The Most Important Thing I Did That Night Had No Billing Code

The ambulance was already on the way. I was early in my career caring for a pediatric brain cancer patient who had deteriorated at home. I had worked her up. We did a CT scan and it showed an enormous brain bleed. She was alive but she was very ill. I contacted her neurosurgery and oncology specialists at our area’s pediatric hospital and had the transfer arranged. Then I stopped, because I understood something the objective protocols didn’t spell out. She was going to die tonight.

I pulled the parents aside and asked them whether they wanted her to take her last breath in a sterile room two hours from home, or at home, surrounded by the people who loved her. They knew the answer before I finished the sentence. We canceled the ambulance. They took her home. On the way out they came back to the department to put their arms around me, with everything they had left, and thank me. Every time I remember that moment, I still tear up.

There is no code for canceling an ambulance. There is no quality measure that rewards it, no RVU attached, no line in any dashboard that lights up green when a physician decides the most healing act available is to stop treating. By every metric my institution tracked that night, I did less. By the only measure that mattered to that family, it was the most important thing I have ever done in a clinical room.

Disease has an ending. Health does not.

Simon Sinek borrowed a concept from the philosopher James Carse: there are finite games and infinite games. Finite games have fixed rules, known players, and an ending you can win. Infinite games have shifting rules and no finish line, and you do not play them to win. You play them to keep playing.

Disease and our healthcare system is full of finite games. The bleed is a finite problem. The fracture, the infection, the potassium of 6.8, these are complicated problems with answers, and medicine is rightly proud of how many of them it has learned to win. But health is not a finite game and it never was. You do not cure a person. You keep them in the game as long and as well as you can, and then you help them leave it on their own terms. The CPT code is a finite-game artifact. It assumes a discrete service with a beginning and an end. Canceling that ambulance was an infinite-game move, and the finite-game machinery had no way to recognize it as care.

Almost everything going wrong in how we fund, measure, and now automate medicine comes from treating an infinite game as if it were a finite one.

We turned the practice of medicine into a game we could score.

You cannot manage what you cannot measure, so we measured. We built a payment system on RVUs, a throughput culture on door-to-disposition times, a quality apparatus on codes and capture rates. None of this was malicious. It was the left-hemisphere doing what it does well, taking a complex thing and making it legible. The trouble is what happens to the thing once you score it.

Dan Pink wrote a book on the mechanism. Drive lays out the research that if-then extrinsic rewards reliably degrade performance and satisfaction on work that is complex and heuristic rather than simple and algorithmic, which is most of medicine. Goodhart’s Law says the rest: when a measure becomes a target, it stops being a good measure. And Arthur Brooks, on a recent podcast, gave the human version without naming either man. Talk about doing something purely for its result, he said, and you strip the love out of it. He was talking about hobbies and friendships. He could have been talking about a clinic.

Brooks also handed me the best conceptual tool in the conversation, a distinction underneath the worn-out left-brain and right-brain talk. Some problems are complicated, which means they are hard but knowable and solvable. “How does my car work?” is complicated but solvable. You could learn it. Other problems are complex, which means they cannot be solved, only lived with and understood. His marriage, he said, after thirty-five years, is a complex problem he will never figure out, and that is precisely why he loves it. Medicine has a complicated layer and a complex core. We have spent twenty-five years pretending the whole thing is complicated, because complicated problems are the only kind our metrics see.

I helped built the machine.

I am not immune to the dopamine jolt that technology provides. I helped create an environment that focused on the tools, not the outcome. I am an Epic physician builder. At one point, I was told that I had created more Epic tools one year in that role than any other physician in the Epic-sphere. One tool that I got really good at building was Epic’s main decision support mechanism, BPAs. I was also told at one point that our organization had more BPAs than any other similarly sized organization. Before I started to puff my chest on either of these “accomplishments,” I realized that these were not things to be proud of. I had mistaken volume and technical prowess for real-life effectiveness and first principles moments like calling off an ambulance.

Pointing AI at the score just helps us lose faster.

Here is where it gets dangerous, because the tool we are now deploying at scale is very good at exactly the wrong thing.

Most healthcare AI pitches I see are finite-game accelerators. Optimize the throughput. Lift the coding capture. Shave the documentation time. Move the metric. And they will, which is the problem. Deploy AI on top of a system that mistook the score for the game, and you do not fix the foundation, you industrialize the dysfunction. You run the same flawed logic faster and call the speed progress.

