Guayoyo Tech Rebrands in 48 Hours with AI Design Skills

Guayoyo Tech transformed its AI-generated website aesthetic in under 48 hours. The founder used Claude Code as a designer, feeding it explicit design constraints to move beyond generic 'AI Slop' interfaces.

Guayoyo Tech, an AI automation company, overhauled its website's visual identity in less than 48 hours. The process involved modifying 62 files and executing 30 commits. The core of this rapid rebrand was the strategic deployment of Claude Code, an AI agent, which was explicitly instructed with a set of design constraints to act as the project's designer. This approach moved the site from a generic, AI-generated aesthetic to a distinct brand identity featuring an amber palette, editorial typography, and asymmetric layouts.

Previously, the Guayoyo Tech website exhibited what the development community terms "AI Slop." This aesthetic is characterized by predictable elements: the Inter font, cyan gradients, symmetrical cards with generic box shadows, and perfectly balanced three-column layouts. The site used #06b6d4 (cyan-500) over #0f172a (slate-900) backgrounds, resulting in a visual experience described as "completely forgettable" and the "visual equivalent of elevator music." For a company specializing in AI systems and automation, this generic appearance was a counter-indicator, signaling that an unconstrained AI agent had designed its own public face.

Curbing Generic AI Slop

The problem of "AI Slop" stems from how large language models (LLMs) approach design. LLMs lack inherent aesthetic judgment. When tasked with generating an interface, they default to statistically probable patterns, often resulting in bland, repetitive designs. This was the exact issue Guayoyo Tech faced. Their initial site was functional but lacked any unique personality, resembling countless other AI-generated projects.

The founder recognized that simply changing the AI model or writing longer, more elaborate prompts would not solve the fundamental issue. The solution required a different strategy: providing the AI agent with explicit, codified design constraints. This insight formed the basis of their rebranding experiment, which aimed to test this theory on their own product.

Explicit Design Constraints

The catalyst for Guayoyo Tech's rebrand was an internal analysis of how to imbue AI agents with better design taste. This research culminated in an article titled "Skills de Diseño para Agentes de IA," which explored five tools for teaching aesthetic judgment to code agents. The central premise articulated in that piece was clear: "Los LLMs no tienen criterio estético. Cuando generan una interfaz, aplican patrones estadísticamente probables. La solución no es cambiar de modelo ni escribir prompts más largos. Es darle al agente constraints explícitas de diseño." This translates to: LLMs lack aesthetic judgment; they apply statistically probable patterns. The solution is not to change models or write longer prompts, but to give the agent explicit design constraints.

This principle guided the entire rebranding effort. The experiment involved a four-step process: defining a comprehensive design system in an executable file, feeding this file to Claude Code, instructing the AI to rewrite each section of the website, and then refining the output through iterative commands. This structured approach ensured that the AI operated within a predefined aesthetic framework, preventing it from reverting to generic patterns.

CLAUDE.md as the Design System

The foundational step was establishing absolute design restrictions. This was not about vague instructions like "make it more modern" but concrete, enforceable rules. These rules were codified within a CLAUDE.md file, which served as the executable design system for the AI agent. This file contained specific directives, acting as guardrails for Claude Code's design output. Key rules included:

• Never use Inter, Roboto, or Open Sans fonts.
• Never use cold gradients, specifically purple, cyan, or blue.
• Never use more than three colors per component.
• Never use generic box-shadow values such as 0 4px 6px -1px.
• Avoid symmetric cards in layouts.

These explicit negative constraints forced the AI to explore alternatives outside its default, statistically probable patterns. By defining what not to do, the founder guided Claude Code towards a more distinctive and less predictable visual language. This method transformed the AI from a generator of generic templates into an executor of specific, human-defined aesthetic principles.

Iterative Site Rewrite

With the CLAUDE.md file established, the next phase involved applying these rules across the entire website. Claude Code was given the design system and then instructed to rewrite each section of the site. This was not a single-pass operation but an iterative process. The founder provided refinement commands, guiding the AI to adjust and improve its output based on the established constraints and desired aesthetic.

This iterative feedback loop allowed for continuous alignment with the new visual identity. The result was a complete transformation of the site's front-end code, reflected in the 62 modified files and 30 commits logged over the less-than-48-hour period. The process demonstrated that an AI agent, when properly constrained and directed, could efficiently implement a complex design overhaul without requiring manual design work or direct business logic modification.

WHAT WE'D CHANGE

The Guayoyo Tech playbook for leveraging AI as a designer provides a clear path to escaping "AI Slop." However, its direct application for all founders in 2026 requires consideration of specific contexts and potential modifications. While effective for a rapid rebrand, relying solely on a CLAUDE.md file for a comprehensive, evolving design system might introduce maintenance challenges. Traditional design systems, often managed through tools like Figma or Storybook, offer robust version control, collaborative features, and component libraries that facilitate ongoing development and scaling across larger teams or more complex products. A CLAUDE.md approach, while agile for a rebrand, could become cumbersome for long-term component management and cross-functional design handoffs.

Furthermore, the success of this method hinges on the founder's ability to articulate and codify "good taste" into explicit, machine-readable rules. Not every founder possesses this specific blend of design acumen and technical precision. The initial investment in understanding and defining these design "skills" is substantial, as evidenced by Guayoyo Tech's prior article on the subject. Founders without a strong design background or the capacity to translate aesthetic principles into granular constraints might struggle to achieve similar bespoke results, potentially reverting to generic outputs or requiring external design consultation to define the initial CLAUDE.md ruleset.

Finally, the specific reliance on "Claude Code" suggests a model-dependent workflow. While the underlying principle of explicit constraints applies broadly to LLMs, the exact implementation and effectiveness may vary with different AI agents. Future iterations of this playbook might need to account for model-specific prompt engineering or tool integrations to ensure portability and consistent results across various AI platforms.

Guayoyo Tech's experience demonstrates a clear path for founders to move beyond the generic outputs of unconstrained AI. By treating the AI agent as a highly efficient executor rather than a creative director, and by providing it with a meticulously defined set of aesthetic rules, founders can achieve a distinct visual identity with unprecedented speed. This approach underscores that the future of AI-assisted design lies not in more powerful models, but in the human capacity to translate nuanced aesthetic judgment into explicit, machine-readable constraints, thereby transforming AI from a source of "slop" into a tool for bespoke creation.

Pull quote: “Los LLMs no tienen criterio estético. Cuando generan una interfaz, aplican patrones estadísticamente probables. La solución no es cambiar de modelo ni escribir prompts más largos. Es darle al agente constraints explícitas de diseño.”