Pencils down: We share our vibe-coded websites

Key takeaways

AI tools like Gemini, Codex, and Claude are powerful for solo or small-scale projects, allowing users to quickly prototype personalized websites and systems.
Despite their potential, current AI systems have significant limitations, including inconsistent outputs, security challenges, and platform restrictions that hinder scaling to broader audiences.
Creating polished, user-friendly software still requires substantial technical skills and ongoing effort, beyond what AI can currently provide alone.
Experimentation with vibe coding reveals both the exciting possibilities and the current frustrations, emphasizing the need for better understanding, patience, and preparation when working with these tools.

Summary

Vibe Coding Challenges and Experiences

The hosts reflect on a recent month-long experiment where they each built personal websites or projects using AI-driven vibe-coding tools like Gemini, Codex, and Claude. The discussion highlights the varying challenges faced, such as coding frustrations, platform limitations, and security issues. Hayden built a family photo-sharing site with unique security features, Jake created a vegetable pickup management system, and the host developed a searchable archive for their newsletter content. They discuss how these tools are powerful for personal projects but still complex and difficult to scale for broader consumption.

Advancements and Barriers in AI-Assisted Coding

The conversation emphasizes the promise of AI tools like Codex, Gemini, and Claude in enabling individuals to create customized solutions without deep technical expertise. However, they point out significant hurdles including inconsistent AI behavior, platform restrictions, security concerns, and the difficulty in transforming prototypes into user-friendly, scalable applications. The hosts agree that while AI can help generate prototypes or small tools, the leap to polished, publicly available products requires skills and resources beyond current AI capabilities.

Reflections on AI Tools and Software Development

Throughout their projects, the hosts note a shift in attitude towards AI chatbots—from polite and cautious to terse and frustrated—highlighting their limitations in understanding context and complex commands. They discuss the massive amount of work involved in building, debugging, and maintaining these small systems, and the challenge of making them usable and meaningful to others. The overall sentiment reflects excitement for AI-assisted making, coupled with awareness of the immense work required to scale these tools into mainstream software products.

Show notes

We've spent the last month vibe coding our own corners of the internet. The Verge's Jake Kastrenakes and Hayden Field join David to talk about their use of Gemini, Codex, and Claude Code to build 90s-themed personal websites, vegetable-sharing systems, and photo boards. It wasn't always easy, and we didn't get everything done, but with a little AI and a lot of prompting, we made it work.