Four Malaysian developers demonstrated the power of AI-assisted coding at Google’s Gemini 3 Hackathon in Singapore, building a fully functional feng shui app in just seven hours using primarily AI-generated code. The team—comprising accountant Chan Wei Khjan, AI lecturer Chan Ler-Kuan, IT professional Loh Wah Kiang, and CTO Lee How Sien (Benny)—embarked on what’s known as “vibe coding,” where developers rely heavily on AI models to generate and debug code.
The hackathon, organized by Google DeepMind and Singapore’s AI builder collective 65labs, challenged 189 participants to build working demos by 5:30 p.m., with projects competing for a 100,000-credit Gemini API prize pool. The team’s app, dubbed “Feng Shui Banana,” used the Gemini Live API to analyze users’ outfits and workspaces through real-time camera feeds, providing feng shui assessments based on traditional Chinese practices.
The development process showcased both the capabilities and limitations of AI coding assistants. Wei Khjan led the prompting effort, primarily using Claude to generate the initial codebase and workflow. When bugs emerged—as they inevitably did—the team simply copied error messages back into the AI for fixes. A critical insight emerged around 12:20 p.m. when Wei Khjan discovered that changing his prompting style from issuing commands to asking the AI to “discuss it with me” transformed the model into a more effective collaborator.
The app successfully identified clothing colors, analyzed feng shui compatibility based on birth timings, and provided real-time recommendations. The team used Google’s AI Studio to design logos, Gemini to generate video storyboards and scripts, and various AI models to create voice output that mimicked a Chinese feng shui master’s speaking style.
Despite the “vibe coding” label suggesting effortless development, the reality proved more demanding. The team worked through lunch, constantly refined AI outputs, and manually corrected feng shui logic when the AI’s understanding fell short. Ler-Kuan had to manually adjust the underlying dictionary and mappings where color analysis intersected with traditional feng shui principles.
While the team didn’t win prizes, they successfully demonstrated how AI development tools can dramatically accelerate prototyping. By 5:30 p.m., they had created a working app with real-time camera analysis, workspace evaluation, audio output, an animated landing page, and a professional demo video—all built primarily through AI assistance in a single workday.
Key Quotes
For people who don’t know how to read code, it’s helpful to have people who do
Wei Khjan explained the team composition strategy, highlighting that while AI can generate code, human developers who understand programming remain essential for reviewing, refining, and debugging AI-generated output—a key limitation of current vibe coding approaches.
Instead of issuing commands, he asked the AI to ‘discuss it with me.’ The shift changed how the model reasoned, and it worked more like a collaborator
This insight from Wei Khjan reveals a critical discovery about effective AI interaction: conversational prompting can produce better results than directive commands, suggesting that prompt engineering skills are becoming as important as traditional coding knowledge.
Sometimes, the best experiences come from saying ‘yes’ without overthinking. Innovation starts with curiosity and a little bit of spontaneity
Ler-Kuan reflected on the hackathon experience, capturing how AI tools are lowering barriers to experimentation and enabling rapid prototyping that encourages developers to test ideas quickly rather than spending extensive time on planning.
Seven hours of vibe coding turned out to be anything but effortless
The reporter’s observation challenges the notion that AI makes coding trivial, revealing that while AI accelerates development, creating functional applications still requires sustained effort, problem-solving, and human judgment throughout the process.
Our Take
This hackathon story perfectly encapsulates the current reality of AI-assisted development: transformative but not magical. The team’s experience demonstrates that we’re entering an era where rapid prototyping becomes accessible to broader audiences, but the “vibe coding” label is somewhat misleading—this was intensive, focused work requiring constant human intervention.
What’s particularly significant is the hybrid workflow that emerged: AI generating initial code, humans debugging and refining, AI fixing errors, humans providing domain expertise. This pattern likely represents the future of software development for the next several years. The discovery about conversational prompting suggests we’re still learning optimal human-AI collaboration patterns.
The fact that an accountant could contribute meaningfully to building a functional app in seven hours signals a genuine shift in who can participate in software creation, though the continued need for developers who “know how to read code” shows we’re far from AI replacing programmers entirely.
Why This Matters
This hackathon experience provides a real-world case study of AI-assisted development’s current state and limitations. The story matters because it demonstrates how generative AI tools like Claude and Gemini are fundamentally changing software development, enabling non-traditional developers to build functional applications rapidly. The team included an accountant and lecturer, showing how AI is democratizing coding beyond traditional software engineers.
However, the experience also reveals critical limitations: human expertise remained essential for domain knowledge (feng shui concepts), debugging complex logic, and making creative decisions. The team’s discovery that prompting style significantly impacts AI performance highlights the emerging skill of “prompt engineering” as crucial for effective AI collaboration.
For the broader tech industry, this represents the future of rapid prototyping and MVP development. Companies can potentially reduce development timelines from weeks to hours, though the need for human oversight, domain expertise, and quality control remains paramount. The story also illustrates how AI coding assistants are tools that augment rather than replace developers, requiring new collaborative workflows between humans and AI systems.
Related Stories
- Andrej Karpathy Reflects on ‘Vibe Coding’ Revolution in AI
- Google Founders’ $511B Fortune Soars on AI Breakthrough Success
- PwC Hosts ‘Prompting Parties’ to Train Employees on AI Usage
- Harvard Students Behind Facial Recognition AI Eye Construction Industry
Source: https://www.businessinsider.com/vibe-coding-team-embed-google-gemini-hackathon-singapore-2026-2