We love vibe coding at SaaStr. We spend 1.5-2 hours most days building production apps with AI-assisted tools. SaaStr.ai itself was partly built this way. But let’s be honest about something we’ve all noticed: if you’ve seen one AI-generated site, you’ve kind of seen them all.
You know the look. Those same Lucide React icons. Inter or similar sans-serif fonts. That particular shade of purple-blue. Rounded corners everywhere. The same card shadows. The same button styles. It’s like everyone went to the same design school, and that school was “Whatever Claude Thinks Looks Good University.”
Gemini 3 has changed that. Now you can vibe code an app without a designer that really looks … designed. Not like every other vibed site. That’s a big deal.
Design Has Lagged Function in Vibe Coded Apps
Here’s what happened: Most prosumer vibe coding tools – Replit, Cursor, Bolt, Lovable, V0, you name it – use Claude under the hood. And Claude has opinions about design. Strong ones.
Claude defaults to:
- Lucide React icons (always)
- Tailwind utility classes (always)
- Specific color palettes (purple, blue, teal gradients)
- Specific fonts (Inter, system fonts)
- Specific component patterns (cards with shadows, rounded corners, gradient backgrounds)
So when millions of people are using different tools that all use the same LLM with the same design preferences, we ended up with a generation of websites that all look… exactly the same. Once you’ve seen one or two, you can just see it instantly the next time you see a vibe’d site.
It’s not Claude’s fault. It’s actually doing exactly what it should do – generating clean, functional, modern-looking interfaces. But “modern-looking” became “identical-looking” at scale.
You can literally spot a Claude-generated site from across the room. That’s a problem.
What Changed With Gemini 3 + Design-First Tools
I built that SaaStr Annual 2026 landing page mockup in under 30 seconds with Replit’s new design mode powered by Gemini 3. Not 30 minutes. Not 3 hours. 30 seconds. And it doesn’t look like every other AI-generated site out there.
Here’s what’s different:
- Starting with actual design, not just having the AI guess. When you use Replit’s new design mode, you’re working in a visual design environment FIRST. You’re defining colors, typography, layout, spacing. Then Gemini 3 generates code that matches YOUR design system, not its default preferences.
- Gemini 3 has different default aesthetics than Claude. This matters more than people realize. Different base model = different design opinions = actually differentiated outputs. It’s not that Gemini is “better” at design – it’s that it’s different, which breaks the monoculture.
- Design systems that aren’t Tailwind-defaults-by-default. You can still use Tailwind (I do constantly). But now the AI is generating code based on your defined design system, not just reaching for the same utility classes every other AI tool uses.
- Real brand differentiation at AI speed. This is the unlock. You can still build at AI speed – that 30 second timeline isn’t an exaggeration. But now you can build something that looks like YOUR brand, not like “Claude-generated SaaS site #47,392.”
Why This Matters for B2B Apps
Here’s what I’ve learned building SaaStr.ai with AI and shipping 10+ vibe coded apps into production:
First impressions matter even more than they used to. When every site looks the same, the ones that DON’T stand out immediately. Your homepage is often the first interaction a potential customer has with your product. If it looks exactly like your competitor’s homepage (because you both used Claude-powered tools with similar prompts), you’ve already lost differentiation.
The “Claude look” became a tell for “this was built by non-technical founders with AI.” And look, I have nothing against non-technical founders – I’m an investor in dozens of them. But when your site SCREAMS “AI-generated,” you’re signaling something to potential customers, investors, and hires. Not always something good.
Vibe coding is still the future, but it needed this evolution. The promise of AI-assisted development isn’t that everyone builds the same site faster. It’s that everyone can build THEIR site faster. That requires breaking free from Claude’s default design opinions.
Screenshot
The Practical Reality
Tools like Replit with Gemini 3 + design mode will change the speed/quality tradeoff. You used to have to choose: fast and generic-looking (Claude-powered tools), or custom-designed but slow (actual designers and developers). Now you can have both. That’s not hyperbole – that SaaStr Annual mockup took 30 seconds and looks like it came from a design team.
The design-first approach works better for iteration too. When you start with design, then generate code, then iterate on the design again, the AI has better context for what you’re trying to achieve. It’s not just defaulting to Lucide icons and purple gradients because that’s what it always does.
This is where Figma integration gets really interesting. Design in Figma (where designers actually work), export to AI-generated code that preserves your design system perfectly, iterate in code, push design changes back to Figma. That workflow is becoming real, and it’s powered by models like Gemini 3 that can understand both design specs and code.
The Pace of Change and Improvement in B2B Vibe Coding is Breathtaking
We started on our vibe coding journey at SaaStr less than 180 days ago. The agents were limited, vibe’ing B2B apps was hard. Hard. So much has gotten so, so much better since then. And now, design has too.
The promise of AI-assisted development was never “everyone builds the same thing faster.” It was “everyone can build what they actually want, faster.”
For the last year or so, prosumer vibe coding tools all used Claude, so everything looked the same. Same icons. Same fonts. Same colors. Same component patterns.
Gemini 3 + design-first tools like Replit’s design mode are finally breaking that pattern.
Your site doesn’t have to look like it was vibe coded anymore. It can look like it was actually designed. And you can still build it just as quickly.
That’s the unlock.
P.S. – If you’re coming to SaaStr AI London on December 1-2, we’ll be diving deep into practical AI implementation like this. Not theory. Not demos. Actual tactical deployment. Get tickets here.
And if you’re building AI-native B2B/SaaS and looking for funding, check out SaaStr.ai – we’ve onboarded 300+ top VCs and are matching founders with capital in real-time.
