The first patient is in the lobby. The clinician opens the chart and sees what they always see: a wall of past data, organized for storage rather than for what’s about to happen. They scroll. They synthesize. They check the labs from the last visit. They guess at care gaps. Nothing on the screen is wrong; nothing on the screen is helping.
The EHR was supposed to be the workday. After two decades of it being the workday, the workday hasn’t gotten any easier.
The bargain.
The EHR we have today was a bargain. In 2009, the federal government promised tens of billions of dollars to providers who adopted electronic records. The bargain was simple: digitize the chart, get paid; don’t, don’t.
The bargain worked. Within ten years, paper records had effectively disappeared from American medicine. What replaced them was a database: a system of record. Patient demographics, claims, schedules, labs, notes, problem lists, allergies. All stored. All retrievable. All compliant with documentation standards a regulator could audit.
What the bargain did not deliver was a clinical workflow. The database was treated as if it were a workflow, as if storage and retrieval were the work itself. They aren’t. The work is the visit, the documentation as it happens, the coding while the assessment is on the screen, the order drafted from the conversation, the prior auth started before the clinician asks, the inbox cleared, the letter sent, the loop closed.
None of that is what a database does. The database stores the artifacts of work. It does not do the work.
We have spent twenty years asking the system of record to do something it was never built to do. Every chart prep tool, every smart phrase, every dictation app, every macro, every templated note: a workaround for the fundamental category error. The patient needs a clinician who is present. The chart needs documentation that is accurate. Neither needs the same database to do both jobs.
The category error.
A category error happens when one kind of thing is mistaken for another kind of thing. Treating the EHR as the workday is a category error.
Storage is one kind of activity. Action is another. A database stores; an action system acts. The two have different shapes: different verbs, different time horizons, different relationships to the patient encounter. A database holds state and waits to be asked. An action system changes state, and anticipates what’s about to be asked.
The point is not that the database is bad. The database is excellent at being a database. The problem is that we asked it to be a workflow. We bolted “intelligent” features onto the chart (population health alerts, BPAs, smart sets, embedded coding suggestions) and called the resulting kludge a “clinical workflow.” It is not a workflow. It is a database with extra widgets. The clinician still does the work; the database still gets in the way.
Every AI scribe, every coding assistant, every prior auth bot built on top of the EHR is a patch on the same error. Each tool tries to make the database do something the database was never built to do. The patches accumulate. The clinician now uses six tools to chart one visit. The chart still doesn’t write itself.
The fix is not another patch. The fix is to recognize that storage and action are different systems and to build the second one.
The missing layer.
A system of action is not an AI feature, and not another widget bolted onto the chart. It is a separate layer that runs on top of the database, scribe and coder and copilot working as one system, designed to do the clinical day, with the database underneath as the reliable source of truth it was always meant to be.
A system of action prepares the visit before the clinician walks in. It listens during the encounter and writes the note as the visit happens: to the right fields, in real time, in structured form. It surfaces the codes in-visit, attached to the right problems, with the assessment still on the screen. It drafts orders from the conversation that just took place. It generates the after-visit letters and closes the loop. And between visits, it answers the questions that come up on a real clinical day (the drug interaction, the dose check, the chart lookup) without requiring the clinician to switch contexts.
Each of these is a verb. Prepare. Listen. Write. Surface. Draft. Generate. Close. Answer. The database holds nouns: patients, problems, codes, orders, claims. The system of action handles the verbs. The two work together because they were never supposed to be the same thing.
The reason this layer is “missing” is that nobody built it. The EHR vendors built more database. The point solutions built more patches. The category itself, a system designed to do the clinical day, has been waiting for someone to build it.
What changes.
When the layer is in place, the shape of the day changes.
The clinician walks into the room with the chart already prepared. The visit happens. The note writes itself, accurately, to the right fields. The codes land. The orders draft. The follow-up letter generates. The patient leaves. The clinician moves to the next room.
What does not happen: the after-hours documentation block. The forty-five minutes of post-visit cleanup. The Saturday morning chart catch-up. The compromise between a complete note and a quick note. The accumulating denials from coding errors. The locum shifts to cover provider burnout. The notes in the inbox at midnight.
The database is still there. It still does what a database does: it stores the artifacts of care. But it is no longer being asked to do the work of caring. That work belongs to the system of action. And the system of action belongs to the clinician.
Healthcare built a system of record because that was the bargain the federal government offered in 2009. The system of record did its job. The job that remains, the day a clinician actually has, needs a different kind of system. That system isn’t another tool to add to the stack. It is the missing layer.
It is what we should have built next. We are building it now.