A senior programmer doesn’t just see bugs in someone else’s code as isolated: they infer underlying thought patterns (common assumptions, habitual shortcuts, structural misunderstandings) that help them predict where else to look. Across professions, expertise relies on forming a theory of mind: an internal model of how someone thinks. But when that “someone” is an AI agent, a single mistake may not reveal a stable pattern. Errors can be independent and uncorrelated, making them harder to diagnose or anticipate. Unlike humans, AI systems are not constrained by shared cognitive biases or developmental trajectories. Their “thought patterns” can be opaque and discontinuous. Bridging this gap between machine reasoning and human theory-of-mind inference will be essential to keeping AI behavior aligned and predictable.