Background

In recent years, I have found myself in short supply of 'free' time. My wife and I moved into her childhood home ~3 years ago, and ever since it has been an endless string of home renovation projects. As such, I have not had the time I would have liked to tinker with some personal AI projects outside of work.

Prior to the very website you are reading this post on, I had not touched Claude Code. In fact, I felt fairly out of touch with AI tooling in general. I've used Github Copilot for work, but rarely for anything beyond debugging or minor tweaking. There are a few reasons for this, which I'll outline and expand on below

Knowing the Codebase

I've always held the belief that knowing how all the pieces in a system is crucial to effectively working on that system. Moreover, knowing HOW those pieces work in detail can make you an invaluable resource. One of the best ways to accomplish this is by either

 - Writing the code yourself
 - Reading through the code you didn't write

Using AI tooling makes it all too easy to bypass both of these points. I have stood this entire website up via prompting. I have done <1% of 'manual' coding on it (at least, at the time of this writing). The result is a very vague and incomplete idea of how this code is written and how it works. What I do know is largely predicated on my prior knowledge.

Forfeit of Learning

I have no experience with Svelte. The extend of my knowledge is a 'sample project' I spun up years ago that consisted of running the default create command and then abandoning the project. All of the code in this website has been the result of Claude's efforts. I have a functioning, deployed app with a fraction of the understanding originally required. It almost feels like going to a foreign country and relying entirely on a translation app to interact. Sure, it gets the job done, but the cost is steep. Not only are you not learning/retaining this information, but you are also giving up an opportunity to challenge and exercise your brain.