For the first few years after residency, Jason Levine was doing charts after 8 PM. Not occasionally. Routinely. By his own estimate, documentation was consuming more than ten hours of his week, and the hours were coming from wherever he could find them, which meant evenings, and then home, and then the slow accumulation of what he now recognizes as burnout.
Levine is a pediatrician, and the patients he sees are children. The conversations he needs to have are with families, often anxious ones, often with a lot to say. But during those early post-residency years, he has been candid about where his attention actually was. He deployed Epic's smartsets and templates aggressively, building whatever efficiency he could into the workflow. It was not enough. "My focus was 70% on documenting, 20% gathering my thoughts, and 10% listening to my families," he said. That arithmetic is worth sitting with. The family across from him, the child on the table, the reason for the visit, all of that was receiving one tenth of his available attention.
"I definitely was not as present as I should be with the families and kids I worked with."
This is the burden that rarely appears in conversations about electronic health records, which tend to center on interoperability, data quality, or billing compliance. What Levine describes is something more intimate and harder to quantify: the way a documentation requirement, applied hour after hour across a busy primary care panel, can restructure a clinician's cognition inside the exam room itself. The chart was not just taking time. It was taking the room.
His practice has since adopted Doximity's scribe feature, an AI tool available at no additional cost to physicians who already hold an NPI and carry a Doximity account. Levine credits it with a significant shift. Evening charting has become rare. The weight that had settled into his post-residency routine has largely lifted. His endorsement, though, comes with a specific caveat: the tool's transcriptions are not always accurate and require editing before they can be trusted in the record. The solution exists, he argues, but only if it is used with enough discipline to catch what it gets wrong.
That tension is probably where the conversation needs to stay for a while. The technology has arrived. The workflow around it, the quality checks, the professional habits required to keep an AI-assisted note from becoming an AI-authored record, is still being worked out in practices like Levine's in real time. Whether the scribe becomes a genuine corrective or just relocates the documentation burden from the evening to the review queue is a question his experience raises but, so far, cannot fully answer.
