A common founder frustration: AI can code, but its designs often feel… off. Think generic three-card layouts, those ubiquitous indigo purple gradients, or just a general lack of intentionality. Amadou Wace, co-founder of Command Code, calls this "design slop." He argues it’s not an unsolvable problem, extending his proven "repair logic" from fixing AI coding agents to cleaning up AI-generated user interfaces. This isn't just about tweaking prompts; it's about giving LLMs a deterministic framework for good taste.

Wace saw the numbers. While models like DeepSeek struggled with specific coding tasks against a Claude Opus 4.7, his "repair logic" could fix tool-calling errors deterministically, closing the performance gap. The same principle, he found, applies to design. Just as LLMs make finite, predictable coding errors, they also produce a finite set of common design "smells."

Mario Zechner, a respected developer, recently highlighted this exact issue with viral examples of AI design slop. Wace says the solution lies in a structural intervention: teaching the AI how to think about design, not just what to design. Instead of iterating on bad outputs, you give the LLM a "compositional framework" and specific rules. It’s about building guardrails so the AI can’t fall into the same traps.

Key Takeaways

  • AI-generated design often suffers from predictable "design slop," like generic layouts and jarring color schemes, akin to predictable errors in AI-generated code.
  • Command Code extends its "repair logic" system from fixing AI coding agent errors to deterministically correcting design flaws, drastically improving output quality.
  • By providing LLMs with "compositional frameworks" (e.g., 7 specific patterns for dashboard design) and enforcing strict rules (like using OKLCH color space), AI can generate designs much closer to human taste.
  • This approach means you don't need endless iterations; you can achieve high-quality design outputs with minimal reference documents—as few as 24 documents, 10 design smells, and 7 patterns.
  • Command Code's Design Slop Repair Framework offers a structured way to instill design intelligence into LLMs, closing the quality gap between AI and human designers.

The Command Code's Design Slop Repair Framework

This framework provides a structured approach to guide LLMs towards generating high-quality, human-like designs by addressing common flaws directly and deterministically.

  • Identify Design Slop Patterns: Recognize common AI design flaws such as the 'indigo purple gradient thing', generic three-card layouts, or designs lacking intentionality.
  • Apply Compositional Frameworks: Guide the LLM with a framework for design thinking, such as the 'seven patterns' for dashboard surface areas, to instill an understanding of design intent.
  • Enforce Specific Design Rules: Deterministically fix common design 'smells' and enforce better practices, such as requiring the use of OKLCH color space instead of HSL for better control over lightness and overall color palette.
  • Minimal Reference Documents: Leverage a small set of reference documents (e.g., 24 documents, 10 design smells, and 7 patterns) to encapsulate design taste and apply it consistently.

When This Works (and When It Doesn't)

This framework shines when you need consistent, high-quality design outputs for structured problems, especially those with established best practices like dashboards or landing pages. Wace highlights its effectiveness in closing the "contract gap"—the difference between what a user asks for and what an LLM delivers—by embedding specific design intelligence. It's powerful for transforming generic AI outputs into something that meets a specific aesthetic standard.

However, this approach may be less effective for highly experimental, artistic, or truly novel design challenges where existing patterns don't apply. If your goal is to push the boundaries of visual design or create something completely unprecedented, relying heavily on a framework of known patterns might limit breakthrough creativity. It's a system designed for precision and consistency within defined parameters, not necessarily for spontaneous innovation in uncharted aesthetic territory.

What to Do With This

This week, consider an internal tool or dashboard you need to build. Instead of just prompting an LLM to "create a sales dashboard," apply Wace's framework. First, identify potential "design slop patterns" you want to avoid—maybe you’ve seen those bland grey interfaces or unreadable data visualizations. Then, provide a "compositional framework." For instance, instruct the LLM to use one of the "seven patterns" for dashboard layouts that are proven to be effective for data display. Finally, "enforce specific design rules"; mandate it use OKLCH for all color palettes to give you superior control over the visual tone and ensure consistency. By doing this, you're not just correcting output; you're teaching the AI taste and structure from the start, dramatically improving the final design's utility and aesthetic appeal.