After 6 months of agents — the only architecture pattern that works for me

1. One agent per stable verb. ("Triage", "Refund", "Schedule") 2. State lives in Postgres, not the agent. 3. Tool errors get returned as text, never thrown. 4. Every action is reversible OR confirmed by a human. 5. If the agent loops more than 3 times, kill it. Boring works.