AI Agents in Production: What Actually Breaks

Demos of AI agents are easy; running them reliably in production is not. This article covers the failure modes we've hit running LLM-based agents in production — hallucinated tool calls, runaway loops, cost spikes — and the guardrails that fixed each one.