General agents are not universally capable, and their abilities are specialized across a fragmented world model. This limitation is formalized through proof, demonstrating that standard worst-case analysis is insufficient for distinguishing between critical bottlenecks and irrelevant failures1. The big-world regime highlights the necessity for a more nuanced understanding of agent capabilities, as uniform guarantees are inadequate. In this context, structural certification emerges as a crucial component for evaluating agent performance. By acknowledging the specialized nature of agent abilities, researchers can develop more effective frameworks for analyzing and improving agent behavior. The implications of this research extend beyond the technical realm, influencing policy, security, and workforce dynamics. So what matters to practitioners is that this newfound understanding of agent limitations can inform the development of more realistic and effective AI systems.