This paper explores how humans' pre-existing beliefs about their own abilities and AI competence influence their decisions to delegate tasks to AI. It uncovers a systematic "AI optimism" bias and a disconnect between perceived efficacy and actual collaborative outcomes, suggesting that merely providing contextual information may not improve human-AI team performance.