Principles · v1

Five things we believe about software for physicians.

These are not commandments. They are working positions. We hold them firmly until we find better ones, and then we will change them in public.

Principle 01

The physician is the user. Not the chart.

Software in medicine has spent twenty years optimizing for the chart. The chart cannot consent. The chart cannot heal. The user is the human in the white coat, and the user we serve next.

Every screen, click, and modal is a tax on the only person in the room who can decide what to do next. The product question stops being “what does the record need?” and becomes “what does the clinician need to do this well, today, in the next ninety seconds?”

When the physician is the user, latency matters. Defaults matter. The number of decisions surfaced before a decision is made matters. We design for the cognitive budget of someone who has not eaten lunch.

Principle 02

Documentation is exhaust. Not work.

Notes, codes, orders, and letters are byproducts of care. They are not the care. Software should produce them from what already happened, not extract them by interrogating the doctor afterward.

If a system can listen, retrieve, and reason, the chart should write itself — with the clinician as editor, not author. The bar is not “good enough draft.” The bar is “signable in under a minute.”

Billing is a downstream consequence, not an upstream interrogation. Quality measures are computed from the record, not pre-typed into it.

Principle 03

Agents act, then explain. Not the reverse.

A useful clinical agent submits the refill, drafts the referral, calls the pharmacy, queues the prior auth, and shows its work. It does not produce a transcript and ask the human to do the next ten steps.

The unit of value is a finished task, attributable and reversible. “Here is what I did, here is the evidence, here is the undo button” — not “here is a summary of what you might want to do.”

Explanations are first-class artifacts: every action has a citation, every citation is auditable, and the clinician can intervene at any step.

Principle 04

Interop is a verb performed by software. Not a committee.

FHIR, HL7, CDA, USCDI — these are vocabularies. The work of moving the right information to the right place at the right moment is engineering work. Standards do not perform it; programs do.

If a record cannot be read by the next clinician who needs it, the system has failed, regardless of what was conformant. Compliance is a floor, not a ceiling.

Patient data belongs to the patient. Movement of that data should be cheap, fast, and permissioned by them — not gated by integration fees.

Principle 05

Time at bedside is the only metric that matters.

Throughput, RVU, click counts, screens per encounter — every operational metric is a proxy for one thing: minutes a clinician spent looking at a person instead of a screen. Optimize for that, or do not call yourself a clinical product.

If a feature does not return time, lower cognitive load, or prevent harm, it is decoration. If it adds clicks, it is a regression.

We will publish the time we return. We will publish the time we cost. We will be measured by the difference.

In summary