AI and the Physician Shortage: The Strong Case for Urgent, Careful Progress

The following is a guest article by Ahmed Omar and Amit Kumthekar, Co-Founders of Sully.ai

Patients have raised important questions about consent and AI scribes, including in lawsuits alleging doctor-patient conversations were recorded without permission. Such concerns must be taken seriously, and they point toward a clear path forward, where transparency and consent is embedded into how these tools are used from the get-go.

It matters that we get this right as AI scribes increasingly provide value for doctors and patients: they cut back time doctors spend writing notes by as much as 21 minutes per day – time doctors can give back to patients and fewer late nights catching up on documentation. 

Technology can make care safer. One federally funded review found hospital drug orders entered by computer were almost half as likely to contain an error than paper orders – preventing an estimated 17 million estimated medication errors per year. With clear consent, AI scribes can build on that progress, so patients get both privacy they expect and the attentive, accurate care that they deserve.

The shortage

The U.S. is expected to be short 86,000 physicians by 2036, according to the Association of American Medical Colleges. Federal projections show nonmetro areas will have just 42% of the physicians they need by 2038, compared with 95% in metro areas. It takes a decade to train a doctor; the doctors we do have need time back right now, not later. 

Paperwork eats time

For each hour physicians spend with patients, almost  two go to screens and desk work, plus one to two hours at home nightly. And family physicians spend 5.9 hours of an 11.4-hour day in the electronic health record.

When The Permanente Medical Group gave 7,260 physicians AI scribes, they saved an estimated 15,791 hours in about a year, or 1,794 workdays. Almost half of patients said their doctor spent less time at the computer and not a single one reported their visit got worse.

Why we built Sully

Ahmed: I watched my mother suffer serious harm from a medication error. This wasn’t an isolated event; the right information didn’t reach the right person in time. Such errors injure more than 1.5 million Americans in hospitals every year.

Amit: I was diagnosed with type 1 diabetes at 2. My care has always depended on data being correctly read. I began working on technology for automated insulin delivery with my dad when I  was t 8, began research at Stanford at 16  and joined Sully at 19.

Sully builds AI agents for the work that surrounds each visit, like scheduling, intake, documentation, coding and follow-up. Sully also shares what a health system learns about every patient, so that every doctor walks in with a full picture. Doctors remain in charge, and, when our models disagree, a human being reviews each case. 

Washington works on supply

We’re encouraged that Congress is weighing the bipartisan Resident Physician Shortage Reduction Act to fund more residency slots, as well as a bill extending visa waivers for rural doctors. The Centers for Medicare & Medicaid Services is additionally sending states $50 billion over five years through its Rural Health Transformation Program, including for tech. These efforts have our full support, but realistically, it will take years to experience the fruits of these policy changes. 

While these bills add doctors, the bipartisan COMPLETE Care Act would stretch psychiatrists we have by paying primary care practices to bring them in as consultants. That’s critical– by 2037, the U.S. is projected to have only enough adult psychiatrists for under half patients who need one. 

Risks and a blind spot

With AI, it’s imperative we move carefully. States have introduced more than 280 health AI bills this year and enacted 33. Many of them require that patients are told when AI is involved in their care. Meanwhile, the gloves are off in Washington, which is slugging it out as to whether to override them. Regulators have also proposed dropping rules that make vendors disclose how clinical AI is tested. The industry should keep disclosing anyway.

Trust is falling. Only 42% of adults say they’re open to AI in their own care, down from 52% in 2024. Yet 51% have used AI for an important health decision without a medical professional. When people can’t see a doctor, they turn to chatbots alone.

Federal health IT regulators should remain highly skeptical, because scrutiny is essential for robust public health. 

A full stop on AI in health care is the wrong prescription for the moment

Some thoughtful people want to halt AI in medicine until every question is answered. We understand the instinct. But the status quo is the much bigger safety risk. Instead of pausing medication errors, let’s move forward carefully.  A halt would not fix medication errors, the rural or mental health practitioner shortage or the hours upon hours lost to paperwork and ceded to physician burnout. Instead, it would lock them in, leaving patients with the most human-error prone tools.

The better path? The one medicine takes with any new treatment: first test carefully, next disclose honestly, always ask for consent, and keep the human physician in charge. Publish the results transparently. Hold companies like ours to that standard, now and always.

About Ahmed Omar

Ahmed Omar is a co-founder and the CEO of Sully. He started the company after his mother was seriously harmed by a medication error, and he set out to build technology that could help prevent similar mistakes. Today he leads a company developing autonomous AI agents that help physicians manage clinical operations and spend more time with patients. His focus is moving healthcare AI beyond chatbots to agents that handle real-world clinical work, while keeping physicians in control.

About Amit Kumthekar

Amit Kumthekar is a co-founder and the Chief Product Officer at Sully. Diagnosed with type 1 diabetes at age 2, he jailbroke insulin pumps to build automated insulin delivery systems with his dad at 8 and became one of the youngest sponsored principal investigators at Dexcom when he was 18. His early research now underpins several of Sully’s patents and papers. At Sully, he leads product, engineering, and applied AI, turning clinical AI research into systems that fit real healthcare workflows, expand clinicians’ capacity, and improve the care patients receive.⁠

The following is a guest article by Ahmed Omar and Amit Kumthekar, Co-Founders of Sully.ai

Patients have raised important questions about consent and AI scribes, including in lawsuits alleging doctor-patient conversations were recorded without permission. Such concerns must be taken seriously, and they point toward a clear path forward, where transparency and consent is embedded into how these tools are used from the get-go.

