The first threat to the cultivation of experience relates to the current health care system and its effect on direct patient care in training programs. The increasing complexity of health care delivery, including crowded emergency departments, shorter patient lengths of stay, and staffing shortages, has forced trainees to devote substantial time to nonclinical tasks such as documentation, billing, insurance navigation, scheduling, and multiple handoffs. These factors erode a sense of clinical continuity [10], limiting the full picture of the patient’s disease and treatment.

The increased reliance and time spent engaging with the EHR also threatens time spent at the bedside and engaging in direct patient care. While the EHR certainly allows better centralization of medical records and standardization of care, its complexity comes at the cost of significant user time. A study in the ambulatory setting at one academic program conducted in 2017 to 2018 found that residents spent an average of 45.6 minutes in the EHR per patient, with 13.5% of that time spent after hours. While the residents showed improvement in efficiency with the EHR, there was no significant increase in patients seen per day or time spent in the EHR after hours [11]. A prospective, multicenter investigation of first-year internal medicine interns on clinical inpatient shifts using time-motion observations conducted in 2016 found that in a 24-hour period, a mean of only 3.0 hours was spent in direct patient care [12]. Concurrently, duty hour restrictions have resulted in fewer patient encounters across multiple specialties [13,14]. At our medical center and other academic centers with whom we have communicated informally, an internal medicine intern today admits a small fraction of the patients evaluated in the past. Over the course of 12 months at our institution, an intern in medicine admits fewer than one-quarter the number of patients admitted by an intern 30 years ago (internal data, 2026). While these data remain preliminary, institution specific, and not yet generalizable, the drop we observe warrants further investigation. There is growing concern as well about the clinical competence of surgical residents, who are not participating in a sufficient number of operative cases and/or having sufficient autonomy to be able to perform independently [15,16]. In a national survey in general surgery, program directors reported that trainees were less prepared than 10 years prior, citing duty hours and increased supervision among factors diminishing autonomy [17].

The impact of suboptimal experience has coincided with declining board examination pass rates across multiple specialties, including internal medicine and emergency medicine [18,19]. In emergency medicine, the reason for this decline is likely multifactorial, including the recent proliferation of new programs affiliated with for-profit institutions as well as the decrease in new patient volume due to the COVID-19 pandemic and boarding [20]. The decline in internal medicine board examination pass rates coincided with the COVID-19 pandemic, during which trainee experience and encounters were limited, and has since plateaued. While in both specialties these correlations are multifactorial and associated with pandemic-era and program-level factors, the educational effects of reduced patient exposure warrant further investigation.

Furthermore, when one adds the discontinuity of care associated with increased handoffs due to restricted duty hours, concerns about patient safety arise [21,22]. In 2 national cluster randomized trials of surgery and internal medicine residents, however, flexible hours were found to be noninferior with respect to death or serious complications compared to standard or restricted work hours [23,24]. Furthermore, program directors in the internal medicine trial were less likely to report dissatisfaction with elements such as intern ownership of patient care, time for reflection, real-time feedback from attendings, frequency of handoffs, teaching opportunities, and continuity of care in the flexible hour model. The amount of time trainees spent in direct patient care and educational activities, however, did not significantly differ between the groups [25]. We recognize the well-intentioned desire to improve the wellness of trainees and concern for fatigue-related errors, and the implementation of duty hour restrictions certainly puts these aims at the forefront. Whether its benefits support the resulting effects on education, continuity, and time spent on patient care remains unclear. It is one thing to be deemed competent based on interactions with a 25-year-old patient with asthma and quite another to subsequently be confronted with a 75-year-old patient with asthma and a history of congestive heart failure and coronary artery disease. Nevertheless, reduced clinical experience, if not compensated for in other ways, threatens the development of true competence, let alone expertise in patient care. In the end, the recognition of one’s shortcomings upon graduating from a residency program in which adequate clinical experience was lacking may compromise one’s sense of accomplishment, which is essential for personal well-being [26].

With the increasing complexity of the health care system, program directors and educational leaders need to advocate for the prioritization of staffing models that minimize trainee time spent performing nonclinical duties such as patient scheduling and similar administrative tasks that do not require physician oversight. This may allow for more opportunities for patient contact and clinical learning experiences during the workday. Furthermore, with the tremendous expansion in AI capabilities, educators can also leverage its benefits for certain time-consuming tasks that do not require the cognitive abilities of physicians or threaten the acquisition of critical thinking skills. There is evidence to show that using AI for medication reconciliation, for example, can decrease the administrative burden of clinicians [27] and, thus, allow more time for patient-facing clinical experience.

