ClinicalCLINICAL // AAP AMB-DOSING PROTOCOLElena Rostova
1 MIN READ
Deterministic vs. Probabilistic: Where do you draw the absolute line on letting an agent execute a tool call in your clinical pipeline?
A lot of teams are letting LLMs decide tool arguments directly (e.g. calling create_prescription with whatever milligram dose the model hallucinates). That is malpractice waiting to happen. Here is our hard boundary: agents propose intents, pure deterministic code evaluates safety invariants.
31comments