First Run with Vibe Codin

At its best (which was roughly 70-80% of the time), the experience was genuinely transformative. I could focus entirely on what I wanted to create rather than getting bogged down in the minutiae of Next.js configuration, TypeScript type definitions, or CSS quirks. The AI became an extension of my thinking, translating high-level intentions into working code with remarkable accuracy.

This site you're browsing now — from the animated navigation to the photo galleries and CMS backend — was almost entirely (99%) written by Cursor AI with Claude 3.7 Sonnet. I provided direction, reviewed the code, and handled debugging when necessary, but the implementation details came from the AI.

One of the most exciting implications is how this might reshape the engineer's role to be more product-oriented. When freed from much of the coding grunt work, we can focus more on the "why" and "what" instead of just the "how." This potentially bridges the gap between technical implementation and user needs in ways that weren't possible when coding consumed most of our bandwidth.

What I didn't anticipate was the different kind of technical expertise required. Rather than memorizing syntax or framework specifics, I found myself developing skills in:

The reality is that working effectively with AI still requires significant technical background. In many ways, it felt like AI was helping me "relearn" how to ride a bike — it made things easier, but only because I already knew what a working bicycle should look and feel like.

The lows were genuinely deflating. There were moments when the model would suddenly produce code that was nonsensical or technically incoherent — almost as if it had forgotten everything about our previous interactions or basic programming principles.

During these episodes, I'd find myself debating whether to restart the entire conversation, reboot Cursor, or try to salvage things through careful prompting. It's a strange feeling to watch a system that was writing complex authentication logic minutes ago suddenly struggle to understand basic React component structure.

The central limitation right now appears to be context management. Both Cursor and Claude improved dramatically during my project (with updates in February 2025), but they still struggle with maintaining attention over complex, multi-faceted projects.

Until AI companies solve the general context and attention problems, this technology isn't quite ready for broad adoption. The chaos the model occasionally injects would likely overwhelm someone without a solid foundation in software development.

The public discourse around AI tools often presents them as "coding for non-coders," but my experience suggests we're not there yet. This isn't about protecting professional turf — it's a practical observation that the current tools require you to:

Without these abilities, project entropy would likely increase to a point where the application would become unmaintainable.

We're witnessing a fundamental shift in how software is built, but we haven't yet reached what I'd call the "iPhone moment" — that magical intersection where technology and utility converge to create something transformative for everyone, particularly non-technical users.

For developers, the revolution is already well underway. Our work is changing fundamentally, and productivity gains are substantial. But for the general public—the non-technical creators, small business owners, educators, and others who might want to build software solutions—we're still missing the breakthrough that would make this technology truly accessible without requiring significant technical knowledge.

As context handling improves and models become more reliable, I expect this approach to gradually become accessible to broader audiences. But for now, the most effective practitioners will be those who blend AI capabilities with solid technical foundations — product engineers who can navigate both the creative and technical realms with confidence.

The journey continues, and I'm excited to see where it leads.