Hugh Williams once led engineering teams at Google. Now he has built a fully functional search engine without typing a single line of code. The project, called Zettair, indexes 1.5 million Wikipedia articles. It delivers autosuggest, query-biased snippets, related searches, trending topics and AI-generated summaries.
Williams relied entirely on Anthropic’s Claude Code. He described his intent in natural language. The AI generated, refined and debugged the entire system. Zero lines of code from him.
But there is a catch. The underlying information-retrieval engine draws from technology Williams helped develop in the early 2000s. That foundation gave Claude Code a strong starting point. Without it, the experiment might have faltered. Williams made this clear in his account.
The former Google leader detailed the effort in a recent Business Insider article. He previously used the same AI assistant to construct an AWS-based system in 48 hours. This time the goal was search. And the results surprised even him.
Williams does not claim AI has replaced engineers. Quite the opposite. “Building with AI feels less like programming and more like coaching,” he said. Experienced engineers, he added, still make the best coaches. The observation lands with weight inside Silicon Valley circles where hype around autonomous coding runs hot.
The term “vibe coding” traces back to Andrej Karpathy. The AI researcher coined it in early 2025. He described a workflow where developers forget the code exists and focus instead on intent, conversation and iteration. Google Cloud documentation explains the practice in detail. Users describe goals in plain language. AI generates code, users test and refine through feedback loops. Pure vibe coding treats the process as exploratory. Responsible versions keep humans in the loop for review and ownership.
Williams operated somewhere in between. His deep expertise in search algorithms shaped every high-level prompt. Claude Code handled implementation. The combination produced production-like features quickly. Yet Williams stressed that novices would struggle to achieve similar outcomes. Context matters. Domain knowledge matters more.
Interest in vibe coding has exploded this year. A June 5 MarkTechPost analysis ranked 15 tools. It placed Atoms at the top for its multi-agent approach that tackles everything from architecture to deployment. Cursor earned praise for its AI-native IDE capabilities. Google’s own AI Studio and Antigravity agent system also appear in developer conversations. But Williams chose Claude Code. The choice reflects Anthropic’s strength in following complex instructions across large projects.
Developers have taken notice. Some report building entire applications in hours. Others warn of hidden costs. Security flaws surface in one of every three AI-generated snippets, according to experiments shared on LinkedIn. Maintenance becomes tricky when no one fully understands the generated codebase. Williams avoided those traps because he understood the retrieval mechanics at a fundamental level.
His Zettair project highlights a broader shift. Traditional search engines once demanded armies of indexers, rankers and relevance engineers. Today a single expert with the right AI partner can stand up a credible alternative. Scale remains limited. Zettair handles Wikipedia, not the open web. Still, the barrier has dropped.
But. The drop applies mainly to those who already possess the expertise. Newcomers may generate code. They rarely generate correct, efficient or secure code at scale. Williams’ takeaway carries a cautionary tone. AI amplifies talent. It does not create it from nothing.
Google itself has leaned into similar ideas. At I/O 2026 the company showcased agentic search features and generative UI that can build interactive elements from natural language. Its official blog post described agents that act on user questions across workflows. The demos included coding examples and fitness trackers built through conversation. Momentum builds across the industry.
Critics point to limitations. Generated code often lacks long-term maintainability. Debugging complex failures still requires human insight. Williams saw both sides. His prior AWS project succeeded because of his systems knowledge. The search engine succeeded for the same reason.
So what does this mean for software teams? Many companies now evaluate AI coding tools with fresh urgency. They weigh speed against quality, autonomy against oversight. Williams offers a practical test. Give the AI a domain you know intimately. Coach it relentlessly. Measure the output against production standards.
The results can impress. Zettair includes trending topics and AI summaries that feel modern. Autosuggest works smoothly. Snippets highlight relevant passages. These capabilities once required teams of specialists. Now they emerge from dialogue with an AI model.
Williams has not stopped at one project. His experiments continue. Each one tests the boundary between human guidance and machine execution. Other former Big Tech engineers pursue parallel paths with different models and use cases. The pattern spreads.
Yet the coaching analogy persists. Great coaches spot weaknesses, adjust strategies and push for excellence. They do not merely issue commands. Williams coached Claude Code through architectural decisions, relevance tuning and feature integration. The AI executed. He evaluated and redirected.
This model may define the next phase of software development. Not replacement. Augmentation grounded in expertise. Novices gain new entry points. Veterans gain velocity. The distribution of capability changes. Who benefits most remains an open question.
Industry observers watch closely. Venture firms pour money into AI coding startups. Enterprise teams pilot tools in sandboxes. Regulators eye security and intellectual property risks. Williams’ Zettair stands as one data point in a rapidly expanding set.
It works. It works because the builder brought decades of search experience to the conversation. Remove that experience and the project likely collapses under its own complexity. That reality tempers the excitement. It also sharpens the opportunity.
Engineers who master both domains — deep technical knowledge and effective AI collaboration — will hold clear advantages. The rest face a steeper climb. Williams has shown one viable path forward. Others will test variations in the months ahead.
The search engine he created may never challenge Google or Bing. Its value lies elsewhere. It demonstrates what becomes possible when experience meets the latest generation of coding agents. The bar has moved. Teams that recognize the shift early stand to gain the most.
WebProNews is an iEntry Publication