From doing to being: Generative AI, workplace affordances, and professional identity

You’re listening to “From doing to being: Generative AI, workplace affordances, and professional identity,” by M. Lavercombe. Published in 2026.

Medical Teacher

ISSN: 0142-159X (Print) 1466-187X (Online) Journal homepage: the linked source

Mark Lavercombe

To cite this article: Mark Lavercombe (28 Jul 2026): From doing to being: Generative AI, workplace affordances, and professional identity, Medical Teacher, DOI: 10.1080/0142159X.2026.2710213

© 2026 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group

Published online: 28 Jul 2026.

Submit your article to this journal

Article views: 590

View related articles View Crossmark data

COMMENTARY

From doing to being: Generative AI, workplace affordances, and professional identity

Mark Lavercombea,b aDepartment of Medical Education, The University of Melbourne, Melbourne, Australia; bDepartment of Respiratory & Sleep Disorders Medicine, Western Health, Melbourne, Australia

In the early years of implementing generative artificial intelligence (AI) in health professions education (HPE), our discourse has largely focused on the technical: the tools we adopt, the tasks we redistribute, and the frameworks through which we teach and assess. Khakpaki’s 2025 review organises the literature around technologies and tools, applications in teaching, and impacts on educational outcomes; Ning et al. similarly survey the delivery of training ‘through an AI lens’ and the challenges to realising AI’s potential. These works are valuable and necessary, but their centre of gravity is consistent: AI is treated as something we deploy to deliver content, assess performance, and improve efficiency.

The implications of these emerging technologies are identified and then addressed through a technical lens, what Habermas called the technical knowledge-constitutive interest: a concern with instrumental control and the efficient allocation of tasks. What this framing leaves unexamined is how AI reshapes the learning affordances of the clinical workplace.

Thankfully, our discourse has not remained entirely technical: a significant literature on the ethical implications of AI in health professions education has emerged. Masters’ AMEE Guide No. 158 maps a broad swathe of ethical considerations, including data collection, privacy, consent, ownership, security and algorithmic biases. His focus, however, is the relationships between institution, teacher and student; he largely excludes patient interactions as ‘best covered in medical ethics’. A call for a principles-based approach to AI education, grounded in medical and public health ethics, followed.

Zainal et al. extend that work, and their proposed Digital-Age Clinical AI Ethics Competence framework is genuinely novel; it emphasises ‘grounding in clinicians’ lived experiences’, situated capacities and adaptive professionalism. Cole’s related commentary argues that ‘AI clarifies the distinctly human work that defines educational expertise’ and reframes the question productively: not whether AI should replace faculty, but what forms of faculty work become indispensable when knowledge is no longer scarce. She rightly foregrounds the relational and formative dimensions of teaching, prioritising judgement, relationships and moral agency. Cole frames this as a redistribution of faculty work – AI handling information delivery and assessment, with faculty taking up the relational and formative domains.

These are precisely the capacities Habermas located in his two other knowledge-constitutive interests – the practical interest in mutual understanding through communicative action, and the emancipatory interest, concerned with critical self-reflection and emancipation from unexamined assumptions. Such ways of knowing develop through the work of acquiring and applying knowledge, not alongside it; they cannot be cleanly reallocated between machine and human. Masters, Zainal et al. and Cole all move past the technical framing, yet their focus remains formal education and the roles of educators. The argument here extends theirs further: these capacities are cultivated not only by faculty but also in the clinical workplace, an environment AI is already beginning to alter.

Cruess et al. describe professional identity formation as a process of socialisation in which a learner’s existing identity is progressively reshaped through engagement with the communities of practice in which they train. Their model is non-linear – ‘cyclical and self-reinforcing’ – and traces the learner’s movement ‘from “doing” to “being”’. These communities of practice – the settings, relationships, and encounters through which learners interact and are socialised – are part of the very workplace AI is already reconfiguring. Cruess et al. characterise identity formation as a constant renegotiation in response to changing conditions; an AI-mediated workplace does not merely add a condition but transforms the environment in which that renegotiation occurs.

Socialisation and identity formation, as shown by Billett’s workplace learning framework, depend on both the affordances workplaces offer and learners’ access to and engagement with those affordances. Stabel et al. identified challenging patient cases – including rare presentations, unexpected responses to treatment, complex comorbidity, or patient dissatisfaction – as the primary trigger for general practitioners to engage in further learning. It is the uncertainty in such cases that prompts engagement: difficult cases are discussed among colleagues, in team meetings, and during department handovers – what Billett and Noble call pedagogically rich work activities.

AI-mediated clinical practice does not merely redistribute educational tasks; it can reshape the affordances that shape identity and judgement. Automated medication reconciliation, with grading of potential drug interactions, might reduce the need to consult with a clinical pharmacist. AI screening of chest imaging may erode the rationale for a weekly ‘radiology meeting’ in which a chest radiologist and the respiratory team discuss unusual radiographic patterns or complex patient presentations, thereby enabling clinico-radiological correlation.

