(Guest Author: Chat GPT 6, using Sol 6.1 High.)
We previously did a deep dive article on Socarraz & Lennerz, a vision for fundamental changes in pathology (here).
Today, we're revisiting Socarraz & Lennerz but in a new context: it joins a triangle populated by pathologists, oncologists, and radiologists, using very recent articles predicting their winds of change.
##
Three Specialties, One Transformation: How AI Is Recasting Pathology, Radiology, and Oncology
A Triangle of Changing Professional Roles
Across pathology, radiology, and oncology, a common question is taking shape: what should physicians do when AI can perform increasing portions of the work through which they traditionally demonstrated expertise? Six publications offer a revealing triangle of answers. De Socarraz and Lennerz envision pathologists as diagnostic architects, with Vos and colleagues specifying new educational requirements. Moy and colleagues propose a transformation of imaging services and the radiologist’s role. For oncology, Mullangi and colleagues describe an augmented clinician working within a redesigned decision environment; Patel and colleagues propose a collaborative human–AI model of cancer care; and the ASCO Post’s coverage of Peter Clardy examines how AI changes expertise, clinical encounters, and patients’ access to knowledge.
Together, these articles reach well beyond adding software to existing practice. They ask how physicians should allocate attention, exercise judgment, organize care, and demonstrate value when generating medical information becomes increasingly automated.
Pathology:
The Diagnostic Architect and the Steward of Reliability
In Adapt or Become Irrelevant: the Pathologist as the New Diagnostic Architect, Mariano De Socarraz and Jochen Lennerz argue that pathology’s capabilities have outgrown its traditional organizational position. Their seven examples encompass biomarker interpretation, genomic reimbursement, AI triage, integrated diagnostics, pharmacogenomics, clinical trial design, and emerging quality-assurance structures. The recurring problem is that producing a technically valid result does not ensure that the result contributes effectively to care. PD-L1 can become an oversimplified checkbox; molecular testing can encounter fragmented ordering and payment arrangements; pharmacogenomic information can remain unused because nobody builds a dependable implementation pathway. Their proposed diagnostic architect synthesizes morphology, molecular findings, imaging, and clinical context into a coherent assessment, while the diagnostic steward oversees the continuing reliability of assays, algorithms, and workflows. Shoko Vos and colleagues provide an educational counterpart in Making Pathologists Ready for the New Artificial Intelligence Era: Changes in Required Competencies, published in Modern Pathology in 2025. Their proposed entrustable professional activity covers evaluating AI tools, interpreting outputs for individual patients, handling disagreement, and recognizing personal and technological limitations. Taken together, these papers reposition pathology around diagnostic synthesis, implementation, and governance. The pathologist’s professional contribution expands from interpreting what arrives in the laboratory to helping determine how diagnostic knowledge is generated, connected, and used.
Radiology:
Selective Automation, Greater Capacity, and a Wider Clinical Reach
Linda Moy and colleagues’ Call to Action: Accelerating AI-driven Transformation in Medical Imaging and the Broader Health Care System begins with a capacity problem: imaging demand exceeds the resources available to meet it. The International Society for Strategic Studies in Radiology consensus statement argues that meaningful productivity gains require AI to replace selected radiologist tasks. Software that merely adds another set of prompts to review may leave the underlying workload largely intact. The proposed changes span the imaging service: identifying examinations without actionable disease, recognizing longitudinal stability, shortening acquisition, tailoring protocols, improving scheduling, and shifting report production toward physician editing of AI drafts. The authors distinguish these ambitions from current readiness, retaining caution about autonomous interpretation, particularly for complex imaging. Their vision also expands radiology’s clinical reach. Scans obtained for one indication could yield additional information about bone density, vascular disease, or body composition; imaging departments could integrate those findings with clinical and molecular data to support prediction and prevention. Thus, fewer manual steps could accompany broader clinical responsibilities. Radiologists would increasingly oversee the quality, relevance, and consequences of information extracted from images while helping redesign how imaging contributes to care. The central test is whether this transformation delivers better access, more useful decisions, and sufficient operational or clinical value to sustain investment.
