In a recent LinkedIn post, Hiten Shah discusses the practical realities of using AI for web design, sharing insights from an extensive experiment conducted by his team at Base 44. Shah and his colleagues ran 180 AI web design prompts across multiple platforms, including Bolt, Figma Make, and Lovable, to determine if AI tools could truly design full websites in minutes as often claimed.
The Limitations of AI in Web Design
The core finding of Shah’s experiment is that while AI can generate visually clean layouts, it struggles with the nuanced design process that a human brings. “AI can generate clean layouts, but it can’t design in the way a person can. It needs direction. It needs a system to follow,” Shah states in his post. This highlights a critical gap between AI’s current capabilities and the holistic approach required for effective web design.
Shah’s team discovered that the structure of the prompt was paramount. When prompts were ordered, starting with layout, then styling, and finally interactions, the AI-generated designs began to show coherence. Conversely, prompts that attempted to cover all aspects simultaneously led to disjointed and unusable results.
“When we asked for everything at once, they fell apart.”
Prompt Engineering: The Key to Usable AI Designs
The experiment revealed that AI-generated prompts often surpassed those written by humans, frequently being more comprehensive and including details that designers might overlook. Shah’s team adopted a new strategy: letting AI draft the initial prompt and then refining it with human judgment. This iterative process significantly improved the quality of the output.
“AI-generated prompts often beat the ones we wrote ourselves. They were more complete. They included the small details that people forget to mention, like spacing rules, hierarchy, and interaction logic,” Shah observed.
The Crucial Role of Brand Identity
A significant factor in achieving specific and high-quality results was providing the AI with comprehensive brand identity context. When supplied with colors, fonts, tone of voice, and reference screenshots, the AI produced designs that felt tailored. Without this information, the tools tended to default to generic templates.
“Brand identity made the biggest difference. When we gave AI our colors, fonts, tone, and a few reference screenshots, it produced results that felt specific.”
Ultimately, Shah concludes that the effectiveness of AI in design is directly correlated with the user’s own clarity of thought and articulation. “AI design only gets good when your thinking does. The clearer the intent, the stronger the result,” he argues.
AI as a Collaborator, Not a Shortcut
Shah emphasizes that AI tools are best viewed as collaborators rather than shortcuts. While AI can handle the more mechanical aspects of design, such as grids, layouts, and repetitive tasks, it frees up human designers to focus on higher-level judgment and creativity. He notes, “AI takes over the mechanical parts of design such as the grids, the layouts and the repetition, so your judgment has more room to work.” Treating AI as a collaborator, rather than a way to bypass the design process, leads to increased speed and better outcomes.
“If you treat it like a shortcut, you’ll get templates. If you treat it like a collaborator, it will make you faster.”
Shah shared a link to the full results and learnings from the experiment in the comments section of his original LinkedIn post, offering further resources for those interested in mastering AI-assisted design processes.
📝 About This Content
This article is based on insights shared by Hiten Shah on LinkedIn.
📅 Originally posted on November 12, 2025 | View original post on LinkedIn →