I Spent 9 Years on Telemedicine Calls That AI Could Have Handled

During my years as a hospitalist, I watched the same frustrating pattern play out constantly. The ER would admit a patient with three days of worsening cough and fever. By the time they arrived, they were genuinely sick enough to need inpatient care. But if someone had seen them sooner, evaluated their symptoms properly, and started appropriate outpatient treatment, they never would have needed hospitalization at all.

Fast forward to 2017 when I started practicing telemedicine with SteadyMD. Over the past nine years I've done thousands of telemedicine visits. A significant portion end the same way: "I need you to go get examined in person." Those visits still provided value — but here's the thing that keeps nagging at me: I'm an expensive, limited resource spending 15 to 30 minutes determining that someone needs to see me in person. That's backwards. We need intelligent systems that can make that initial determination before consuming physician time, not after.

What a Digital Front Door Actually Means

A digital front door to healthcare isn't a website or a patient portal buried behind seventeen login screens. It's an intelligent system that meets people where they are, understands what they're experiencing, and guides them to the right care at the right time for the right price.

AI chatbots are already providing diagnostic results comparable to physicians, with safety ratings approaching 97% when properly designed and deployed. More importantly, these systems preserve precious physician time for the cases that truly require clinical judgment while scaling healthcare access to populations who currently have none.

The Evidence Is Promising, But Incomplete

Recent research demonstrates that well-designed symptom checkers achieve remarkable accuracy. A 2020 study evaluated eight symptom assessment apps against 200 clinical scenarios — the leading system provided the correct diagnosis in its top three options 71% of the time and delivered safe triage advice 97% of the time, matching physician safety ratings. A 2024 emergency department study found that AI chatbots matched physician triage decisions 70% of the time when using identical patient information.

But here's where I pump the brakes. The difference between 70% accuracy and 97% safety isn't just numbers on a page. It's the gap between a useful tool and one that's truly ready for universal deployment. We need to understand not just whether AI triage works on average, but precisely where it breaks down: which patient populations, which symptom presentations, which clinical edge cases expose its limitations.

Why This Matters Beyond Convenience

Physician time gets deployed where it matters. Every unnecessary ER visit, every telemedicine consultation that just determines someone needs in-person care — these waste the scarcest resource in healthcare: trained clinical judgment. When intelligent systems handle initial assessment, physicians spend their time on diagnosis, treatment planning, and complex decision making. That's not replacing physicians. That's respecting what physicians actually do.

Access barriers collapse. When you can assess symptoms at 2 AM from your phone, geography and clinic hours stop determining whether you get care. The patient in rural Montana gets the same triage capability as someone blocks from a major medical center.

The entire system becomes more efficient. Emergency departments see countless patients who don't need emergency care but had nowhere else to turn. Administrative costs consume roughly 25% of total U.S. healthcare spending, much of it tied to inefficient triage and routing.

What Actually Has to Happen

Building an effective digital front door demands more than accurate algorithms. It requires universal design that works for everyone — from pediatric cases to geriatric, mental health to physical symptoms. Research shows significant disparities in coverage among existing symptom checkers, with some unable to handle basic scenarios involving children, pregnancy, or mental health. A true front door does not lock out entire populations.

It demands transparent limitations. Studies demonstrate that chatbots tend toward overconfidence. A responsible digital front door knows when to say "you need a human clinician" rather than pretending it can handle everything.

The Real Win: Physicians Doing What Only Physicians Can Do

I became a doctor to diagnose complex cases, develop treatment plans, and help patients navigate difficult medical decisions. I did not become a doctor to spend 20 minutes on a telemedicine visit determining that someone needs to be seen in person.

Digital front doors don't replace physicians. They protect us. They preserve our capacity for the work that actually requires years of training and clinical judgment. But these systems only work if they're rigorously validated by the physicians who understand what actually matters in clinical practice. A chatbot that performs well on benchmark tests but misses subtle red flags in real patient presentations isn't just useless. It's dangerous.

That's why the work we're doing at Automate Clinic matters. We connect the companies building these systems with practicing physicians who can identify the failure modes that matter clinically, not just statistically.