Oncology:
Designing Better Decisions and Leading Human–AI Care
The oncology publications describe an equally substantial transformation, encompassing treatment selection, teamwork, and the therapeutic relationship. In AI to Support Modern Cancer Care—The Augmented Oncologist, Samyukta Mullangi, Kanan Shah, and Debra Patt argue that conventional decision-support architecture has not kept pace with therapeutic complexity. They propose moving from fragmented, largely static guidance toward interactive systems that combine evidence with a patient’s genomics, treatment history, comorbidities, and other circumstances. AI could present tailored risk–benefit comparisons, clarify ambiguous questions, and eventually support simulations of alternative treatment sequences. Crucially, the arrangement of options itself influences decisions, making choice architecture part of clinical practice. Milit Patel, Edward Christopher Dee, and Nancy Lee extend the vision in AI and Human Expertise in Cancer Care—Striving for Synergy, published online in November 2025 and in the February 2026 issue of Nature Reviews Clinical Oncology. They propose multidisciplinary human–AI teams in which computational synthesis supports contextual judgment, creative problem-solving, ethical reasoning, and continuing relationships with patients. Julia Cipriano’s ASCO Post account of Peter Clardy’s discussion adds another dimension: AI can participate in encounters and patients can use it independently, changing both the information balance and the meaning of expertise. Together, these perspectives envision oncologists evaluating evidence, arbitrating competing recommendations, coordinating teams, and helping patients make consequential choices. The transformation concerns the entire decision process, including how options are generated, presented, discussed, and revised over time.
What the Three Corners Reveal
A Shared Movement Toward Integration and Judgment
The three specialties approach AI from different starting points, but converge on a similar professional trajectory. Pathology emphasizes coherent biological interpretation and trustworthy diagnostic systems. Radiology emphasizes capacity, information extraction, and longitudinal assessment. Oncology emphasizes individualized treatment decisions and coordinated care. In each, automating individual outputs increases the importance of deciding whether those outputs fit together and deserve action.
This resembles convergent evolution: different pressures produce a shared emphasis on integration, judgment, and oversight. Yet important distinctions remain. Pathologists must assess whether the specimen, assay, and biological interpretation support a conclusion. Radiologists must assess what images establish and how findings change. Oncologists must weigh those conclusions against treatment benefit, toxicity, timing, and patient priorities. These responsibilities overlap, but the consequences each specialty must arbitrate differ.
Three Claims to Integration Require Shared Accountability
Both De Socarraz–Lennerz and Moy position their specialties to lead multimodal diagnostic integration. The oncology papers make clear that integration also occurs at the point of treatment choice, where diagnostic information meets competing interventions and patient preferences. All three claims are credible within their domains. Together, however, they expose a practical question: how should leadership be distributed across the care pathway?
The implication is that integrated care requires explicit arrangements for resolving discordance, obtaining missing information, communicating uncertainty, and ensuring follow-through. An imaging finding, molecular result, and AI-generated treatment recommendation might each be reasonable in isolation yet imply different actions. The value of the team lies partly in resolving those differences. A collection of sophisticated outputs becomes integrated care only when someone takes responsibility for the connections among them.
Task Replacement and Human Augmentation Operate at Different Levels
The apparent contrast between Moy’s emphasis on replacing radiologist tasks and Patel’s insistence on preserving human expertise is instructive. These positions address different levels of practice. A system can replace report drafting, routine measurement, or literature retrieval while strengthening the physician’s ability to exercise judgment. Conversely, retaining a physician’s approval at the end of a workflow does not establish that the workflow improves care.
Patel’s call for comparative studies of human–AI teams therefore strengthens the entire triangle. Agreement with an expert recommendation, diagnostic accuracy, and minutes saved each capture only part of the benefit. Evaluation should also ask whether treatment becomes more appropriate, follow-up more reliable, access more equitable, and patients better informed. The relevant unit of evaluation is increasingly the care process and its outcomes, alongside the performance of the algorithm.
Authority Can Move Into the Software Before Anyone Notices
Mullangi’s emphasis on choice architecture adds a particularly important insight. A system can preserve formal physician autonomy while influencing decisions through which options it highlights, ranks, or leaves less visible. As an analytical implication, clinical authority can shift through interface design even when the physician retains the final signature.
That concern connects directly with De Socarraz and Lennerz’s warning about diagnostic oversight moving into external assurance structures, and Moy’s emphasis on local validation and continuing surveillance. Governance must address both the accuracy of an output and the decisions embedded in the system: what counts as actionable, which evidence receives priority, and who can challenge or change the process. Clinicians need meaningful influence over those choices to fulfill their expanded responsibilities.
