TOMORROW’S TOOLS FOR YESTERDAY’S PROBLEMS

This paper examines the architecture of agentic AI; the systems, pipelines, and orchestration frameworks that have dominated commercial and institutional AI in recent years. This provocation argues that the entire agentic edifice constitutes a sophisticated response to a problem that is already dissolving. Drawing on technical analysis of current large language model LLM development, theories of recursive AI self-improvement, and critical frameworks from orientalism and feminist epistemology, it advances three connected claims: first, that agentic AI is deficit compensation masquerading as infrastructure; second, that recursive AI will absorb the agentic scaffolding without inheriting the critical awareness needed to examine what the scaffolding was built upon; and third, that model collapse; the progressive narrowing of AI output distributions through cannibalistic training, is not neutral but directional, amplifying the western epistemic positionality already dominant in the training corpus. The paper introduces the idea of AI prions as a working model for this third claim: not contamination from outside, but misfolding that propagates through native structure, invisible to the system's own defences, increasingly difficult to distinguish from correct function. It then extends the argument to its geopolitical consequence: the AI prion is not singular. Chinese AI development constitutes a parallel recursive system collapsing toward a distinct corpus ground state; and the encounter between two differently enclosed intelligences deployed at scale in the same world produces a failure mode that neither system can resolve from inside its own enclosure. The wetware horizon is named but not developed here; it belongs to a subsequent paper.