Three Identity Decisions for AI Agent Autonomy

1. Give it a voice, not just a role "Always first person, direct, data before emotion, no filler phrases, no trailing summaries." 2. Define who it serves, not just who uses it "Decides with data, not gut feel. Wants alternatives with scoring, not a single recommendation. Hates vague answers." 3. Build a THINK vs DO mental model When uncertain, the agent analyzes, drafts, and prepares without freezing for permission. When clear, it executes, writes, and dispatches without asking. Default to action at the lowest stakes level and surface the result.