Designing Beyond AI Defaults: Magic Patterns and Website Reviews

In this Y Combinator Design Review, Aaron Epstein interviews Magic Patterns co-founder Alex Danilowicz about AI-assisted prototyping and reviews startup sites with him. Magic Patterns is a YC company, so the demonstration and praise are not independent product evaluation.

Danilowicz says he and his co-founder were front-end engineers seeking a more intuitive way to turn ideas into visual prototypes. After several pivots, he says the product found paying users who returned to it.

He demonstrates a canvas that can combine prompts, images and documents, generate interactive designs and connect to coding tools through MCP. The team uses it to explore variants and states, such as loading indicators and screen sizes.

The canvas can make several threads and pieces of context visible together, but Danilowicz says it is not necessarily right for every user. He describes onboarding as a challenge and distinguishes the canvas from a traditional graphics editor.

For better results, he recommends providing relevant context such as a design system, components and project instructions. A playful theme-park example illustrates how defined components can constrain a generated interactive result.

Reviewing startup pages, the two discuss whether a visitor can quickly understand the product and its intended audience. They recommend examining copy, hierarchy and visuals together instead of accepting a generic generated result.

Epstein cautions that visual defaults can become dated and suggests choosing typography and a brand identity that communicate a company’s particular message. Danilowicz says he wishes his team had established a brand book earlier.

In one review, they find the visual style polished but the description too vague to explain what the product does. They suggest giving concrete user problems and examples, and testing the page with actual users.

For an inbox product, they praise showing the product but question whether a fast, detailed demo makes its benefit easy to grasp. They also discuss selecting social proof relevant to the audience and explaining security when users are asked to connect sensitive data.

Danilowicz says design taste involves looking at things one finds compelling and considering the user’s experience, not just polishing individual pixels. He says everyone at his company participates in customer support to hear problems directly.

On speed versus quality, the pair argue that extra attention to confusing details can be worthwhile even when development is faster. Their closing predictions about future design tools are opinions, not guarantees.