Why Observability Is Not Enough: Building a True Kill Switch for AI Agents

If you are building AI agents with Python, you have probably experienced the moment of panic when an agent gets stuck in an infinite loop. It keeps calling the OpenAI API, burning through your credits, and you only realize it when you check your billing dashboard hours later. The standard advice in the industry right now is to use observability tools. We are told to wrap our LLM calls in tracing libraries so we can see exactly what the agent is doing. While tracing is incredibly valuable for debugging, it has a fundamental flaw when it comes to cost control and safety. Observability is entirely passive. It records the disaster perfectly, but it does absolutely nothing to prevent it. When an agent goes rogue, you do not just want a log entry. You want the execution to stop immediately. To solve this, I built AeneasSoft. It is an open-source circuit breaker designed specifically for AI agents. Instead of just logging the requests, it actively monitors the traffic and blocks calls in appl