I built a circuit breaker for LLM agents after seeing someone lose $200 overnight

A student created a circuit breaker for large language model agents after witnessing two alarming incidents in the LangChain Slack. One user was unexpectedly billed for 211 continuous runs on their first query, while another lost $110 due to LangChain's default recursion limit of 9999, which can lead to endless loops. These experiences highlight the potential risks of unchecked AI agent behavior, underscoring the need for safeguards to prevent costly mistakes and ensuring responsible AI usage.