A decision that could not be undone.
2am.
A donor hospital.
Mine alone.
UNOS data in my head.
The cold ischemic clock running.
A donor family upstairs.
My recipient across the country,
their family deciding
whether to say goodbye.
The AI had done its part.
Risk scores. Ischemic windows. Offer processed.
Then it stopped.
Take this organ, for this person, now.
Or not.
No way to document why.
No record of what I overruled.
Nothing between the output
and the irreversible call.
Just everything I knew,
carried in my head.
That is every clinical AI workflow today.
The AI ingests, analyzes, flags, generates.
Then it stops.
A clinician picks it up.
Decides alone.
Carries the accountability
with nothing underneath.
That moment has a name.
The Handoff.
Where the AI stops and the clinician starts.
Not the model. Not the data.
The Handoff.
Because no one designed for it.
The Handoff is the junction between two seats.
A Governance Owner who authorizes the agent.
A Decision Owner who holds the call.
That night, I held the Decision Owner seat alone.
No Governance Owner. No Handoff. Nothing underneath.
Faster output was never the need.
Judgment architecture was.
Without it, AI produces outputs.
With it, AI produces decisions.
The Accountability Gap™ (TAG™)
does not live at the model
or the data.
It lives at the Handoff.
If the clinician figures it out alone,
you bought a workflow.
If the Handoff is designed, with a name
and a trail, you built infrastructure.
Every clinician has had their own 2am.
You already know which one was yours.
Mo Johnson, MD MBA is a cardiothoracic surgeon and the founder of GPe Research. Field Notes are short dispatches from the clinical AI accountability frontier, published alongside the MedicoVigilance™ newsletter at medicovigilance.org.
Follow the work on LinkedIn: linkedin.com/in/mo-johnson