Watch how this plays out with burnout, which everyone agrees is the crisis and almost no one diagnoses correctly. Brooks splits well-being into enjoyment, satisfaction, and meaning, and points out that only meaning has collapsed in the general population. His tell is revealing: depression has tripled and the number of therapists has tripled over the same period. More of the apparent solution, more of the disease. Clinicians still have satisfaction, the hard goals and the real struggle are intact. What got buried under the clicks and the prior auths and the RVU targets is meaning, the right-hemisphere experience of why any of it matters. When we deploy AI to remove clicks and lift throughput, we are servicing the satisfaction machinery and never once touching the wound, because the wound is not an efficiency problem. A left-hemisphere answer to a right-hemisphere question will always look like it is helping and never quite help.

The drudgery and the medicine are tangled together.

The instinct, then, is to point AI only at the complicated, finite, left-brain work and leave the rest alone. Right instinct, harder execution, because the two are braided at the level of the individual task.

Take the most purely administrative thing in a visit: booking the follow-up. That is left-brain work, transactional, rules-based, the ideal target for automation. But watch what actually happens at the front desk. The receptionist schedules the appointment and, handing over the card, says good luck with your daughter’s recital next week. That sentence is right-brain contact, the oxytocin moment Brooks describes, the thing the patient’s nervous system registers as I am known here. It rode along on a transaction. Automate the transaction cleanly, route it to a patient portal and a confirmation text, and you have not just removed friction. You have deleted the only warm human moment on the way out the door, and you will never see it on a dashboard because it was never measured to begin with.

This is the real argument for ambient AI, and it is not the one the vendors lead with. The pitch is a better note, faster. The actual prize is that a physician who is no longer typing can look the patient in the eyes, which is the only way, neurologically, that a human brain registers another human as present. Done well, ambient AI hands the right hemisphere back to the doctor. Done badly, it is just a third thing in the room listening.

Not all friction is waste.

So the design principle is not remove friction. It is sort it.

The dating apps, of all things, prove the point. They spent a decade engineering human friction out of courtship, the swipe reduced to a left-brain curve-fit, and the result was a generation that felt lonelier the more efficiently it matched. The ones getting better are getting better by adding friction back, and specifically by adding a human right brain to the process: routing your matches through a friend who decides who you actually meet, or putting a group of people in a room together and letting chemistry do what no algorithm can. They did not optimize the friction away. They learned which friction was the product.

Medicine needs the same sorting, and AI is the right tool for it if we are honest about the job. The fax, the prior auth, the redundant click, the alert that fires for the four-hundredth time today: wrong friction, pure drag, rip it out and point the models straight at it. The eye contact, the room that goes quiet while a family decides, the receptionist who remembers the recital, the physician who cancels the ambulance: right friction, the friction that is the medicine. Protect it. Where we have already automated it away, consider adding it back. The question for every deployment is not how much friction it removes. It is which friction.

The most powerful medicine has no code.

Push this far enough and you arrive somewhere uncomfortable for a system built on billing. The interventions with the most leverage over a population’s actual health are largely right-hemispheric and almost entirely unbillable. The Surgeon General’s advisory on loneliness put the mortality risk of social isolation in the range of smoking. Sitting with friends, getting outside, seeing the sun, being needed by someone: there is real evidence these move outcomes more than a good deal of what we prescribe, and there is no CPT code for any of them. This is not an argument against medication. It is an argument that a finite-game payment system is structurally blind to its own highest-leverage moves. It cannot see them, cannot measure them, cannot fund them, so it pretends they are not medicine.

I learned this again, recently and gently, outside the ED. A physician I had worked alongside for decades was retiring, and I knew the transition was hard for him. I had a family member seeing him, so I went along to the visit. Beforehand I had done the quiet work of finding out what would actually land, and we gave him a card with a donation receipt to a charity that mattered to him. He had not expected any of it. He fought back tears and said he had goosebumps. None of that was reimbursable. None of it took an algorithm. It was the kind of thing that is the entire reason people go into this work and the last thing any of our systems are built to notice.

The CPT codes and the RVUs pay the bills. They are not why we do this, and they were never going to be. We keep building cleverer tools to win a game that has no ending, and the cleverness is real, and the win condition is a category error. The point was never to win. The point is to keep someone in the game, and to stay human while we do. If the AI we are pouring into healthcare cannot tell the difference between the friction that is killing us and the friction that is the medicine, then all we have built is a faster way to score a game. But healthcare, medicine and the relationships and trust that should be part of them are infinite. Scoring that game is missing the point.

John Lee is an emergency physician and Epic consultant who helps health systems bridge the gap between Epic’s capabilities and operational reality. He specializes in data architecture, registry optimization, and making Epic’s tools actually deliver results.

If you need help configuring your Epic environment to support these capabilities, connect with him on LinkedIn or via his website.