Tech Blog

Field notes from real projects: the trade-offs we made, and why

2026-09-22
Multi-tenant Data Isolation: Adding tenantId Is Only the First Layer
We build an AI agent platform for grassroots public health, shared by several districts. In phase two we went back through data isolation and found that putting tenantId on every table solves only half the problem. This is the second layer — data scope inside a tenant — and the three places where it leaks most often.…
Read the article → A Liang · Chada Software　|　Building AI Agents in Production · part 2　|　Multi-tenancy · Data Isolation · Authorization · RBAC · Architecture
2026-09-21
Permission Isolation for an AI Agent: the Architecture Fix After We Found a Way Around RBAC
We build a healthcare-facing AI assistant: clinical staff talk to it inside WeCom, with a web admin console alongside. Once the product worked end to end we reviewed the permission model — and the most dangerous path turned out to be the entry point we added last. This is the write-up of that architecture fix, and of how "separate decision from execution" lands in practice.…
Read the article → A Liang · Chada Software　|　Building AI Agents in Production · part 1　|　AI Agent · Authorization · Multi-tenancy · Architecture · Fail-closed

Working on something similar?

Agent permissions, multi-tenant isolation, healthcare data — happy to compare notes.