Why is AI so bad at design?

Video thumbnail: Why is AI so bad at design?
Jul 2, 202655s video lengthLenny's Podcast

The Signal

Current AI in design struggles to move beyond stylistic imitation, failing to bridge the gap between simple component updates and true semantic abstraction. While machines reliably replicate successful aesthetics—like the widely mimicked Linear website—they lack the capacity to preserve underlying interaction patterns across visually diverse interfaces, leaving deeper design automation currently unreachable.

The Case

Design vs. Software Engineering

  • Software engineering benefits from over-indexing on known, repeatable patterns, whereas effective design generation requires an element of randomness and genuine novelty to be successful.0:17
  • Recalling the flood of websites that simply copied Linear—a software project management tool widely praised for its design and taste—the speaker argues that mere stylistic reproduction is an insufficient benchmark for true design intelligence.0:01

The Abstraction Challenge

  • A shallow rebrand might only involve mechanically updating hundreds of components, such as the 263 distinct pieces mentioned as a baseline task, but this fails to address the deeper problem of consistency.
  • The core difficulty is maintaining semantic abstraction, where two visually different elements must still convey the same interaction pattern and function to the user.0:38
  • Current technology remains limited in its ability to handle this level of abstraction reliably; it can mimic a look but not necessarily translate the intent behind the structure across new visual forms.

The 1 Minute Signal Take

Do not confuse visual imitation with meaningful design capability. True automation in this space requires models that understand semantic relationships rather than just reproducing the surface style of popular incumbents.

Pro Analysis

Why It Matters

Design generation is the current frontier for AI, representing the leap from deterministic pattern matching to creative problem solving. If we misdiagnose the difficulty of this task, we will continue to pour resources into model training that produces beautiful, yet empty, visual echoes.

Strategic Implications

Businesses should be wary of viewing 'AI-design' tools as creative partners until they move beyond component-level modification. For product managers, the implication is that current automation can reduce drag on mundane tasks, but it cannot displace the human need for the abstraction and semantic intent necessary for a cohesive brand identity.

Evidence & Hype Audit

The content relies on an anecdotal observation of market trends—specifically the 'Linear-copy' phase of web design. While the claims about the difficulty of semantic abstraction are theoretically sound based on current transformer limitations, the content lacks empirical data or technical proofs. It functions more as an expert opinion/framework for evaluating capability rather than a technical demonstration.

Counterarguments

One could argue that if an AI can consistently produce 'Linear-tier' design, the semantic abstraction problem is a matter of prompt engineering or few-shot learning rather than a fundamental limitation. If a model can be trained to 'understand' the design system tokens, it may solve the abstraction problem implicitly without needing explicit semantic reasoning.

Who Should Care

  • Product Designers: To set realistic expectations for AI tooling.
  • Technical Founders: To distinguish between commodity style-cloning and genuine value-add automation.
  • Model Developers: To prioritize abstraction-aware architectures over visual pattern matching.

What To Do Next

  • Audit current design tools to see if they perform mechanical component updates or conceptual refactoring.
  • Define formal semantic relationships within your design system to guide AI augmentation.
  • Shift evaluation metrics from 'visual output similarity' to 'functional consistency across different styles.'
  • Stop prioritizing AI models solely for their ability to replicate popular aesthetic benchmarks.

Share this

Tags

Written by: 1 Minute Signal Editorial Team