I Tested Gemini 3 as a Designer. It’s Terrifyingly Good.

Key Concepts

• Gemini 3.0 Pro: A large language model developed by Google AI, used here for design generation within Google AI Studio.
• Google AI Studio: A platform for building and prototyping with Google's AI models, including design capabilities.
• Vibe Coding: A term used to describe the process of generating code or designs based on a general feeling or aesthetic, often with less precise instructions.
• Prompt Engineering: The art of crafting effective prompts to guide AI models to produce desired outputs.
• Reference Images: Visual inputs provided to AI models to influence the style and aesthetic of generated designs.
• Product Requirements Document (PRD): A document outlining the purpose, features, and functionality of a product, useful for providing detailed input to AI design tools.
• Tailwind CSS: A utility-first CSS framework used for rapid UI development.
• React: A JavaScript library for building user interfaces.
• Annotation: The process of adding visual feedback or instructions directly onto a generated design preview.
• Microsoft XP Aesthetic: A specific visual style reminiscent of the Microsoft Windows XP operating system, characterized by its color palette and UI elements.
• Teenage Engineering Aesthetic: A design style inspired by the minimalist, functional, and often retro-futuristic aesthetic of Teenage Engineering products.
• AI Agents: Automated systems that can perform tasks, such as searching for information or generating content.

Personal Website Design

The video begins with a test of Gemini 3.0 Pro's design capabilities by attempting to redesign the presenter's personal website.

• Initial Prompt: The presenter provided a screenshot of their existing, "basic" personal website and asked Gemini 3.0 Pro to design it with "more color, more freshness, and more and just all around more interesting." A specific stylistic request was added: "make it look like an experience in Microsoft XP except it's a personal website." The website domain, gregisenberg.com, was also provided for content retrieval.
• Gemini 3.0 Pro's Process: The AI generated code, including metadata.json and index.html, utilizing Tailwind CSS and a React application entry point. The presenter noted that Gemini 3.0 Pro "shows its work" by displaying the development process.
• Generated Design: The output closely resembled the Microsoft XP aesthetic, featuring elements like a notepad interface and a tagline, "The internet is my playground." It incorporated content from the presenter's website, including links to subscribe to a newsletter, download guides, and view portfolio companies. The design was surprisingly functional on mobile.
• Feedback and Iteration:
  • Icon Improvement: The presenter requested more realistic icons for applications like "About Greg" and "Greg's Letter," noting that the initial icons didn't feel like "real Windows or Mac app icons." Gemini 3.0 Pro updated these icons successfully, though one icon for "Popular Guides" bugged out.
  • Background Annotation: The presenter found the white background "boring" and used the annotation feature to request a background reminiscent of the "classic blue skies and green mountain" of Microsoft XP. This involved drawing on the preview and adding a text comment.
• Results and Rating: The presenter was "pretty damn impressed" with the personal website design, rating it 9 out of 10. They acknowledged that the background change took longer than expected and that providing a reference image for the XP background might have yielded better results.

SAS App Design

The next test involved designing a Software as a Service (SaaS) app, specifically a dashboard for a restaurant analytics and AI application.

• Prompt and Reference Images: The presenter provided a screenshot of a clean, modern SaaS dashboard from Dribbble as a primary reference. They also included a screenshot of a Teenage Engineering product, requesting that the buttons on the SaaS app have a similar tactile, "real-world" feel. The app's purpose was described as "analytics meets AI app for restaurants."
• Input for Detail: The presenter mentioned the importance of providing detailed information, referencing the use of Product Requirements Documents (PRDs). They suggested using tools like ideabrowser.com or templates from resources like Lenny Rachitsky's to generate PRDs, which can then be fed into Gemini 3.0 Pro for more comprehensive design.
• Generated Design: Gemini 3.0 Pro generated an app called "AI analyst chef OS," which appeared to be a functional dashboard for a restaurant. The design incorporated the requested Teenage Engineering-inspired buttons, which the presenter found "super super super good." The app displayed metrics like active tables and average ticket prices, suggesting a high level of detail and functionality.
• Results and Rating: The presenter was highly satisfied with the SaaS app design, rating it 8.5 out of 10. They praised the button design and the overall aesthetic, noting that it was a significant improvement over typical "vibe coded" SaaS apps. The ability to move elements within the generated app was also highlighted.

Mobile App Design

The final test focused on designing a mobile app, inspired by an existing app called "Brain Rot."

• Prompt and Inspiration: The presenter used app store previews of "Brain Rot" as a reference. The goal was to design a mobile app that encourages people to work out, in the "style of Brain Rot," with a mascot and warm colors, specifically for iOS.
• Potential Limitations: The presenter expressed uncertainty about Gemini 3.0 Pro's ability to design mobile apps, fearing it might be limited to web apps.
• Generated Design: Gemini 3.0 Pro generated a React application named "Gains," described as a fitness app. It featured a mascot that reacts to workout habits, a streak counter, goal maintenance mode, and sections for activity and gains calendar. The presenter noted that the mascot design was "tough" and looked "really really good."
• Results and Rating: Despite some minor interaction issues (e.g., inability to log workouts directly in the preview), the presenter was impressed with the mobile app design, rating it 8.3 out of 10 for a single prompt. They acknowledged that providing more data on how to log activities would improve functionality.

Overall Synthesis and Conclusion

The presenter concludes that Gemini 3.0 Pro, when used within Google AI Studio, is a powerful and increasingly capable design tool.

• "Scary Good" Capabilities: The AI demonstrated impressive design generation across personal websites, SaaS apps, and mobile apps, often exceeding expectations with minimal prompting.
• Importance of Prompting and References: The success of the designs is heavily reliant on the quality of prompts and the provision of relevant reference images. The presenter emphasizes that users need to have "taste" and provide good visual examples to achieve beautiful results.
• Shift from "Vibe Coding": Gemini 3.0 Pro is enabling a transition from generating aesthetically uninspired "vibe coded" applications to creating well-designed and visually appealing products.
• Empowerment for Non-Designers: The tool empowers individuals without formal design training to create sophisticated and attractive digital products.
• Recommendation: The presenter strongly encourages viewers to experiment with Gemini 3.0 Pro and similar AI design tools to build their own projects.

The presenter expresses bullishness on Google AI's trajectory, noting their rapid progress from Google Bard to more advanced models and tools like Anti-Gravity (though they found Gemini 3.0 Pro in Google AI Studio to yield better design results). The overall sentiment is that AI is making it possible to achieve high-quality design with just good ideas and effective prompting.