Saving time, improving training: The right way to use AI in primary care

Primary care already has a pipeline problem. Medical Economics reported that the 2026 residency match offered 5,491 family medicine positions but filled 83.6% of them, leaving 899 unfilled and prompting a new review of the specialty's growth and sustainability. Those family medicine match results should sharpen the way practices think about artificial intelligence (AI), which can help physicians recover time. However, practices should use some of that gain to strengthen the next generation of clinical judgment.

The technology is already moving into routine medical work. The American Medical Association’s (AMA's) 2026 physician survey found that 81% of respondents use AI professionally, with common uses including research summaries, discharge instructions, documentation, chart summaries, patient-message drafts and assistive diagnosis. That rapid growth in physician AI use makes the training question immediate: What happens to early-career learning when software increasingly produces the first draft of work that physicians once had to construct themselves?

A useful warning comes from outside medicine: The Stanford Digital Economy Lab's August 2026 update found that employment among workers ages 22 to 25 in highly AI-exposed occupations stood about 19% below where it would have been if it had kept pace with less-exposed peers. The researchers said the divergence appears mainly through reduced hiring and is concentrated in occupations where AI tends to automate tasks, while stressing that their findings are descriptive rather than proof that AI caused the change. Those young-worker employment patterns do not map directly onto residency, but they illustrate a broader risk: If technology removes too much entry-level work, organizations can weaken the path by which novices become experts.

Medicine has a built-in defense against that pattern: supervised training. Residents and fellows still work within structures designed to turn formal knowledge into judgment through repeated cases, feedback and graduated responsibility. The AMA has noted that residents and fellows are already using AI and that residency programs are deciding how to incorporate it into graduate medical education. That push to integrate AI into residency training creates an opportunity to redesign learning before weak habits become embedded.

The practical rule should be straightforward: Let AI do more preparation, then require developing physicians to do more interpretation. Medical Economics has documented how ambient AI scribes can reduce documentation burden and restore time during and after visits. That shift in clinical documentation is valuable precisely because it can free attention for higher-value work. Practices should resist the urge to use every saved minute simply to add more volume.

A resident can receive an AI-generated chart summary, then identify which source facts deserve verification before the visit. A trainee can review a proposed differential, then explain what evidence would raise or lower each possibility and what cannot safely be inferred from the available record. A physician can use an AI-drafted patient message, then ask the resident to revise it around the patient's history, health literacy, adherence barriers and the specific decision the patient needs to make. These exercises concentrate training on judgment, uncertainty, communication and accountability.

The supervising physician's role should shift as well. Instead of spending scarce teaching time correcting routine formatting, the attending physician can make tacit reasoning visible. Why did one symptom change the plan? Why was an apparently abnormal value clinically unimportant? What triggered escalation? Which question unlocked the diagnosis? A skills-based approach to AI-era work helps physicians and practice leaders distinguish tasks that technology can support from the human skills that continue to create value.

Verification needs to become a trainable skill, not an afterthought. Medical Economics recently described real errors in AI-generated notes, including an incorrect diagnosis and an unprescribed medication, while also reporting that most notes in the cited pilot were free of serious errors. That combination makes AI verification in clinical documentation especially important. A tool that is usually right can be harder to supervise well because routine accuracy encourages complacency.

Practices should therefore measure more than minutes saved. Medical Economics has framed AI scribe performance around concrete outcomes such as documentation time, after-hours electronic health record work, patient interactions and practice finances. Alongside those measures, track time to independent competence. Can a developing physician verify an AI-generated summary, spot missing context, challenge a plausible recommendation, explain uncertainty and make sound decisions with progressively less review? Supervisors can sample AI-assisted cases, document recurrent failure modes and use those patterns to target teaching. The goal is to make AI-assisted practice more educational per hour, not merely faster per encounter.

That matters economically as well. More than 80% of surveyed independent primary care physicians were concerned about near-term financial stability, even as AI use was already common in their practices. Those financial pressures on independent primary care create a strong incentive to capture productivity gains quickly. But a practice that saves time today by hollowing out tomorrow's expertise is borrowing against its own workforce.

Primary care should use AI aggressively where it removes clerical friction, supports retrieval and creates a useful first draft. At the same time, physicians should deliberately preserve the learning loops that produce clinical judgment: exposure to cases, explanation, challenge, feedback and supervised decisions. That is a sound standard for AI adoption at work in any field, and it is especially important in medicine because the quality of future care depends on how today's developing physicians learn.

Primary care cannot solve its workforce challenge by narrowing the entry path. It needs more capable physicians to reach independence with stronger judgment. Use AI to give them that path.

Gleb Tsipursky, Ph.D., a behavioral scientist called the “Office Whisperer” by The New York Times, helps tech-forward leaders stop overpaying for AI while boosting engagement and innovation. He serves as the CEO of the AI consultancy Disaster Avoidance Experts and is the author of eight books. This article was adapted from his book, “The Psychology of AI Adoption at Work: From Resistance to Results” (Georgetown University Press, 2026).