AI Website Builder Claims $15K Results for $3 in Tokens

Nick Saraev wants you to know that building premium websites has become absurdly cheap. In a video demonstration, he produces four animated sites with 3D scroll effects in roughly 15 minutes, claiming the whole operation costs under $5 in API tokens—work that would have commanded $5,000 to $10,000 just a few years ago.

The pitch is seductive: combine Claude Code with a design framework from a 16-year-old developer, add AI-generated video from Kling 3.0, and you've got websites that look expensive. Saraev, who built his reputation automating businesses with tools like Make.com, positions this as democratization—taking what used to require specialist knowledge and making it accessible to anyone willing to follow three steps.

The question isn't whether this works. Saraev's demonstration is real enough, and he leaves one of his test sites live for verification. The question is what "works" actually means when you're building websites this way.

The Three-Step Promise

Saraev's process strips web development down to its components. First, feed Claude Code a few bullet points about what you want. Second, point it at Leon Lin's "taste skill"—an open repository that encodes high-end design principles into prompts. Third, generate animated assets using Kling 3.0 through a platform called Higgsfield, then integrate them into the site.

The taste skill does the heavy lifting on design decisions. "This skill just standardizes a few things like spacing, high-end luxury looks and and so on and so forth," Saraev explains. It's essentially a design system compressed into instructions that Claude Code can interpret, producing what he calls "one-shot" websites—sites generated from a single prompt with minimal iteration.

For the animations, he uses Kling 3.0 to create rotating globes, exploding house diagrams, and panning interior shots. The cost breakdown: roughly 7.5 credits per animation on Higgsfield's platform, which he calculates at about 36 cents per video. Generate three variations to pick the best one, add Claude Code's API costs, and he arrives at his $3-5 total.

Free hosting on Netlify completes the package. "Free forever," as Saraev notes, with a global CDN that caches content at edge nodes for faster loading worldwide.

What the Demo Actually Shows

Watch past the headline and you'll see something more complicated than "websites in 10 minutes." Saraev's interior design site requires multiple iterations to get the text readable, the gradients working, and the scroll animations smooth. He tells Claude Code to "make it faster," and the AI extracts video frames as optimized JPEGs, implements preloading, and reduces file sizes—compression from 5.3 megabytes to 252 kilobytes in one case.

"As a former web developer, some of the stuff that this thing does blows my mind," he says. "It would have taken me like 3 days to do what this just did in 30 seconds."

That's the interesting part—not the speed of the initial generation, but the speed of optimization. Traditional web development involves understanding how browsers parse resources, where bottlenecks occur, and which compression algorithms to apply. Claude Code handles this through natural language: "make it faster" becomes a technical implementation without requiring the user to understand the implementation.

But optimization only matters if the initial output is worth optimizing. Saraev's sites have a particular aesthetic—smooth, minimal, heavy on whitespace and 3D effects. They look like portfolio pieces for a design agency, which makes sense given the taste skill's focus on "high-end luxury looks." Whether they work for actual businesses depends entirely on what those businesses need to communicate.

The Skills That Haven't Changed

Saraev dismisses the complexity that other creators add to this process: "Everybody in their mom's just like, 'Oh my god, these website design frameworks are the reason why we're capable of doing this these sites so well,' and then they show you like 35,000 steps, and it's a massive pain in the butt."

He's right that the technical barrier has dropped. You don't need to understand CSS animation timelines or video codec optimization. But you still need to understand what makes a website effective, which animations enhance rather than distract, and whether the aesthetic matches the message.

The taste skill provides one answer to these questions—a specific point of view about contemporary web design. It works if your goal is to look like every other site using the taste skill. It fails if you need to stand out from that aesthetic or serve audiences who want something different.

Saraev notes that he hasn't mobile-optimized his demonstration site, suggesting you can just tell Claude Code to "mobile optimize the site three or four times" and it will handle it. That's probably true in a technical sense—the layouts will adapt, the touch targets will enlarge, the fonts will scale. Whether the experience actually works on a phone depends on choices the AI can't make: which information matters most on a small screen, whether those scroll animations make sense with touch gestures, whether the page weight is reasonable on cellular connections.

These aren't technical problems. They're judgment problems dressed up as technical problems.

The Economics Worth Examining

Saraev's $15,000 comparison point comes from his experience selling websites. Whether that's a fair baseline depends on what those $15,000 sites included—discovery, strategy, custom design, content creation, testing, training, ongoing support. If you're comparing the cost of generating HTML and CSS to the cost of a full-service web project, you're measuring different things.

The $3-5 token cost is accurate for what Saraev demonstrates: API calls to generate code and animations. It doesn't include the time spent refining prompts, the iterations needed to get animations that work, or the knowledge required to recognize when something needs fixing. His video runs nearly 14 minutes to build one site, and that's with experience knowing which prompts work and which tools to chain together.

For someone building their first AI-generated site, the learning curve isn't the tools—it's understanding enough about web design to evaluate what the tools produce. That knowledge used to come bundled with the technical skills required to build sites. Now it's separated, which means people can produce things that look professional without understanding whether they are professional.

That gap between appearance and function is where most of these projects will fail. Not because the technology doesn't work, but because knowing how to use the technology isn't the same as knowing what to build.

What This Means for Web Development

Saraev is documenting a real shift. The technical implementation of websites—the code, the optimization, the deployment—has become a commodity service that AI handles competently. This was inevitable. Web development always involved too much boilerplate, too many decisions with obvious right answers, too much work that could be automated once someone wrote the automation.

What remains is the work that can't be automated: understanding what a client actually needs versus what they say they need, recognizing when a beautiful animation makes the message clearer or just makes it slower, knowing which conventions to follow and which to break.

These were always the valuable parts of web development. The difference is they used to be bundled with technical execution in a way that made them hard to separate. Now they're exposed, and it's not yet clear how the market will value them.

Saraev offers one answer: teach people to use these tools through free courses and demonstrations, then monetize through affiliate deals and sponsored content. That works for him because he's already established credibility building seven-figure businesses. For everyone else watching these videos, the calculation is harder.

The tools work as advertised. Whether they solve your problem is a different question entirely.

—Bob Reynolds, Senior Technology Correspondent