agentic coding tools are amazing. But even the best models make mistakes. Are we perhaps becoming too reliant too fast? Especially in sensitive applications.