<p>Delegation to agentic artificial intelligence (AI) is increasing as these systems can plan, decide, and execute tasks through external tools across everyday and organizational settings. However, users do not judge delegation only by outcome quality or speed. They also judge whether the AI acted within the goals, constraints, permissions, and remedies that they authorized. We develop Delegated Agency Theory (DAT) to explain this judgment through the concept of delegation fidelity, which refers to users’ perceived alignment between the delegation contract they believe they set and the behavior they later experience. We argue that delegation outcomes depend on four higher-order design dimensions: contract specification quality, capability-context fit, governance visibility, and recovery capacity. These dimensions reduce adverse selection before handoff and moral hazard during execution, thereby increasing delegation fidelity. Higher delegation fidelity improves users’ functional, symbolic, and ethical evaluations, which together shape value-in-delegation, continued reliance, complaint intention, and switching. By positioning delegation fidelity as the central mediating mechanism, DAT offers a more precise explanation than that provided by adjacent concepts such as trust, transparency, or algorithmic appreciation alone. We also outline an empirical pathway through the user Delegation Quality Index (uDQI), which captures the quality of delegation design and supports future measurement and validation across contexts and cultures.</p>