jairooh
Re:Ask HN: What's the costliest mistake you've made with LLM agents in...
on Wednesday March 18, 2026 @09:06AM (#47425326)
Ours was a LangChain agent we deployed to automate customer research. No budget cap, no monitoring, no approval gates. It ran a recursive loop overnight querying external APIs — we woke up to a $600 bill and zero useful output.
That incident is literally why we built AgentShield (useagentshield.com). Runtime risk scoring on every agent action, cost tracking per run with kill switches, and human approval gates before high-risk decisions.
Curious what others have run into — cost overruns seem common but prompt injection incidents in production are the ones that scare me most.
Ask HN: How are you monitoring AI agents in production?
With the recent incidents (DataTalks database wipe by Claude Code, Replit agent deleting data during code freeze), it's clear that running AI agents in production without observability is risky.Common failure modes I've seen: no visibility into what the agent did step-by-step, surprise LLM bills from untracked token usage, risky outputs going undetected, and no audit trail for post-mortems.I've...