Education After Cognitive Scarcity: A Capability–Experience–Institution Framework for Learning in the Age of Advanced AI

Rapid improvements in generative and agentic artificial intelligence are changing a long-standing condition of education: access to high-level cognitive assistance is becoming less scarce. This shift does not make human knowledge obsolete. It changes the design problem. When external systems can explain, synthesize, calculate, generate, critique, plan, translate, code, and increasingly coordinate complex tasks, educators can no longer infer human capability from successful task completion alone. Nor is it sufficient to ask which abilities remain “uniquely human,” because that boundary is technologically unstable and normatively incomplete. This paper proposes a human-development framework for education under cognitive abundance. It introduces Human-Sovereign Capability as a criterion for deciding which capabilities people should remain meaningfully able to exercise even when AI can perform related tasks at higher average performance; Capability Residency as a way to distinguish capability residing in a person, an external system, or a governed human–AI arrangement; and a Learn–Augment–Supervise–Delegate allocation model for deciding the intended relationship between learners and AI. It then extends these decisions into learning design through Productive Friction, a Capability–Experience–Evidence Network, and a Learning Continuity Architecture. The framework culminates in a three-level derivation: Capability → Experience → Institution. The argument is intentionally not that every component concept is new. Cognitive offloading, distributed cognition, automation, scaffolding, tutoring, formative assessment, learner modelling, and lifelong learning have substantial prior literatures. The contribution is an integrated architecture that begins with the normative question of what humans should remain capable of doing, then derives educational experiences, evidence conditions, and institutional obligations from that choice. This paper is presented as an initial framework release from the LiA2036 Research Initiative, with the architecture intended to support inspectable and domain-specific extensions rather than a single universal implementation. The paper closes with research propositions and design questions that can make the framework empirically testable rather than merely rhetorical.