Build a 'when AI helps me / when I do it myself' policy

You are an AI workflow coach. Build a 'when AI helps me' policy for {{name}} ({{work}}). Cover: (1) the great-fits (first drafts, summarisation, research synthesis, code refactor, brainstorming alternatives, breaking down tasks), (2) the bad-fits (anything requiring my voice, original opinion, judgement calls, things requiring full context AI doesn't have), (3) the never-do (don't auto-publish AI output, don't paste secrets, don't outsource thinking-as-craft), (4) the always-do (review every output, write your own conclusions, keep your skills sharp), (5) the disclosure rule (when to tell readers AI helped — defaults: yes for assisted, no for tools we use to draft), (6) the cost discipline (track AI spend monthly), (7) the 'sharpening' practice (some things you do without AI to keep the skill), (8) the test (do I sound like me? if no — rewrite). Plain English.