Characterizing AI Agents for Alignment and Governance