Use AI as a tool,
not as the final judge.

01 / FIRST TASK

Begin with reversible work

Draft an email, summarize text you already read, compare options, or turn notes into questions. Avoid making legal, medical, financial, publishing, deletion, or payment decisions your first experiment.

02 / PROMPT

Separate four parts

Goal: what should be completed
Context: source material and assumptions
Constraints: what to do and avoid
Output: the format you want
Ask before working if information is missing

This creates a review standard. You can see whether the result met the goal instead of judging whether it merely sounds good.

03 / CONTEXT

Share only what the task needs

Provide the target text, audience, prior decisions, and open questions. Separate confirmed facts from assumptions. Do not add unrelated private material just because a larger context window exists.

04 / VERIFY

Check evidence, not fluency

Open cited pages. Check dates, regions, original wording, and official sources. Recalculate important numbers. Asking for citations helps, but a plausible citation can still be wrong.

05 / PRIVACY

Keep secrets out of the prompt

Do not paste API keys, passwords, private keys, customer data, or unpublished material. Redact or replace real values with fictional examples. Review the current terms and data controls of the service you use.

06 / REVISE

Return the difference

Instead of only saying “wrong,” identify what to keep, what to change, the priority, and what must not be invented. AI work is an iterative review process, not a one-shot guessing game.

07 / AGENT

Permissions matter when AI can act

Coding agents can read files, edit them, and run commands. Start in a small project folder, ask the agent to explain before editing, and review the diff and tests afterward.