, while learning tasks that AI has taken over might still be needed for when AI fails. The personal craftsmanship component of development and reflection stays in place, but again, AI might change how this is enacted as AI could be a mirror for action. Also, the need for autonomy seems to remain, but might become mediated by AI that creates a reliance of the physician on the model. Further, the involvement of a physician with their patients, colleagues, and profession seems to remain important, while being mediated in positive and negative ways. Finally, an ethical orientation seems to stay important and might even become more important since AI requires a new in-depth reflection itself.

So, our findings suggest that craftsmanship may evolve through the introduction of AI, requiring physicians to adapt to this changing context. Our results resemble recent literature, showing that changing working conditions require attention to ethical, organizational, and educational challenges [1,2,6,29,30]. By relating these insights to the concept of craftsmanship, an urgency to address them presents itself.

At the same time, the present study does not aim to fully explicate the mechanisms through which AI mediates the different dimensions of craftsmanship. Rather, the findings suggest that these influences are likely to be context-dependent, varying across medical specialties, organizational settings, and types of AI applications. For example, the way in which AI affects autonomy, knowledge development, or collaboration may differ substantially between clinical domains and technological use cases.