AI alone may outperform your care team by 2030. Is your practice ready?

Artificial intelligence (AI) may soon provide better patient care than human physicians alone or those who use AI to supplement their work, and when it does, there will be effects across medical practices.

“Will Autonomous AI Exceed AI-Aided Physicians as the Best in Medical Care?” is a perspective essay published Aug. 14 in JAMA. The “unsettling” answer: quite possibly, at least in some parts of health care, and likely with ramifications for practice management.

“Data from medicine and other fields suggest that when AI alone performance is consistently superior to human-alone performance, AI alone surpasses human-AI hybrids,” the authors said. “Paradoxically, hybrid care in which humans are in (or on) the loop to correct AI errors is likely to worsen rather than improve AI performance.”

Now is the time to begin considering effects on workflow, liability, regulation, reimbursement and medical education, the authors said, because “superior autonomous AI” could be ready to deploy in some, maybe many, workflows by 2030, the authors said.

Do you do these tasks as well as AI?

Large language models (LLMs), the AI systems behind tools such as ChatGPT, have only been public since November 2022. But authors Ezekiel J. Emanuel, M.D., Ph.D.; Abe Baker-Butler; Neal Khosla, M.S.; and Vinod Khosla, M.S., MBA, argue the technology is improving faster than physicians' own performance is. They cite at least five areas where AI is overtaking human doctors:

• Gathering patient information
• Diagnosing conditions
• Selecting tests to pinpoint diagnoses, while staying on budget
• Recommending treatments
• Managing chronic diseases, at least for hyperlipidemia, osteoarthritis, diabetes and breast cancer

On budget or better

The numbers behind that claim have direct financial relevance to practice budgets. In one study the authors cite, a Microsoft diagnostic tool working within an $8,000 test-ordering budget reached the correct diagnosis on 56 complex cases roughly four times as often as physicians working without access to colleagues, textbooks or the internet — and did it at 19.1% lower cost per case ($2,396 vs. $2,963). Similarly, Cedars-Sinai's AI system produced optimal treatment recommendations in 77.1% of 461 real patient cases, compared with 67.1% for physicians. At scale, gaps like that in diagnostic accuracy and test utilization translate into fewer repeat visits, less unnecessary testing and less exposure to claims denied for lack of medical necessity.

Go with the workflow

The authors make a claim that could reshape how practices think about "AI-assisted" workflows. The article's central assertion is that once AI alone gets better than physicians at a task, adding a physician back in to review the AI's work tends to make outcomes worse, not safer. A 2024 meta-analysis of 106 human-AI studies found that pairing a person with AI helps only when the person was already outperforming the AI on their own; when AI was already ahead, human review dragged its performance down. A separate 2025 review of 52 clinical studies found physician-AI teams "neither outperformed medical AI alone nor surpassed the best of clinicians or medical AI alone."

That has a workflow implication practices. Many AI vendors are currently selling a "physician reviews every AI output" model as the safe, responsible way to deploy AI in a practice. But that configuration may not be the one that produces the best results as the underlying models keep improving. Practices investing in AI tools built entirely around a mandatory physician sign-off step should ask vendors directly how that architecture is expected to change as the technology improves, rather than assume today's guardrails are permanent or that more physician oversight is automatically safer.

AI is not welcome here

Two forces work against smooth AI adoption in a practice, according to the article. One is algorithm aversion, a documented tendency, found in about 75% of studies on the topic, for people to distrust and override algorithmic advice even after being told it is more accurate. The effect is strongest among highly trained experts, physicians included, which means staff pushback on AI tools may run counter to what the data actually show.

The other is deskilling. As physicians and staff lean on AI for tasks like documentation, triage or diagnosis, their own proficiency at those tasks can erode, an effect already documented among endoscopists who rely on AI-assisted colonoscopy tools. For practice leaders, that is a training and staffing question as much as a clinical one. Decide now which skills the practice wants staff to keep practicing independent of the software, and build in the time to do it.

For staff training in the AI era, protect time for physicians and clinical staff to keep core diagnostic and documentation skills sharp, rather than letting AI quietly become the only version of those skills anyone in the practice still exercises.

Caveats against computer care

The authors cite clinical trials, but most of the supporting evidence comes from simulated exam scenarios rather than real patient encounters. While AI may be good a thinking about medicine, many elements of patient care, such as surgeries and deliveries, remain hands-on. Liability and reimbursement rules also have not caught up. If an autonomous or lightly supervised AI tool misdiagnoses a patient, it is not yet settled how fault and malpractice exposure would be divided among practice, physician and vendor.

There are at least two more practice management elements that may be worth considering now.

• Vendor contracts: Know what your AI tools' liability, data-use and update terms actually say before adoption expands further into the practice.
• Malpractice coverage: Ask your carrier how autonomous or semi-autonomous AI use is treated under your current policy, before a claim forces the question.