<p>Agentic AI does not create a harder version of the evaluator problem that existing AI literacy frameworks address; it creates a different problem. When an AI system plans, decides, and executes multi-step actions on behalf of a user, the user is no longer the evaluator of AI outputs. The user is the principal who has delegated agency to an AI system. Existing AI literacy frameworks, built for the evaluator paradigm, do not equip users for this relationship, and the governance roles that emerging agent-safety frameworks prescribe are not exercisable without corresponding principal-side literacy. I introduce the Agentic AI Literacy Framework (AALF) to address this gap. The framework comprises six competency domains structured across three proficiency levels, grounded in both principal-agent theory and documented agentic incidents, with relational autonomy, the capability approach, meaningful human control, and trust-calibration research supplying normative and operational content. The framework is both a literacy model for users and a design constraint for builders. Because most users will predictably remain at a baseline level of agentic competency, system design must compensate rather than push the burden further onto them. The paper develops three adoption artifacts to support this design-constraint framing: an incident-to-competency map, a stakes-by-autonomy proportionality matrix, and a mapping to existing regulatory instruments including EU AI Act Article 4, GDPR, NIST AI RMF, and OWASP, alongside a twelve-question Readiness Self-Assessment. Documented incidents from 2024 and 2025 illustrate the specific competency gaps the framework addresses.</p>