Firms are increasingly adopting predictive artificial intelligence (AI) to improve decision-making by combining advanced data analysis with managerial judgment. While AI provides more precise information to support managerial decision-making, its adoption can nevertheless reduce shareholder profits and the aggregate welfare, when both managerial and AI-provided information on project quality are highly precise and managerial bias is uncertain ex ante. In such cases, investment misallocation worsens, with overinvestment in low-quality projects and underinvestment in high-quality ones, even after the manager's compensation contract has been re-optimized. We propose remedies through information design and contracting approaches. Our analysis underscores the importance of governance reforms to guide AI adoption beyond simply enhancing AI's precision.