AI as a Design Partner
A semester-long exploration of AI tools in design practice—from journey mapping to prototyping. This guide distills real experiences, successful patterns, notable failures, and actionable strategies for designers navigating AI collaboration.
Guide Navigation
Jump to any section below, or read sequentially for the full story.
  • Where AI Excelled (what works)
  • When AI Failed (what doesn't)
  • Red Flags: When Not to Trust AI
  • Prompting Strategies That Worked
  • Tool Performance Matrix (which tool for what)
  • Designer's Field Guide to AI (human vs AI roles)
  • Sprint Mapping Success (bill-splitting journey map)
  • Prototyping Breakthrough (Figma Make + Lovable)
  • The Storyboard Breakdown (what went wrong)
  • Risk Assessment, Success Metrics, & Roadmap for V2
FOUNDATIONS
Understanding What Works and What Doesn't
Before you start experimenting with AI, understand its strengths, limitations, and red flags. These lessons come from real failures and successes.
Where AI Excelled
Structuring Content
ChatGPT transformed team discussions into polished Team Charters and organized proposals. Saved hours by creating scaffolding we refined together.
Expanding Thinking
Generated wide-ranging CMU-relevant challenge spaces and helped refine fuzzy topics into crisp "How might we..." statements for ideation.
Revealing Complexity
Service blueprints gained depth when AI identified hidden backend processes—data sync, conflict resolution algorithms, ML heuristics we'd overlooked.
When AI Failed
Journey Map Confusion
Only one person prompted AI—rest of team couldn't distinguish AI-generated content from human writing. Made editing difficult and interpretation unclear. Added unnecessary work instead of streamlining.
Storyboard Disaster
AI struggled with glare removal, failed to generate correct captions after multiple attempts, broke after few prompts producing repetitive images. Couldn't maintain visual consistency across frames.
Usability Testing Flop
AI performed extremely poorly as simulated user. Provided generic observations, failed to identify specific buttons or interactions. Feedback was vague and useless compared to real user testing.
Red Flags: When Not to Trust AI
1
Generic Advice
Suggestions like "add gamification feedback loops" without context. AI defaults to buzzwords when lacking specifics.
2
Emotional Interpretation
AI cannot reflect on personal experiences. Diary entries and emotional nuance require human authorship.
3
Visual Consistency
Repeated image generation breaks down. AI struggles maintaining style, color palette, character consistency across frames.
4
Complex Interactions
Prototypes break with advanced features. Missing breadcrumbs, undo buttons, error prevention—design staples AI overlooks.
5
Social Nuance
AI assumes unrealistic automation like resolving all disputes without human oversight. Misses power dynamics and decision authority.
PRACTICAL STRATEGIES
How to Prompt, Which Tools to Use, and How to Collaborate
Now that you understand the landscape, here's how to actually work with AI. Proven prompting techniques, tool comparisons, and the human-AI partnership model.
Prompting Strategies That Worked
Provide Rich Context
Share class context, assignment briefs, mention you're CMU students. Specificity yields relevance.
Dialogue Over Dump
Let AI ask clarifying questions. Claude's facilitation approach produced better results than one-shot prompts.
Designate AI Facilitator
One team member guides prompts and translates results. Prevents confusion about AI vs. human content.
Question Everything
AI sounds logical but misses emotional truth. Keep ideas that clarify complexity, reject oversimplifications.
Tool Performance Matrix
Each tool has distinct strengths. Success depends on matching tool capabilities to specific design tasks and maintaining human oversight throughout.
Designer's Field Guide to AI
What Stays Human
  • Empathy: Understanding emotional nuance behind user needs
  • Judgment: Deciding which AI suggestions serve the vision
  • Craft Intimacy: Small discoveries from pushing pixels around
  • Decision Authority: Final call on trade-offs and priorities
What AI Accelerates
  • Scaffolding: First drafts, structures, templates
  • Exploration: Multiple variations for curation
  • Synthesis: Organizing transcripts, notes, research
  • Technical Translation: Turning concepts into code

Core Truth: AI amplifies designers who wield it strategically. It's a tool for acceleration, not replacement. The future isn't "AI replaces designers"—it's "AI empowers designers who understand its limits."
CASE STUDIES
Real Projects, Real Failures, Real Breakthroughs
See how these strategies played out in practice. Three detailed case studies from a semester-long bill-splitting app project—what worked, what didn't, and why.
Sprint Mapping Success
The Challenge
Reworking our sprint map for bill-splitting between friends. We needed to identify missing steps and pain points in the user journey.
AI's Contribution
ChatGPT identified a crucial missing step: tracking payments across Venmo, Cash, and Zelle. Also surfaced specific touchpoints like card machines and kiosks we'd overlooked.

Key Insight
"AI helped us cover specific touchpoints like card machines, kiosks or point of sale, which we missed earlier. It pushed us to think of emotional aspects of the splitting experience."
Success Pattern
  • Feed existing work for refinement
  • Ask AI to identify gaps
  • Use output to trigger team discussion
  • One person facilitates AI, team questions outputs
Prototyping Breakthrough
1
Input
Fed future-state blueprint into Figma Make with three-word prompt: "design this app"
2
Output
Captured intended flow with accurate logic. Basic visuals but solid starting point for refinement.
3
Refinement
Claude acted as design facilitator, asking clarifying questions about thresholds, confirmations, must-have features.
4
Result
Semi-functional prototype in hours instead of days. Enabled earlier technical conversations with developers.
Both Lovable and FigmaMake captured core golden path well but exposed critical assumptions we hadn't interrogated—like how users share dishes (wanting 1/3 of appetizer, not binary all-or-nothing).
The Storyboard Breakdown
What We Tried
Uploaded whiteboard sketch photo to ChatGPT, asked for refinement. AI struggled with glare and reflection marks for extended time.
The Problems
  • Failed to generate correct captions despite clear instructions
  • Broke after few prompts—repetitive, inaccurate images
  • Couldn't establish consistency across panels
  • Ignored explicit requests to remove "Panel 1," "Panel 2" labels

Lesson Learned: Narrating story first, then generating images would have been better approach. Visual consistency requires human oversight.
Made with