The most interesting AI work in medicine is not happening inside the encounter.

It is happening between encounters.

Current AI lives inside the visit. Ambient scribing. Documentation. Differential support. Useful. Narrow.

The work that actually determines outcomes happens in the longitudinal space: tracking a patient's trajectory, maintaining context between visits, flagging dangerous drift before it becomes a crisis, simulating how different paths play out.

That is the second gear. The simulation layer.

Current AI has almost no presence there.

A physician's real cognitive load is not typing. It is carrying a mental model of dozens of patients across weeks and months. That is what a world model is for.

Whoever builds the longitudinal simulator finally separates physician cognition from physician administrative labor. Those two things have been fused together by decades of workflow design that never questioned why they had to be.

Full post: rewskidotcom.substack.c…