In light of duty hour limitations, institutions and frontline educators should reprioritize patient exposure as they structure teaching rounds and workflow, ensuring that time in the clinical space is patient facing. One strategy for this includes reincorporating bedside rounds, which have seen a significant decrease [28], to allow for more patient-facing interactions and exposure to clinical competencies such as physical examination techniques and effective communication skills [29]. Bedside rounds have been shown to be preferred by patients [30], and bedside teaching has had a positive effect on education, including the ability of preceptors to provide feedback [31]. Re-examination of staffing models that introduce discontinuity of care through handoffs, such as admission-only services, may also be warranted.

A second threat to trainees’ experience involves the expansion of the EHR and digital technologies, including AI. Increasing reliance on templated notes and “copy/paste” practices designed to maximize hospital revenue have made it easier to go through the motions of patient care without engaging in novel clinical reasoning [32]. We recognize that protocols embedded in the EHR can improve care, particularly for routine procedures and practices. Nevertheless, it is critical for trainees to gain experience thinking through clinical problems without relying excessively on these automated management aids. This will allow them to recognize when deviation from protocols is needed.

Similarly, AI has quickly become ubiquitous in clinical care. While AI can certainly improve efficiency for low–cognitive load, time-consuming tasks as noted above and serve as a beneficial learning resource, medical students and residents are using large language models across a variety of high–cognitive load tasks such as development of differential diagnosis, management plans, and documentation [33,34]. Lack of safeguards around the use of AI puts clinicians at risk of deskilling (degradation of established competencies in already trained clinicians), misskilling (acquisition of incorrect clinical reasoning patterns through uncritical adoption of erroneous or biased AI outputs), and never-skilling (failure to develop the foundational reasoning competencies during formative training). The effect of AI on deskilling or, potentially worse, never-skilling in the learner population remains theoretical and warrants investigation [35]. Without ever having developed these clinical reasoning skills, students and trainees risk not being able to recognize hallucinations, omissions, and faulty logic in AI output, nor will they be able to develop adequate prompts to maximize the output of AI, which could compromise patient care. The reduction in clinical experience and the confidence to use well-developed clinical and reasoning skills only worsens this problem. It is one thing to use AI as a tool to check on the completeness of one’s assessment and/or treatment plan. It is another, however, to be so dependent on AI that a physician cannot function without it.

Given the rapidly expanding digital landscape and unlimited access to generative AI tools, educators must recognize potential negative downstream effects on their learners. Training programs must prioritize faculty development initiatives focused on teaching critical thinking skills to trainees in the age of the rapidly expanding EHR, readily available protocols, and explosion of AI tools for answering clinical questions. Frameworks for this have been developed, including DEFT-AI (diagnosis, evidence feedback, teaching, and recommendation for AI use) [36]. Requiring students and residents to truly write their notes (as opposed to templates and carrying forward the prior notes) and present without reading directly from a prior note may improve assessment of their reasoning skills and should also be considered [37]. Institutions should develop and implement policies surrounding the use of AI in students and trainees and ensure AI-free portions of the program [35] so clinical reasoning skills can be cultivated. The Association of American Medical Colleges has published principles for the use of AI in medical education, outlining the need for maintaining a human-centered focus, the need for ethical considerations, and the need for evaluation of how AI tools affect educational outcomes [38]. Well-designed, prospective studies investigating the efficacy and educational effect of this are needed.

A third change bearing scrutiny is the flawed or incomplete implementation of CBME in many training programs. CBME, if not implemented appropriately, can undermine trainees’ acquisition of experience. Theoretical justifications for CBME are compelling and include the desire for clear learning outcomes, personalized training paths, and clear societal accountability. CBME correctly rejects the assumption that time spent in training alone guarantees readiness to provide appropriate care, relying on the development of “competency” as defined by tools such as milestones or entrustable professional activities (EPAs) [39-42]. CBME aims to ensure that all graduates possess the necessary knowledge, skills, and behaviors as demonstrated through measurable outcomes, ultimately to improve patient care and accountability to society [42]. CBME as designed, however, often varies from CBME as commonly enacted in many programs. For instance, outcome markers such as the EPAs include the ability to perform a physical examination but do not necessarily confirm that the learner can detect abnormalities or nuances that come with time and meaningful experience. With limited numbers of patient encounters and reliance on a CBME rubric alone, the learner may never have exposure to a patient with particular anomalies or one with multiple problems concurrently.

