Summary of 1006: Can AI Make Good Design?

Overview of Syntax Episode 1006: Can AI Make Good Design?

Wes Bos and Scott Tolinski explore a central question in the AI era: can AI actually make good design, or does it mostly produce polished-looking sameness? Their take is nuanced. AI can be extremely useful for speeding up repetitive work, exploring variations, and generating decent starting points—but it struggles with true creativity, UX judgment, and taste. The episode argues that AI is best treated as a tool for busywork and pattern execution, not a replacement for human designers who understand users, goals, and context.

Main Question: Can AI Make Good Design?

Short answer

• AI can make passable or even attractive design outputs
• It does not reliably create original, thoughtful, or user-centered design
• It tends to produce derivative, statistically common, “probable” design, not genuinely creative work

Core distinction

• Code is often deterministic and testable, so AI helps there more easily.
• Design is subjective, contextual, and tied to business outcomes and user behavior—making it much harder for AI to do well on its own.

AI and Creativity: Helpful, But Not Truly Creative

What AI does well

• Generates many variations quickly
• Helps explore ideas and directions
• Can accelerate brainstorming and iteration
• Useful for starting from a rough concept or mood

What it does poorly

• Inventing genuinely new ideas
• Developing a distinctive point of view
• Creating something that feels intentionally crafted
• Avoiding “samey” outputs trained on existing patterns

Notable example

• Wes describes seeing the same AI-generated testimonial names and patterns across many sites, especially names like Sarah Chen and Marcus Rodriguez.
• The point: AI is often recycling common data patterns, not being creative.
• Their view: if a site uses fake AI-generated testimonials, that crosses into fraudulent-looking slop.

Design Trends Are Becoming Table Stakes

Example: YouTube thumbnails

• AI can now generate the classic high-contrast, “Mr. Beast-style” thumbnail:
  • exaggerated expressions
  • bold split colors
  • huge text
  • dramatic composition
• The hosts argue this is already becoming overused and less effective because everyone can do it now.
• Once a style becomes instantly reproducible, it stops being a differentiator.

Broader takeaway

• AI accelerates trend adoption.
• As a result, design trends may cycle faster and become more generic faster.

Can Good Design Be Extrapolated Programmatically?

The answer: partially

AI can extend a strong design system or set of patterns, especially when:

• the rules are clear
• the inputs are structured
• the goal is repetition, not invention

Examples mentioned

• Google Stitch
  • Can generate decent layouts from structured prompts
  • Works better when given clear feature summaries and design constraints
  • But its outputs quickly become recognizable and repetitive
• Design.md
  • Discussed as a structured design spec/steering document
  • Intended to codify design tokens, rules, and constraints for AI tools
  • Wes and Scott are skeptical about adding yet another file that can drift from the actual codebase

Their concern

• Design rules already exist in many projects as:
  • CSS variables
  • design systems
  • component libraries
• They worry a separate AI-facing design document adds complexity and drift without solving the root problem.

Can AI Make Good UX?

Their answer: mostly no

AI can place components on a page, but it struggles with:

• understanding user intent
• reducing friction
• making the right interaction tradeoffs
• designing flows for real-world tasks

Why UX is harder than layout

Good UX requires:

• context
• empathy
• business understanding
• awareness of edge cases
• knowledge of what users are actually trying to accomplish

Example: U-Haul

• They use U-Haul as a case study for bad UX.
• A useful UX would help users:
  • find a trailer
  • understand availability nearby
  • recover gracefully when an option isn’t available
• AI may create something that “looks like an app,” but not one that handles these complex workflow needs well.

Important distinction

• AI can make things look polished.
• It usually cannot make them easy to use.

Can Good Design Be Programmatic?

Partially yes, but with limits

They draw a distinction between:

• programmatic design
• maintainable design
• good taste

Some things work well programmatically:

• text-wrap balancing
• contrast calculations
• background removal
• layout rules
• component generation from tokens

But other things do not:

• color harmony
• visual taste
• brand feel
• aesthetic judgment
• “this just looks good”

Example: Warp Terminal theming

• A programmatic color/theme system can look technically correct but still aesthetically weak.
• A designer who manually controls the system often gets better results.

Their broader point

• Design systems are useful because they create consistency.
• But even with systems, humans still need to make judgment calls.
• “Programmatic” and “good” are not the same thing.

AI and Business Outcomes

Can AI influence the end goal?

Sometimes, yes—especially when backed by data.

Examples:

• TikTok and Twitter/X-style recommendation systems
• A/B-tested landing pages
• algorithmic optimization for conversions
• time-sensitive information design, like Google’s response to tornado searches

But:

• Marketing and conversion patterns change quickly.
• Once a tactic becomes widely known, it becomes less effective.
• AI can help optimize toward outcomes, but it also makes those outcomes easier to copy.

The Aesthetic Backlash: Human and Imperfect May Win

Emerging trend

As AI-generated content becomes more polished and synthetic, people may gravitate toward:

• imperfect visuals
• handmade aesthetics
• “real” photography
• human texture and flaws

Examples mentioned

• AI video and AI influencers becoming harder to distinguish
• overly perfect lighting or production quality being perceived as “AI-ish”
• a potential push toward more authentic, less glossy visual styles

Implication

• The future may reward design that feels human, not just clean.

Best Uses for AI in Design Workflows

Good use cases

• background removal
• slicing and naming assets
• resizing and trimming images
• generating variations for exploration
• summarizing features into layout-ready copy
• speeding up repetitive production tasks

Example from the episode

Wes used Claude to:

• remove backgrounds from a grid of icons
• trim transparent pixels
• slice them into separate files
• rename them correctly

This saved substantial time and was a strong example of AI handling obnoxious, repetitive work.

Their recommended mindset

Use AI to:

• remove busywork
• accelerate established workflows
• help with exploration and summarization

Don’t use AI to:

• replace taste
• decide UX
• invent brand identity from nothing

Key Takeaways

• AI is good at repeating patterns, not creating original taste.
• Design quality depends on judgment, context, and user needs.
• AI can help with layout and production tasks, but not full UX thinking.
• The best use of AI in design is to eliminate tedious work, not to outsource the creative core.
• As AI-generated aesthetics spread, more human-looking design may become more valuable.

Final Verdict

Wes and Scott’s conclusion is essentially:

• AI can make decent design artifacts
• It can speed up designers dramatically
• It can assist with pattern-based or repetitive work
• But good design still requires a human who understands people, goals, and taste

In other words:
AI can help you design faster, but it cannot yet reliably design well on its own.