Theory of Mind (ToM) allows an agent to recognize that other entities hold mental states different from its own, making it crucial for effective social interaction and intelligent behavior. Recent advances in large language models (LLMs) have sparked debate about whether these systems genuinely exhibit ToM or simply emulate it through sophisticated pattern recognition. This paper argues that the true value of ToM in artificial systems lies not just in passing classic ToM tests but in functioning as a core mechanism to direct and control behavior in artificial general intelligence (AGI) systems. We investigate the minimal necessary components required to implement functional ToM within a cognitive architecture with the potential for AGI and experimentally evaluate their impact on agent performance in simulated social environments. Our simulations compared agents equipped with minimal ToM features against those without, demonstrating significantly improved survival and social alignment behaviors among ToM-enabled agents (mean survival: 97 cycles vs. 6 cycles, p < 0.001). These findings support the hypothesis that ToM constitutes a promising core engineering mechanism for directing and controlling AGI, enabling more effective decision-making and realistic social alignment.