AI FIELD NOTES

A review-first workflow for AI-assisted work

Make the task specific, protect your inputs and check the output.

Define the task before the prompt

Write down the result you need and what would make it acceptable. A clear task such as drafting a product description from approved specifications is easier to evaluate than a request to improve everything.

Share only appropriate context

Remove passwords, personal information and confidential business data from examples. Use sample data where possible and understand your chosen tool’s data handling before sharing business information.

Treat the output as a draft

Check factual claims against reliable source material. Run and review generated code in an appropriate test environment. Look for invented details, missing edge cases and assumptions that were never agreed.

Keep a person responsible

For actions that send messages, publish content or change systems, define a review step. Record what worked, what failed and what needs a fallback before expanding the workflow.

Try it on your own project: write a short checklist for the next step, then explain why each item matters.

← Back to all guides