AI feature brief
Define an AI capability with clear responsibility, human control, and failure boundaries.
Product features · 60 min · Advanced · 39 questions
Model responsibility
Which decision may the model make?
Which recommendation may the model provide?
Which output must remain a draft?
Input rules
Which input is required?
Which input is optional?
Which input is prohibited?
What should happen when required input is missing?
What should happen when input may be biased?
What should happen when input is stale?
What should happen when someone tries to manipulate the model through its input?
Data privacy
Which input may be sent to a model provider?
Which AI interaction may be logged?
How long may AI interaction data be retained?
May interaction data be reused for improvement?
Output confidence
How should low-confidence output appear?
How should disagreement between outputs appear?
What should happen when an output lacks evidence?
Human review and control
Which AI output requires human review?
Who is qualified to review that output?
How quickly must review happen?
How can people correct AI output?
How can people undo an action based on AI output?
How can people reject an AI suggestion?
Safety and fallback
Which harmful request requires refusal?
What fallback exists for misleading output?
What fallback exists when the model is unavailable?
Evaluation
Which realistic examples should we use to evaluate the feature?
Which high-risk user or situation needs separate evaluation?
Which rubric defines an acceptable output?
Which baseline must the model outperform?
What response latency is acceptable?
What cost per completed task is acceptable?
Launch and monitoring
Which signal should pause the AI rollout?
Which signal should reverse the AI rollout?
How will model quality be monitored after launch?
How will the team detect that model quality has changed over time?
How will abuse be detected?
How will provider changes be evaluated?
How will production cost be monitored?
How to use this brief
Use this brief as a bank of questions for the conversation. Choose what is relevant, mark covered questions as you go, and skip anything that has already been answered.
Use this brief when...
Use when you need to define an AI capability with clear responsibility, human control, and failure boundaries.
Best used as: Live call · Async intake
- Category
- Product features
- Discussion time
- 60 minutes
- Depth
- Advanced
- Last updated
- Jul 21, 2026
- Scope
- Design briefs
- Best used as
- Live callAsync intake
- Design roles
- Product designer
- Audience
- Product designersProduct managersDesign leadsFounders and operators
- Platforms
- WebMobileAPICross-platform
- Company type
- StartupB2BB2CEnterprise
- Product stage
- IdeaEarly-stageGrowthMature product
- Tags
- product-featuresai-featuresystems-and-platformsgenerative-aipredictive-systemhuman-ai-workflow