Why apps built with AI look a little… OFF

Why apps built with AI look a little… OFF

More

Descriptions:

Brian Casel addresses one of the most common frustrations in AI-assisted development: “design drift,” where an agent working across multiple sessions gradually produces UI that diverges in spacing, button styles, and typography from screen to screen. His diagnosis is structural — agents lack a persistent source of truth about what the app is supposed to look like — and his fix is to embed a living design system directly in the codebase rather than relying on Figma mockups or per-prompt style instructions.

The video walks through the “BM Design System,” a free agent skill Casel built and released via Builder Methods. It’s installable in Claude Code through the plugin marketplace with a single command and establishes a shared component library covering colors, typography, spacing, buttons, forms, badges, modals, listings, and iconography (primarily using Lucide icons). Each component includes documented variants and usage guidance that AI agents can reference during generation. Casel demos the system across four separate internal tools he’s actively using — a content strategy manager, an ideation tool, a YouTube production tracker, and a social post publisher — showing how changing a single primary color propagates consistently through every component across all four apps.

A key insight in the video is how this approach differs from traditional UI component libraries: instead of rigid, hard-to-customize pre-built elements, the design system lives as real, editable code that agents can extend without breaking consistency. For developers building multiple AI-assisted applications, the system provides a practical framework for maintaining brand coherence across long-running projects.


📺 Source: Brian Casel · Published June 05, 2026
🏷️ Format: Tutorial Demo

1 Item

People