Continuously adding exception rules to an AI agent can reach a tipping point where the system becomes too rigid, rejects all inputs, and breaks. This "over-guardrailing" is a form of technical debt that diminishes, rather than improves, the agent's effectiveness.