A research-grounded exploration of how ‘I coherence’—the human capacity to sustain identity across contradiction—maps onto AI coherence, and why the next generation of safe AI depends on building persistent observer states rather than bigger models.