It matters that we get this right as AI scribes increasingly provide value for doctors and patients: they cut back time doctors spend writing notes by as much as 21 minutes per day – time doctors can give back to patients and fewer late nights catching up on documentation. 

Technology can make care safer. One federally funded review found hospital drug orders entered by computer were almost half as likely to contain an error than paper orders – preventing an estimated 17 million estimated medication errors per year. With clear consent, AI scribes can build on that progress, so patients get both privacy they expect and the attentive, accurate care that they deserve.

The shortage

The U.S. is expected to be short 86,000 physicians by 2036, according to the Association of American Medical Colleges. Federal projections show nonmetro areas will have just 42% of the physicians they need by 2038, compared with 95% in metro areas. It takes a decade to train a doctor; the doctors we do have need time back right now, not later. 

Paperwork eats time

For each hour physicians spend with patients, almost  two go to screens and desk work, plus one to two hours at home nightly. And family physicians spend 5.9 hours of an 11.4-hour day in the electronic health record.

When The Permanente Medical Group gave 7,260 physicians AI scribes, they saved an estimated 15,791 hours in about a year, or 1,794 workdays. Almost half of patients said their doctor spent less time at the computer and not a single one reported their visit got worse.

Why we built Sully

Ahmed: I watched my mother suffer serious harm from a medication error. This wasn’t an isolated event; the right information didn’t reach the right person in time. Such errors injure more than 1.5 million Americans in hospitals every year.

Amit: I was diagnosed with type 1 diabetes at 2. My care has always depended on data being correctly read. I began working on technology for automated insulin delivery with my dad when I  was t 8, began research at Stanford at 16  and joined Sully at 19.

Sully builds AI agents for the work that surrounds each visit, like scheduling, intake, documentation, coding and follow-up. Sully also shares what a health system learns about every patient, so that every doctor walks in with a full picture. Doctors remain in charge, and, when our models disagree, a human being reviews each case. 

Washington works on supply

We’re encouraged that Congress is weighing the bipartisan Resident Physician Shortage Reduction Act to fund more residency slots, as well as a bill extending visa waivers for rural doctors. The Centers for Medicare & Medicaid Services is additionally sending states $50 billion over five years through its Rural Health Transformation Program, including for tech. These efforts have our full support, but realistically, it will take years to experience the fruits of these policy changes. 

While these bills add doctors, the bipartisan COMPLETE Care Act would stretch psychiatrists we have by paying primary care practices to bring them in as consultants. That’s critical– by 2037, the U.S. is projected to have only enough adult psychiatrists for under half patients who need one. 

Risks and a blind spot

With AI, it’s imperative we move carefully. States have introduced more than 280 health AI bills this year and enacted 33. Many of them require that patients are told when AI is involved in their care. Meanwhile, the gloves are off in Washington, which is slugging it out as to whether to override them. Regulators have also proposed dropping rules that make vendors disclose how clinical AI is tested. The industry should keep disclosing anyway.

Trust is falling. Only 42% of adults say they’re open to AI in their own care, down from 52% in 2024. Yet 51% have used AI for an important health decision without a medical professional. When people can’t see a doctor, they turn to chatbots alone.

Federal health IT regulators should remain highly skeptical, because scrutiny is essential for robust public health. 

A full stop on AI in health care is the wrong prescription for the moment

Some thoughtful people want to halt AI in medicine until every question is answered. We understand the instinct. But the status quo is the much bigger safety risk. Instead of pausing medication errors, let’s move forward carefully.  A halt would not fix medication errors, the rural or mental health practitioner shortage or the hours upon hours lost to paperwork and ceded to physician burnout. Instead, it would lock them in, leaving patients with the most human-error prone tools.

The better path? The one medicine takes with any new treatment: first test carefully, next disclose honestly, always ask for consent, and keep the human physician in charge. Publish the results transparently. Hold companies like ours to that standard, now and always.

About Ahmed Omar

Ahmed Omar is a co-founder and the CEO of Sully. He started the company after his mother was seriously harmed by a medication error, and he set out to build technology that could help prevent similar mistakes. Today he leads a company developing autonomous AI agents that help physicians manage clinical operations and spend more time with patients. His focus is moving healthcare AI beyond chatbots to agents that handle real-world clinical work, while keeping physicians in control.

About Amit Kumthekar

Amit Kumthekar is a co-founder and the Chief Product Officer at Sully. Diagnosed with type 1 diabetes at age 2, he jailbroke insulin pumps to build automated insulin delivery systems with his dad at 8 and became one of the youngest sponsored principal investigators at Dexcom when he was 18. His early research now underpins several of Sully’s patents and papers. At Sully, he leads product, engineering, and applied AI, turning clinical AI research into systems that fit real healthcare workflows, expand clinicians’ capacity, and improve the care patients receive.⁠