Anyone else had an agentic loop blow through their API budget overnight?

Last month I woke up to an invoice that made me do a double-take. Turned into API costs overnight because of a retry loop bug, the kind of thing that's invisible until you are staring at a bill you cannot explain.

What got me was not the cost itself, it was how late I found out. Every cost tracking tool I checked afterward would have told me the same thing: here's what happened, after it already happened. None of them would've stopped the loop while it was running.

Curious if others have hit this, especially anyone running agents or automated pipelines that make LLM calls without a human in the loop watching each one. How are you catching this stuff now? Alerts? Manual spend caps? Just... vibes and hoping?

(Building something to solve this myself, happy to share more if useful, but mostly want to hear if this is a common pain or if I got unusually unlucky.)