Why AI Agents Cannot Govern Themselves: A Representation-Based Explanation of Enterprise Agent Failure | Synapse