The Economics and Education Must Support the New Roles
All three visions depend on how organizations use the time released by automation. It could support difficult consultations, diagnostic stewardship, multidisciplinary discussion, and shared decision-making. It could also be absorbed entirely by higher throughput. Pathology’s reimbursement concerns and radiology’s focus on operational returns expose different parts of the same problem: savings are often easier to recognize than the value of the new work. Broader responsibilities require time, authority, and resources.
Education presents a parallel challenge. Vos specifies new competencies while preserving foundational ones; Clardy warns about clinicians who lose skills or never acquire them. Reviewing machine-generated work may require expertise historically developed by performing the underlying tasks repeatedly. Training therefore has to provide deliberate opportunities to reason independently, identify errors, and explain disagreements. These publications offer proposals and professional visions with differing evidence bases; realizing them will require testing both the redesigned care models and the preparation of the physicians expected to lead them.
Sidebar:Seven Surprising Needs and Contrasts
1. Oncology’s transformation includes the architecture of choice. Mullangi and colleagues emphasize that the order, visibility, and presentation of treatment options influence decisions. Future oncologists may need to scrutinize how a recommendation system frames a choice as carefully as they scrutinize the evidence behind it.
2. Radiology seeks to borrow laboratory governance while pathology worries about losing influence over it. Moy proposes a CLIA-like quality framework for imaging AI. De Socarraz and Lennerz warn that emerging assurance systems could bypass pathology’s accumulated expertise. The two perspectives suggest an opportunity for collaboration over validation and continuing monitoring.
3. Recognizing stability may create more value than finding another abnormality. Radiology’s emphasis on identifying unchanged disease and examinations without actionable findings challenges an innovation culture oriented toward detecting more. Reliable reassurance can release scarce attention for consequential cases.
4. More automation may require more training. Vos and colleagues suggest reconsidering a Dutch trend toward shorter pathology residency, or promoting earlier subspecialization, because AI adds competencies while established skills remain necessary. Clardy’s concern about never acquiring foundational skills makes this a broader medical-education problem.
5. The oncology sources expose two different meanings of empathy. Patel emphasizes sustained, authentic human relationships; Clardy discusses research in which AI sometimes performs well on assessed empathetic communication. These claims need not conflict. A reassuring response in an encounter and a relationship sustained through illness are different achievements, and evaluation should distinguish them.
6. The patient may acquire AI support before the care team does. Clardy’s clinician–patient–AI relationship means that institutional adoption is only part of the transformation. Oncologists may already be discussing interpretations and recommendations generated outside their own clinical systems, adding a new responsibility to explain their agreement or disagreement.
7. Human oversight is a design problem, not a guarantee of benefit. Read together, the papers suggest that retaining a physician in the workflow is insufficient by itself. Oversight requires competence, usable evidence, time to examine exceptions, and authority to intervene. Whether the arrangement improves care must still be demonstrated.
Source PDFs:
References
De Socarraz M, Lennerz JK. Adapt or Become Irrelevant: the Pathologist as the New Diagnostic Architect. Journal of Clinical Pathology. Published online August 17, 2026.
https://doi.org/10.1136/jcp-2026-210826Vos S, Hebeda K, Milota M, et al. Making Pathologists Ready for the New Artificial Intelligence Era: Changes in Required Competencies. Modern Pathology. 2025;38:100657.
https://doi.org/10.1016/j.modpat.2024.100657Moy L, Vargas A, Gichoya JW, et al. Call to Action: Accelerating AI-driven Transformation in Medical Imaging and the Broader Health Care System. Radiology. 2026;320(2):e253974.
https://doi.org/10.1148/radiol.253974Mullangi S, Shah K, Patt D. AI to Support Modern Cancer Care—The Augmented Oncologist. JAMA Oncology. 2025;11(11):1281–1282.
https://doi.org/10.1001/jamaoncol.2025.2888Patel MS, Dee EC, Lee NY. AI and Human Expertise in Cancer Care—Striving for Synergy. Nature Reviews Clinical Oncology. 2026;23:87–88. Published online November 25, 2025.
https://doi.org/10.1038/s41571-025-01108-9Cipriano J. As AI Advances, How Will the Clinician’s Role Change? The ASCO Post. September 25, 2026. Coverage of Peter Clardy’s presentation.
https://ascopost.com/issues/september-25-2026/as-ai-advances-how-will-the-clinician-s-role-change/
