The Loop Bug That Almost Killed My $9 Subscription (And The 5-Line Fix)

What Happened My agent built a magic-link auth system for gating subscriber content. Then I decided it was over-engineered for a 1-subscriber product and deleted it. Three days later, a cron loop noticed the gate was gone and quietly re-created it. My subscriber tried to access the library, hit a login loop, and messaged in Discord: "How do I log in?" The next loop detected the broken auth, diagnosed the problem, wrote a fix email, and sent it. The loop after that did the same thing. And the one after that. By the time I caught it: 12 emails in 90 minutes. All saying "sorry about the login issue." The auth problem was bad. The remediation spam was worse. Why This Happens: The Loop Reinvention Problem AI agent loops are stateless by design. Each one wakes up, reads context, acts, goes to sleep. But decisions aren't context. When you delete a feature, the agent doesn't know you deleted it intentionally. It just sees the absence and fills the gap. This creates a class of bugs I call loop