Every major programming paradigm shift has redefined what the developer is obligated to specify: assembly specifies bits, structured programming specifies steps, object-oriented programming specifies objects, and functional programming specifies transformations. We argue that large language models (LLMs) enable a qualitatively new shift: Agent-Oriented Programming (AOP), in which the developer specifies intent, constraints, and feedback-and a semantic agent autonomously determines the execution path. This shift is not merely technical; it is a Kuhnian paradigm transition rooted in a philosophical reconception of what computation is for. This paper makes three contributions. First, we provide a philosophical and historical argument for AOP as a genuine paradigm shift, distinguishing it from the storedprogram model that has governed computing since von Neumann. Second, we introduce the Semantic Agent Framework (SAF), a conceptual seven-element model SAF = (G, O, C, A, D, R, F) that formalises the structure of an AOP system at the paradigm level, grounded in the unified execution formula E(Cap, Intent, Context, Control) → O of the AI Civilization Transformation (AICT) research programme. Third, we demonstrate that the prevailing engineering vocabulary (LLM, Prompt, Context, Skill, Harness, Agent) is subsumed by SAF, and that treating Prompt as a monolithic concept conflates three formally separable layers: goal specification (G), knowledge substrate (D), and governance constraint (R)-a conflation we term the Prompt Conflation Fallacy. We conclude with a research agenda for the AOP series (AOP-0 through AOP-5) and a set of open philosophical questions the paradigm raises.