FundamentalsLAST_UPDATED: 2026-03-15
AI Agent Failure Modes: The Complete Taxonomy
A definitive guide to every way an AI agent can fail in a production environment, from hard crashes to silent hallucinations.
Contents
Category 1: Silent Failures
The agent returns a success code, but the database write never happened. This is often caused by 'hallucinated' confirmations from the LLM after a tool call fails silently.