My First Week AI-Coding with Claude: Scuba Diving in a Kiddie Pool

Reading time: 5 minutes

Last week, I strapped on my metaphorical scuba gear and dove into AI-assisted coding with Claude. What I discovered was simultaneously thrilling and humbling—like finding yourself fully equipped for deep-sea exploration, but floating in three feet of water.

The "Hello World" Rush

I started simple. "Claude, help me create a basic website." Within minutes, I had clean HTML, polished CSS, and something actually worth showing off. The speed was intoxicating—this wasn't just functional, it was beautiful.

Riding this wave of success, I immediately started personalizing. I added my partner's name to the site, then my kiddo's, then my best friends'. Each addition took seconds, and I found myself frantically screenshotting and texting everyone: "Look! I'm coding! I made this!" The dopamine hit of rapid progress was real, and I was completely hooked.

Static Site Paradise

What followed was a golden period of static website creation. Claude proved to be an exceptional partner for this work. Portfolio sites, experimental layouts, responsive designs—everything emerged polished and professional. The CSS was organized, the HTML was semantic, and it all just worked.

I started feeling like a real developer, cranking out site after site with growing confidence. Claude has an intuitive grasp of modern design principles and user experience that would have taken me hours to research and implement on my own.

The Beautiful Mirage

Then came the inevitable next step: adding real functionality. Authentication seemed like a natural progression. "Claude, help me add a login system to my site."

Claude delivered exactly what I asked for, and it was gorgeous. Smooth animations, perfect color schemes, professional typography—it looked like it belonged in a premium SaaS application.

There was just one small issue: you could log in with test@test.com and "test" as the password. Didn't have an account? No problem—the code would let you in anyway. Already signed up? The system had no way of knowing or caring. It was a stunning hotel lobby with no actual rooms behind the reception desk.

The "Quick to Please" Challenge

This experience revealed something important about Claude's approach to coding assistance. While it excels at creating beautiful, functional interfaces, it doesn't automatically architect the infrastructure beneath them. Claude is like an incredibly skilled craftsperson who will build exactly what you ask for, but won't necessarily question whether what you're asking for makes structural sense.

When I asked for a login system, Claude gave me a login interface. When I needed authentication, it provided authentication styling. The gap between what I thought I was requesting and what Claude understood I was requesting became crystal clear.

This isn't necessarily a limitation—Claude can absolutely provide excellent architectural guidance when you know to ask for it. The key is starting with the right questions: "What are the best practices for user authentication?" or "How should I structure a secure login system?" When prompted for architectural thinking, Claude delivers thoughtful, comprehensive solutions.

But left to interpret a simple request like "add a login system," it optimizes for speed and immediate functionality rather than long-term architecture.

The Learning Curve

What I discovered is that Claude assumes you know what you want and delivers exactly that. It won't proactively suggest that you need user sessions, database integration, or proper password hashing. It won't remind you about CORS policies or recommend rate limiting.

However, ask it to help you implement JWT authentication with proper password hashing, and it'll deliver clean, secure code. Request guidance on setting up a robust database schema, and it'll provide thoughtful recommendations. The expertise is absolutely there—it just needs to be specifically accessed.

This taught me that successful AI-assisted coding requires a different kind of partnership than I initially expected. Instead of asking "make this work," the magic happens when you ask "what's the best way to make this work?"

Moving Beyond the Shallows

This realization sent me exploring tools that might provide more architectural scaffolding by default. Enter Bolt, and then Lovable—platforms that promised to bridge the gap between rapid prototyping and production-ready applications.

But that's a story for the next post, where I'll share what I learned about the evolving landscape of AI development tools and how each approaches the balance between speed and depth.

The Real Takeaway

My first week with Claude taught me that AI coding assistance amplifies your existing knowledge and intentions. If you understand software architecture and ask the right questions, Claude becomes an incredibly productive partner. If you're learning architecture, you need to explicitly request that guidance—otherwise, Claude will happily help you build beautiful-looking solutions that work perfectly on the surface.

The scuba diving metaphor holds: I had all the right equipment for deep exploration, but I started in shallow waters because I hadn't yet learned to navigate to the depths I actually wanted to reach.

And honestly? That's perfectly fine. Sometimes you need to float around in the shallow end for a while, getting comfortable with your gear and learning how to use it effectively, before you're ready for deeper waters.

The important thing is recognizing when you're ready to dive deeper—and understanding that the tool that got you started has more depth to offer when you know how to access it.

Coming up next: my experiments with Bolt and Lovable, and what I learned about the current state of AI-powered development platforms.