Rating
- size:
3 - complexity:
4
Purpose
Practice turning vague “human in the loop” expectations into real verification standards for a specific workflow.
Scenario
A team wants to use AI heavily in backlog clarification and acceptance-criteria drafting.
Current policy:
A human must review AI output before use.
That is the only standard.
Task
Use AI to help draft a first verification standard for this workflow, then improve it manually.
Your result should define:
- what must be checked
- who should check it
- what evidence counts as real verification
- what kinds of ambiguity should trigger escalation
Expected output
- draft verification standard
- list of failure modes the standard is meant to prevent
- short note on what remains practice-based versus evidence-backed
Good AI uses
- proposing checklist categories
- surfacing missing review dimensions
- comparing weak versus strong standard language
Verification focus
- the standard must be specific enough to change behavior
- “review” must be operationalized, not merely named
- low-observability risks should receive stronger treatment
Anti-patterns to watch
- writing a policy that sounds serious but changes nothing
- using AI to generate a generic checklist detached from the workflow
- forgetting escalation paths for unresolved ambiguity