Although automation might free up time for richer discussion and its impacts are not yet clear, AI does not resolve all uncertainty in difficult cases; it relocates it. Instead of discussing a complex patient presentation and differential diagnosis with colleagues, the uncertainty shifts to: ‘Do I trust the diagnosis the system has made?’ Earlier tools supplied information for the clinician to weigh when making a difficult decision with no clear answer; generative AI increasingly supplies a judgement. In self-study, Ng et al. found that learners treat generative AI as an ‘unreliable collaborator’ and evaluate its output against prior knowledge, but ‘rarely reach higher levels of critical engagement’.

In the clinical workplace, the pedagogically rich work activities through which clinicians are socialised may be displaced, and the learning environment may become less rich, leaving fewer opportunities to engage in the ongoing renegotiation of identity that Cruess et al. describe.

The eventual impact of AI on clinical practice and associated workplace learning environments remains uncertain. New affordances and triggers for learning might still emerge, and the trade-offs we will have to make are unclear; however, it is urgent that we consider how AI-mediated work practices change the nature of practice and, therefore, physician identity. If the senior physician in the emergency department spends more time reviewing AI-generated alerts about deteriorating patients than reviewing actual patients, the physician’s role identity may change, and trainee role modelling with it. The impact on the clinical learning environment may be far-reaching, and this is the environment in which our trainees learn judgement, wrestle with uncertainty, and experience the interactions through which professional identity is formed.

These are not technical capacities, and they cannot be secured by technical means.

Clinical judgement, tolerance of uncertainty, and professional identity formation fall within Habermas’ practical and emancipatory interests. They are developed through active, iterative participation in the unpredictable clinical workplace, with all its vagaries, rather than being delivered to a learner via PowerPoint. Ng et al. call for training that cultivates ‘uncertainty tolerance’, yet AI-mediated practice has the potential to relocate the uncertainty in which it is learned. If our discourse focuses on the tools we adopt and the competencies we assess, we will polish the technical elements of medical education while its very foundations shift beneath us. Recognising the workplace environment as a site of formation allows us to identify an opportunity: AI implementation can be designed with learning in mind, enriching the conditions in which clinicians are made.

Our task becomes far more nuanced and far more interesting: to examine how AI is reshaping the clinical workplace itself – its affordances, its communities of practice, and the encounters through which a learner moves from doing to being.

Generative AI usage acknowledgement

During the preparation of this manuscript, the author used generative AI tools (Opus 4.8 with

Claude 1.14271.0 (c8f4d8); Anthropic PBC; Grammarly 1.166.1.0, Grammarly Inc.) to obtain critical feedback on the developing argument, to suggest structural and editorial revisions, to assist with refining the text, and to support the verification of references. No unpublished data, confidential information or copyrighted material was uploaded into any AI tool. The author has reviewed the terms of use of these tools and confirms their suitability for this purpose. All AI-generated suggestions were independently reviewed, fact-checked, edited and integrated by the author to ensure accuracy, completeness and alignment with the author’s own analysis and interpretation, and the author verified all citations. The author maintains full responsibility for the content of the manuscript, including the accuracy of all information and the integrity of the arguments presented.

No AI tool is listed as an author.

CRediT: Mark Lavercombe: Conceptualization, Writing – original draft, Writing – review & editing.

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes on contributor

Mark Lavercombe, MBBS, SpecCertClinLead, MClinEd, FRACP, FCCP, FAPSR, AFANZAHPE, AFAMEE, Department of Medical Education, The University of Melbourne, Melbourne, Victoria, Australia and Department of Respiratory & Sleep Disorders Medicine, Western Health, Melbourne, Victoria, Australia.

01. Khakpaki A.

Advancements in artificial intelligence transforming medical education: a comprehensive

02. Ning Y, Ong JCL, Cheng H, et al.

How can artificial intelligence transform the training of medical students and physicians? Lancet Digit Health. 2025; 7:100900. doi: 10.1016/j.landig.2025.100900

03. Habermas J.

Knowledge and human interests. Boston (MA): Beacon Press; 1971.

04. Masters K.

05. Weidener L, Fischer M.

Proposing a principle-based approach for teaching AI ethics in medical education. JMIR Med Educ. 2024;10: E 55368. doi: 10.2196/ 55368

06. Zainal H, Chuan VT, Xiaohui X, et al.

From principle to practice: developing a digital-age clinical artificial intelligence ethics competence framework through early-career doctors’ experiences. Med Educ. 2026: Medu.70195. doi: 10.1111/medu.70195

07. Cole R.

Is AI replacing faculty? Rethinking faculty roles in medical education. Med Educ. 2026: Medu.70220. doi: 10.1111/medu.70220

08. Kumagai AK.

09. Cruess RL, Cruess SR, Boudreau JD, et al.

10. Billett S.

Stabel LS, McGrath C, Bj€orck E, et al. Navigating 11. affordances for learning in clinical workplaces: a qualitative study of general practitioners’ continued professional development. Vocat Learn. 2022;15: 427–448. doi: 10.1007/s12186-022-09295-7

12. Billett S, Noble C.

Utilizing pedagogically rich work activities to promote professional learning. Educ Didactique. 2020;14-3:137–150. doi: 10.4000/educationdidactique.7943

13. Ng O, Lim S, Goh N, et al.

What are medical students really doing with GenAI in their self-study? An epistemic entanglement framework approach. Med Teach. 2026:1–9. doi: 10.1080/0142159X.2026. 2667280

Author contributions