AI work intake

A short template for turning fuzzy AI requests into bounded work. Use it before asking an assistant or agent to research, write, inspect, change, or automate anything meaningful.

How to use it

Fill the six fields before starting the AI session.
Keep the task small enough that a human can inspect the result in one sitting.
Make the verification step explicit, even when it is only a checklist.
Save strong versions as reusable prompts, playbooks, or agent skills.

Copyable intake brief

AI Work Intake

Outcome:
- 

Context:
- 

Boundaries:
- 

Success criteria:
- 

Verification:
- 

Return format:
- Files or artifacts changed:
- Evidence used:
- Checks run:
- Risks, assumptions, or open questions:

Outcome

Prompt: What decision, artifact, or change should exist when this is done?
Example: A one-page recommendation with risks, trade-offs, and the next concrete action.

Context

Prompt: What should the AI read, inspect, or treat as source material?
Example: Use the linked project brief, the current customer notes, and the existing page patterns.

Boundaries

Prompt: What must not be touched, assumed, rewritten, published, or automated?
Example: Do not change pricing, do not email customers, and do not edit the newsletter flow.

Success criteria

Prompt: What makes the answer usable instead of merely plausible?
Example: It cites the evidence used, names open questions, and gives one recommended path.

Verification

Prompt: What should be checked before the work comes back?
Example: Run the narrow test or checklist that proves the changed surface still works.

Return format

Prompt: How should the AI report the work so a human can inspect it quickly?
Example: Summarize files touched, decisions made, evidence used, and anything not completed.