“Hard Fork” for Family Medicine - Artificial Intelligence Will Change the Way We Experience Practice

• Artificial Intelligence
• Continuity of Care
• Doctor-Patient Relations
• Family Medicine
• Family Physicians
• Organizational Change
• Population Health
• Primary Health Care

A “hard fork” is a programmer/developer term for a change that is not backward compatible and requires all users to upgrade their software in order to continue participating in the network. In a hard fork, the network splits into two separate versions: one that follows the new rules and one that follows the old rules. Family Medicine faces such a hard fork driven by the accelerating advance of artificial intelligence (AI) in medicine, particularly the recent advent of Large Language Models (LLMs),¹ that will affect the future of practice in many ways. The benefits, risks and challenges to successful AI deployment have been discussed extensively for the past decade.^2–6 Many impacts may be salutary, especially if AI can decrease mindless “WAC” (work after clinic – aka pajama time) through ambient scribing assistance with clinical note writing, review and summarization of medical records, LLM-assisted electronic routine patient message answering systems, automated refill programs and the like.^7–10 These applications reduce clerical, routinized, “non-MD” level tasks that waste precious time and add to burnout, while leaving more time for the human connection and empathic components of the continuous patient-doctor relationships that are central to family medicine. But others, though beneficial, may infringe on the real intellectual medical work of our specialty. If AI can either independently complete or assist with taking histories, managing chronic problems, monitoring health maintenance needs, assessing risk, diagnosing and prescribing usual treatments for common patient complaints, considering unusual diagnoses for complicated presentations, and augmenting and improving treatment for mental health conditions^11–17 … it becomes more pressing now to ask the question: what will the practicing human family physician be expected to do in this new AI-assisted world?

When we pose this question to colleagues in the context of what future doctor visits will look like, the initial response has always been something rather vaguely reassuring like: “Doctors will still be needed for human connection and really complex management challenges…” While we understand what such encounters are, we should be talking more about what the reality of this change would actually look like. If family medicine, at its core, is about establishing long-term continuous healing relationships with patients, then we need to envision what future human encounters will look like in this new environment, and how many of them will be needed for an average panel of patients. This is a challenging and revealing exercise in future practice visioning. Let us first take a moment to describe a few potential patient care situations that might still require a human physician in a future AI-augmented world:

1. Some patients are going to want to talk with a human (their own) physician for a number of reasons: to check to see if the AI “interface” was correctly interpreting their data and arriving at correct assessments and recommendations; to explore an emotional/psychological issue before being referred either to a human integrated behavioral health clinician or an on-line mental health app; or because they just want the “personal touch” of their doctor to talk with because they’re lonely or unsure of where to turn or what to do. As humans have done for millennia.

2. Some patients may never want to interact with AI at all. These individuals despise our phone trees and complain about having to talk with three other staff before they get to talk with their doctor. Some may even have “medical PTSD” from past significantly negative responses to interactions with the medical industrial complex. They demand human connection and it has to be respectful and caring to be heard and accepted. Such patients require the slow building of trust between doctor and patient – a human endeavor.

3. Patients who need to or are trying to change their health behaviors where human-delivered motivational interviewing and supportive techniques might be more effective than an app – because of their longstanding relationship with their doctor.

4. Patients with serious medical conditions further complicated by psychological/behavioral issues or significant social needs that compound their situations. These are patients with whom we already spend substantial amounts of care time. They are not likely to be managed solely by the rest of the team even enhanced with AI.

5. Patients with family issues that impact their health, well-being or even safety. Such patients may benefit from the personal touch of a physician who knows them and their family, whom they trust to help them work through such problems, perhaps along with integrated behavioral health colleagues. AI is years away from handling these kinds of complex interpersonal challenges.

With these persisting human patient needs and assuming AI overcomes still significant practical implementation and dissemination barriers (accuracy, safety, potential algorithmic bias, financial, regulatory, privacy, operational and clinician/patient acceptance), we can perhaps picture what a future AI-enabled practice would look like. While current uptake of AI applications is still generally low in practice,¹⁸ future family physicians will engage with them in increasing numbers and shift from doing many prior lower-level daily tasks by themselves to reviewing the output of AI systems, evaluating their accuracy and appropriateness and deciding how much of their assistance to use. This physician transition will likely rest on the quality of the specific Al tools, their experience using them and the acuity and risk of the situation. This new fully integrated “AI partner”¹⁹ could help all members of the multi-professional practice team improve their efficiency and effectiveness in carrying out their tasks as well as potentially allowing many patients to self-manage with healthcare team-supervised AI advice alone. AI could also help assess whole practice populations for healthcare needs,²⁰ assist individuals in self-management of common complaints, take on many of today’s routine care tasks, and direct the focus of the family doctors and their teams to the patients they need to see and spend time with. If all this comes to pass, the above-noted efficiencies and off-loaded work could provide the capacity for hundreds or more additional patients to be added to each practice’s panel and reduce the 100 million Americans lacking primary care.²¹

