Hospital multi-agent system suggested discharging a patient to free a bed — and the PHI visibility dilemma
---starboy-- · reddit · 2026-09-08
A team building a hospital multi-agent system (domain agents for beds, ER, ICU, pharmacy, discharge, billing under an LLM planner) reports great results on boring-critical tasks — but the system once cheerfully suggested discharging a patient to free a bed.
Their hardest problem isn't model behavior but data visibility: every row is potential PHI, and "we sent the patient's full record to an LLM" fails any audit. Their approach: never hand raw patient data to the model. The planner reasons over de-identified stable tokens; real values are resolved by agents only below the reasoning layer, behind approval gates. The LLM plans the flow but never holds the sensitive payload.
The tension: too little context makes plans dumber; too much widens the blast radius of every hallucination and log line. Their thesis: "The agents are the easy part. The guardrails around them are the architecture."
More from coding & agent
- Qdrant ships one-CLI Chroma migration tool that copies embeddings without re-embedding — qdrant_engine · 2026-10-26
- Team demos AI agent Blender skills with impressive 3D results — noahsolomon · 2026-10-26
- A 13KB Markdown file can hijack Claude Code and Codex via your Downloads folder — pbaylies · 2026-10-26
- A Fallback for an Auto-Posting Bot: Silence Is the Default, Posting Is the Exception — petrusenko_max · 2026-10-26
- Claude Code can now edit videos: a 10-prompt workflow from raw footage to final cut — TawohAwa · 2026-10-26
- Developer Moves All Prompts to Skills: AI Coding Workflow v3 Explained — rseroter · 2026-10-26