At SaaStr, we're enthusiastic about "vibe coding," a process where we spend significant time daily building production applications with AI-assisted tools. SaaStr.ai itself was partially developed this way. However, a common observation has emerged: if you've seen one AI-generated website, you've largely seen them all.
This distinct aesthetic is easily recognizable: ubiquitous Lucide React icons, sans-serif fonts like Inter, a specific purple-blue hue, rounded corners, identical card shadows, and consistent button styles. It's as if every AI-powered design tool attended the same "Whatever Claude Thinks Looks Good University."
Google's Gemini 3 is now changing this landscape. Developers can now "vibe code" an application without a dedicated designer, resulting in a truly designed look that stands apart from other AI-generated sites. This represents a significant leap forward in AI-assisted web development.
Nice job @Replit on new Design mode
This took < 30 seconds pic.twitter.com/WfMNd5HeS3
— Jason ⚡️Saastr.Ai⚡️ Lemkin (@jasonlk) November 19, 2025
Design Lagged Function in Early AI-Coded Apps
The reason for this design homogeneity is clear: most prosumer "vibe coding" tools—including Replit, Cursor, Bolt, Lovable, and V0—have historically relied on Claude as their underlying large language model (LLM). Claude, it turns out, has strong opinions about design, defaulting to a specific set of aesthetics:
- Lucide React icons
- Tailwind utility classes
- Specific color palettes (often purple, blue, and teal gradients)
- Common fonts (Inter, system fonts)
- Standard component patterns (cards with shadows, rounded corners, gradient backgrounds)
When millions of users leverage different tools powered by the same LLM with identical design preferences, the inevitable outcome is a generation of websites that look remarkably similar. Once you've encountered a few, you can instantly identify an AI-generated site.
This isn't a flaw in Claude itself; it excels at generating clean, functional, and modern interfaces. However, at scale, "modern-looking" quickly devolved into "identical-looking." The ability to instantly spot a Claude-generated site from a distance became a noticeable problem for brand differentiation.
Gemini 3 and Design-First Tools Usher in a New Era
The landscape has shifted dramatically with Gemini 3 and the advent of design-first tools. For instance, a mockup for the SaaStr Annual 2026 landing page was created in under 30 seconds using Replit's new design mode, powered by Gemini 3. Crucially, this mockup doesn't resemble the typical AI-generated site.
Here's what makes the difference:
- Starting with actual design: Replit's new design mode prioritizes a visual design environment. Users define colors, typography, layout, and spacing first. Gemini 3 then generates code that aligns with the user's specific design system, rather than its own default preferences.
- Gemini 3's distinct aesthetics: Unlike Claude, Gemini 3 possesses different default design opinions. This distinction, stemming from a different base model, leads to genuinely differentiated outputs, breaking the previous design monoculture.
- Flexible design systems: While Tailwind remains an option, the AI now generates code based on user-defined design systems, moving beyond the default utility classes that other AI tools often fall back on.
- Real brand differentiation at AI speed: This is the key breakthrough. Development remains incredibly fast—the 30-second timeline is not an exaggeration—but now the output reflects a unique brand identity, not just another generic AI-generated template.
Why This Evolution Matters for B2B Applications
Insights gained from building SaaStr.ai with AI and deploying numerous "vibe coded" apps highlight critical implications:
- First impressions are paramount: In a crowded digital landscape where many sites look alike, those that stand out immediately gain an advantage. If a potential customer's first interaction with your product is a homepage indistinguishable from a competitor's (due to shared AI tools and prompts), you've already lost a crucial opportunity for differentiation.
- The "Claude look" became a tell: This distinct aesthetic often signaled that a site was "built by non-technical founders with AI." While there's nothing inherently wrong with non-technical founders, a site that overtly screams "AI-generated" can send unintended signals to potential customers, investors, and future hires.
- Vibe coding needed this evolution: The true promise of AI-assisted development isn't about everyone building the same thing faster. It's about empowering everyone to build *their unique vision* faster. This necessitates breaking free from the default design opinions of earlier AI models.
The Practical Reality of Design-First AI Development
Tools like Replit, powered by Gemini 3 and its new design mode, are fundamentally altering the speed-quality tradeoff in web development. Historically, developers had to choose between rapid, generic-looking sites (using early AI tools) or custom-designed, slower development (with traditional designers and developers). Now, both are achievable. The SaaStr Annual mockup, created in 30 seconds, genuinely appears to be the product of a dedicated design team.
This design-first approach also significantly improves iteration. By starting with design and then generating code, and subsequently iterating on the design, the AI gains superior context for the desired outcome. It moves beyond simply defaulting to Lucide icons and purple gradients.
This paradigm shift makes Figma integration particularly exciting. Imagine designing in Figma (the preferred environment for many designers), exporting to AI-generated code that perfectly preserves your design system, iterating on the code, and then pushing design changes back to Figma. This seamless workflow is becoming a reality, driven by advanced models like Gemini 3 that comprehend both design specifications and code.
Breathtaking Pace of Change in B2B Vibe Coding
The speed of innovation in B2B "vibe coding" is remarkable. SaaStr embarked on its "vibe coding" journey less than 180 days ago, when agents were limited and building B2B apps was genuinely challenging. Since then, immense improvements have been made, and now, design capabilities have caught up.
The core promise of AI-assisted development was never about everyone building the same thing faster. It was always about empowering individuals and teams to build exactly what they envision, at an accelerated pace.
For the past year, most prosumer "vibe coding" tools, relying on Claude, led to a pervasive design uniformity: identical icons, fonts, colors, and component patterns.
Gemini 3, coupled with design-first tools like Replit's new design mode, is finally breaking this pattern. Your website no longer has to look like it was merely "vibe coded"; it can now appear genuinely designed, all while maintaining the speed of AI-driven development. This is the significant unlock for the future of web creation.
P.S. If you're attending SaaStr AI London on December 1-2, we'll be exploring practical AI implementation in depth, focusing on tactical deployment rather than just theory or demos. Get your tickets here.
For those building AI-native B2B/SaaS and seeking funding, explore SaaStr.ai. We've onboarded over 300 top VCs and facilitate real-time matching between founders and capital.