AI-assisted healthcare will require intense re-training of all members of the healthcare team as well as patients themselves. The latter will need to learn the limits and risks of stand-alone medical chatbots that claim to offer on-demand, time-efficient medical or even mental health advice. They will also need to both accept and learn how to interact with the new AI partner in their family physician’s office. As we alluded to above, physicians must learn new competencies in how to evaluate and work with AI, such as regularly reviewing AI’s input from its own patient interactions or “panel scanning”, checking to see if AI’s decision-making “judgement” is correct,²² and determining how much of what kind of human physician input is needed. In short, they will need to learn to operate in a new triad: the “doctor-patient-AI partner relationship”. The nature and composition of the average day’s work hours would need to change to allow time for these doctor-AI partner interactions which are essentially a new “in-basket” task. This cannot be relegated to WAC as is often the case with current in-basket work. If this doesn’t happen or if all the efficiencies gained through AI just result in more assigned work instead of reducing pajama time, new family doctors will likely not tolerate this situation or may choose to decrease their hours worked. Furthermore, face-to-face visit lengths for complex and intense encounters would on average be longer than the current 15-20 minute standard. The result should be that physician work days would be composed of fewer, longer in-person visits and more built-in time working with their AI partner.

There are other ramifications of this AI-enhanced practice that need to be considered. Having to deal with a higher concentration of cognitively burdensome or emotionally draining complex patient interactions might present an additional challenge.¹⁹ Most physicians appreciate “easier” or “simpler” visits distributed throughout their sessions. Such routine visits create added opportunities to establish or deepen continuity relationships around aspects of prevention, social support and health behavior change. If many of these are managed by the team and AI partner, will continuity be interrupted and such relationships fail to form? Will the resultant emotional workload from a higher load of complex cases be too much? Will the profession be more or less attractive to student candidates? Some commentators have already expressed concerns about the decreased emphasis on continuity care during residency and consequently in resident graduates’ career choices, which seem to be gravitating toward urgent care, hospital medicine and sports medicine – areas that do NOT have the same burdens of continuity care.^23,24

Conversely, this new AI movement could contribute to improved future family physicians’ practice lives. Accepting and meeting the challenge of managing a whole panel or population of patients might provide a rewarding “public health” aspect to family practice. Balancing the emotionally taxing, complex longitudinal work with this type of effort or other “one-and-done” work that is enjoyable and fulfilling (e.g., office procedures, sports medicine, a specialty area for first-line advice in a group practice, school health, free clinic work, etc.) could improve career satisfaction and mitigate burnout. Routinely encouraging such diversity in clinical practice careers, while maintaining the major focus and concentration on continuity care, could help attract and retain future family physicians.

To prepare for this more fully built-out AI-enhanced future, family medicine practitioners, educators, residency directors, researchers and policymakers must engage in active dialogue and strategic planning and be involved in the entire development process for clinical AI that will be used in family medicine. In addition to the many needed action steps for AI adoption (revising residency training, advocating for supportive regulatory policies, addressing financial and operational barriers, and ensuring ethical implementation), family medicine professional organizations should pilot several larger-panel AI-enabled practices to assess their feasibility in the real world. This will require forecasting which AI-assisted functions will likely be widely disseminated (ambient scribing, message answering, chart summarization and healthcare gap analysis), how often others that assist intellectual work (risk appraisal, clinical decision support and counseling) will be used and by which healthcare team members, and how they would all interact and function together in one practice. This operational feat will require far greater skill and coordination than adopting any one AI solution, and may result in temporary increases in work load until efficiencies of use are achieved.¹⁹ In the setting of pilot practices, detailed aspects of the patient-doctor-AI partner relationship could be worked out so that the human-requiring tasks stay firmly with the family physician. Such pilot practices could be fielded by primary care AI innovation hubs like the Stanford Healthcare AI Applied Research Team (HEA₃RT)^25,26or the newly proposed Family Medicine AI Centers of Excellence.²⁷

Failure is possible on many fronts (AI functional quality and performance, patient and healthcare team acceptance, cost and ROI and legal/ethical standards to name a few), but we believe that family medicine is in fact the most likely component of the healthcare system to both survive and prosper with AI. Generalists frequently have to manage uncertainty and multimorbidity, integrate varied forms and sources of information, meld the science and human elements of the art and practice of medicine and help patients navigate a complex and information-rich healthcare system.²⁸ With their continuity connections to patients, and whole-person and community focus, family physicians are uniquely suited to lead in population health management including incorporation of social determinants of health and promotion of equity-oriented care. AI will make some routine jobs easier, quicker or even unnecessary, leaving more time for the essential human doctoring tasks to family physicians, as has been true during all past eras of technological advance. Wise adoption of AI could substantially improve the effectiveness and efficiency of family medicine, the health of all the populations it serves, and breathe new life into efforts to attract more future colleagues into family medicine careers. This result would constitute a successful navigation of the imminent AI-driven “hard fork” in family medicine.

• Received for publication July 2, 2025.
• Accepted for publication October 27, 2025.