How to Use AI Agents to Create More Distinctive Designs

Summary

Anshu Chimala argues that AI-generated interfaces often look generic because language models are trained to choose predictable, broadly acceptable continuations. Drawing on his experience leading exploratory AI product work at Apple, he adapts a Double Diamond-inspired process for teams of AI agents: discover, define, and deliver. The discovery stage explores a wide design space through bold briefs, seed strings, and stronger references instead of asking for vague originality. His String Seed of Thought technique has the model generate a random alphanumeric string, derive a visual direction from it, and use that direction to create more varied layouts, colors, and typography. He also recommends starting with broad ideas, actively refining the directions that match the user’s taste, and turning the chosen direction into an implementation prompt. In the definition stage, a separate design-critic agent reviews screenshots without access to the implementation context, evaluates the design against a clear quality bar, and gives focused feedback while a cheaper agent handles revisions. Image generation and video models can add richer visual material, animated graphics, and transitions that coding agents often omit, but the article stresses that generated media should serve the design rather than decorate it. The final delivery stage is largely editorial: remove unnecessary effects, labels, containers, and custom controls, preserve useful native components, and inspect the result for recognizable AI patterns. The overall argument is that distinctive AI-assisted design depends less on finding a different model than on injecting variety, supplying human taste, separating critique from implementation, and applying disciplined restraint.