design-companion
Design Companion Skill
Specialized knowledge for reviewing designs against human-centric AI principles.
When to Use
- Reviewing design briefs
- Evaluating prototype designs
- Checking AI feature implementations
- Accessibility audits
- Trust and emotion assessments
Quick Trust Check
Before approving any AI feature design:
- User understands what AI will do before it acts
- User can see evidence for AI decisions (receipts)
- User can easily undo AI actions
- User can correct AI and see learning
- AI admits uncertainty appropriately
- AI failures are graceful and recoverable
State Design Requirements
Every AI feature needs all states designed:
| State | Required Elements |
|---|---|
| Loading | Progress, "AI is thinking" for screen readers |
| Success | Confirmation, what happened, undo option |
| Error | What went wrong, how to fix, alternative path |
| Low Confidence | Muted styling, hedging language, verification prompt |
| Empty | Encouraging message, actionable next step |
Emotional Design Checklist
Visceral (First Impression)
- Does it look trustworthy and polished?
- Is AI visually distinguished from static content?
- Do animations convey intelligence without gimmicks?
Behavioral (During Use)
- Is response time acceptable (faster than manual)?
- Are interactions predictable and consistent?
- Can users easily dismiss or correct AI?
Reflective (After Use)
- Does user feel augmented, not replaced?
- Would they recommend to a colleague?
- Do they feel more capable, not dependent?
Persona Concerns
Sales Reps Fear
- Replacement ("Will AI take my job?")
- Surveillance ("Is AI tracking my performance?")
- Embarrassment ("Will AI make me look bad?")
Design Response: AI helps YOU, doesn't report ON you. Rep gets credit. Suggestions private until rep acts.
Sales Managers Fear
- Losing touch with team
- AI replacing judgment on people
- Creating surveillance culture
Design Response: AI surfaces insights, manager decides. Coaching, not scoring.
Accessibility Quick Check
- All AI features keyboard-accessible
- Dynamic content uses
aria-live - Color is not sole information carrier
- Reading level ≤ 8th grade for important info
- Animation respects
prefers-reduced-motion - Alternative text for visual AI elements
Anti-Patterns to Flag
🚩 Confident wrongness - AI asserting incorrect info confidently 🚩 Unexplained actions - Acting without user understanding 🚩 Silent failure - No response or vague deflection 🚩 Over-automation - Acting without authorization 🚩 Creepy personalization - Using undisclosed data 🚩 Replacement framing - Suggesting AI replaces human value
Copy Guidelines
Loading States
- Short: No copy needed
- Medium: "Analyzing your conversation..."
- Long: "Processing 45-minute call... this may take a moment"
Error States
- Be specific: "Couldn't connect to HubSpot" not "Something went wrong"
- Offer action: "Try reconnecting"
- Never blame user
Low Confidence
- Hedge: "I think...", "This might be...", "Consider..."
- Invite verification: "Does this look right?"
- Offer alternatives: "Or did you mean...?"
Review Questions
When reviewing prototypes, ask:
- Would a user trust this immediately?
- Does user understand what AI can/can't do?
- What happens when AI is wrong?
- Is AI faster than the manual alternative?
- Can user see WHY AI made this decision?
- Can user easily override or correct?
- Does this work for all users?
Reference
Full research: pm-workspace-docs/research/human-centric-ai-design-research.md