Robo-Prototyping: The Good, the Bad, and the Ugly

05/23/2025 by Craig

Robo-Prototyping: The Good, the Bad, and the Ugly poster image

As we explore or adopt AI-assisted tools to accelerate UI prototyping, it's important to acknowledge their current limitations. While AI can help with speed and surface-level output, we should be aware of potential risks in accessibility, UX design, and maintainability.

Accessibility Is More Than Checkboxes

AI can often handle:

  • Color contrast
  • Basic aria and role attributes
  • Semantic HTML

But it often misses deeper needs, including:

  • Logical keyboard navigation (especially for sighted keyboard users)
  • Focus management in dynamic UIs (modals, menus, form steps)
  • Motor accessibility (e.g. hit areas, gesture redundancy)
  • Cognitive accessibility (e.g. layout clarity, input chunking, predictable flows)

⚠️ Risk: Interfaces may pass automated audits but still be unusable for real users.

Lack of UX Flow Context

AI tends to build in isolated blocks, not end-to-end experiences.

  • No awareness of task flows or user goals
  • Missing transitions, error states, loading logic
  • Poor onboarding or progressive disclosure strategies

⚠️ Risk: We may end up with good-looking screens that don’t actually support the tasks users need to complete.

Pattern Overuse and Misapplication

AI often relies on overused or misapplied patterns:

  • Dropdowns for everything (even when radio buttons would work better)
  • Hamburger menus on large screens
  • Modals overused for content that should be inline

⚠️ Risk: We inherit UX anti-patterns that are hard to reverse later.

Visually Polished ≠ User-Validated

AI-generated prototypes can look finished, which may:

  • Give stakeholders a false sense of progress
  • Delay user testing (“let’s polish more first”)
  • Skip important steps like content modeling and interaction testing

⚠️ Risk: A polished prototype can mask deep usability issues or broken logic.

Inclusive and Ethical Design Blind Spots

AI won’t automatically:

  • Consider underserved users or edge cases
  • Avoid manipulative patterns or dark UX
  • Adapt tone for cognitive readability or inclusive content

⚠️ Risk: We might unintentionally exclude or frustrate key user groups.

Code Maintenance Challenges

Rapidly generated HTML/CSS/JS is often:

  • Poorly modularized
  • Lacking naming consistency
  • Missing state management planning
  • Fragile under future changes

⚠️ Risk: Technical debt builds up quickly and silently.

✅ Recommendation

Use AI as a starting point or "napkin sketch", not a delivery pipeline.

  • Validate AI-generated UI with UX designers, a11y experts, and devs
  • Avoid using AI output in production without review
  • Maintain source of truth in human-readable, modular components