Inherited, Not Invented: Direction as the Missing Primitive in AI Coherence

Intelligence explains what an agent can do. This paper argues that a separate primitive, Direction, explains why an agent's behavior remains coherent toward an end over time — and that in artificial intelligence specifically, Direction has so far always been inherited from its creators rather than invented by the system itself. This is not offered as a permanent or metaphysical limitation but as a current, falsifiable, and actively watched condition: an AI system that originated its own research direction — choosing which problems matter, rather than executing on problems given to it — would count as a counterexample, and its absence is named as the field's own live frontier, not merely assumed by this paper. Unintended or emergent AI behavior is addressed directly and shown not to break the thesis, since such behavior remains a product of a process fully authored by humans even when its specific output was not chosen. The paper closes by arguing that this gap is not incidental: direction-setting is the one capacity deliberately held back precisely because nearly everything else has already been ceded to AI systems.