Python RLS for Agents: Row-Level Security in LangGraph and Multi-Tenant LLM Workflows An agent we shipped last quarter returned a row from the wrong tenant. The agent did what the prompt asked. The database did what the agent asked. The third jurisdiction of every agentic workflow is the database, and Pydantic and TypedDict cannot reach it.
Python I am a TypeScript developer. Python typing has not caught up. A TypeScript developer integrates LangGraph into a Pydantic-typed FastAPI codebase and finds a wall. The wall is real, structural, and identical in shape to walls every modern language has hit. Naming it changes everything.