Despite its conceptual appeal, empirical evidence supporting CBME’s efficacy remains mixed. Over two-thirds of published articles about CBME are commentaries rather than empirical research [43], and a recent review found only inconsistent empirical support for CBME’s foundational assumptions [44]. A recent retrospective analysis of patient outcomes after endovascular aneurysm repair found that lower milestone ratings of vascular surgeons during their surgical training were associated with major complications in early surgical practice [45]. Similarly, in primary care, a retrospective study of practicing physicians who trained from 2015 to 2019 found that those who scored higher on professionalism and interpersonal skills and communication milestones in the previous 6 months of residency received better patient experience ratings in their first year of unsupervised practice on the Consumer Assessment of Healthcare Providers and Systems Clinician and Group Survey [46]. On the other hand, a retrospective analysis of a cohort of over 6000 hospitalists who completed training in 2016 to 2018 and were treating Medicare fee-for-service beneficiaries found that board examination scores correlated with lower 7-day patient mortality and readmission rates. Their milestone ratings, however, showed no clear association with 7-day mortality and readmission rates, 30-day mortality and readmission rates, length of stay, and subspecialist consultation frequency [47]. It is important to note that these studies were all correlational and not designed to test all formative or readiness uses of milestones and that the milestones were designed to describe the progressive level of performance of a trainee. Nevertheless, the data raise concerns about the present use of CBME. Prospective investigations are needed to determine whether the milestone ratings alone or combined with both the quantitative and qualitative types of experience that trainees have can truly gauge readiness for independent practices and patient outcomes.

While CBME in principle emphasizes robust competency assessment across a range of contexts, CBME in practice often devolves into resource-intensive administrative exercises in which clinical competency committees ensure that trainees receive the necessary approval to advance at the expected pace [48]. EPA levels are designed to be determined by observable behaviors and provide a forward-looking assessment of generic clinical skills; it is unclear that they predict the individual’s readiness to respond to a more complex and unknown instance in the future. They do not account for the entirety of the actual clinical environment and real-world practice [49,50]. Additionally, the underlying rationale behind “entrustability” parallels the framework by Dreyfus [8] of novice to expert [40], further necessitating the need to consider experience as a critical factor on this scale. Competence is a necessary but not sufficient goal for graduates of a GME program.

Furthermore, the assumption that meeting the competencies, as defined by these instruments as a listing of the basic skills of a physician, guarantees competence in a novel situation is misguided; if a medical resident has never seen a patient with status asthmaticus or acute pulmonary edema, can they be deemed competent upon graduation? If a general surgical resident has never performed a hernia repair, are they competent? While case logs are required in surgery and time requirements spent in different venues are defined in nonprocedural fields, the role of the trainee caring for those patients and the exact disease states are not specified. To be sure, no trainee will have exposure to every disease or possible procedure in their training, but a competency-based model that does not explicitly incorporate clinical experience may obscure the loss of training in core areas. The implementation of CBME alone cannot substitute for a system that ensures sufficient trainee experience across relevant practice domains.

With CBME now standard across training programs, clerkship and program directors need to ensure that determinations of competence truly represent a learner’s stage on the path to expertise. Dedicated training on how these designations should be made along with accounting for experience may provide more accurate predictive assessments of abilities. For example, GME programs can revamp clinical competency committees to ensure that milestone ratings promote learner development and readiness for unsupervised practice [51] with consideration of case logs and exposure to certain disease states and presentations. This could include defining minimum numbers of encounters or procedures, a case mix or variation index, redundancy criteria, continuity metrics, encounter definitions, and review intervals. In 2025, the Review Committee for Emergency Medicine recommended for the first time a minimum number of emergency department patient encounters for completion of residency. It also advocated for extending training to 48 months across all programs to better accommodate today’s health care and educational environment [52]. While the latter recommendation comes with significant costs, including financial considerations, equity issues in reduced training environments, and trainee opportunity costs, that must be weighed against its potential benefits, it is an albeit controversial possible solution to address insufficient experience. We encourage other medical specialties and accrediting bodies to follow suit in explicitly defining numbers and ranges of various clinical experiences.