Thoughts on AI
After finally seeing some pretty cool things getting vibe-coded, I thought it would be useful to take a deeper dive into technology myself. I wanted to look beyond the hype, better understand how it works, and form a more informed perspective on its potential and implications. My first idea was to build a small step sequencer using Base44. It's built around a programmable synth, with controls for basic parameters like waveform, attack, and cutoff. You work on a grid of quarter notes, clicking to add notes and dragging their edges to change their length. Below, you can see me doing a very poor imitation of Agraph's music. Since making this video, I've extended the grid by an octave in each direction and added the ability to create additional sound layers.
It was cool, and sure, it's pretty impressive that you can more or less speak these things into existence. But I don't get the same satisfaction from this process as I do when I build something more from scratch. In some ways, the current version of AI feels less suited to people like me, or to creatives in general, than to people focused primarily on reducing the time and cost of labor. It can feel as though you need a well-defined workflow before you can make the most of it. Since so much of what I'm doing right now is just exploring, the prospect of another subscription isn't particularly enticing.
There's an underlying mismatch between prompting and the creative process. Prompting puts a ton of weight on knowing how to describe what you want and then waiting for an output. But the crux of creative work is that you don't always know what you want until you start putting things on paper. Carving something out piece by piece is itself a way of discovering and describing it, so relying on prompts can feel like sidestepping the process. Of course, you can end up with a similar result. But at the end of all that prompting, I sometimes find myself thinking, “So what?” Making something becomes a lot less interesting when I have little connection to how it was made, no matter how impressive the outcome is.
The idea for the sequencer actually started a couple of years ago, when I was thinking about creating little music cards you could save in a format like JSON and share with other people. You could load a card into the sequencer and play it, much like inserting a tape into a cassette player. At the time, the idea was partly driven by NFT mania. NFTs seemed like the perfect vehicle for an art project built around creating and trading “scarce” cards. Of course, that trend fizzled out (badly), and so did my motivation to learn the skills I needed to finish the project. Looking back, I can see that part of my interest came from wanting to capitalize on a trend. That's not so different, I think, from how a lot of people are approaching AI today.
Finding a Use for AI
After trying a few ideas in Base44, I subscribed to both Cursor and Codex. That's when I began to see some personal value in these tools. With fewer barriers to experimenting, I finally tackled some overdue updates and bugs on my personal site. I built a new photo portfolio from scratch and added dark mode to my blog. It wasn't that I couldn't have made those changes myself. I just didn't want to spend the time learning the particular architecture of my Gatsby template to implement something like dark mode. My blog and portfolio are personal projects, after all. They're there for fun and as a means of self-expression.
What I like most is that I can keep control of my site, including things like self-hosting, while working in a way that almost feels like using a site builder. It's a little like working with Wix or Framer, only without being locked into either platform. I'm not using a drag-and-drop builder, but it's pretty dang close, and it costs a fraction as much. So why do I keep bouncing between being impressed by what these tools can do and feeling dissatisfied when I use them?
The Cost of Convenience
For one thing, there's a kind of technical debt that comes with this approach. AI removes much of the friction from turning an idea into something that works. But when I finally took a peek under the hood, I was struck by how little I understood about what I'd made. Before, I'd build something brick by brick, gradually learning how each component fit together. Now I can lay down entire buildings at once without necessarily knowing how they stand up. What I gain in speed, I risk losing in understanding. Eventually, I either have to wade through the sludge of generated code or keep paying for a tool to help me maintain it.
This has made me think more about what I enjoy about working on the web in the first place. When I was younger, I built websites purely for fun. I used platforms like Homestead, one of the early drag-and-drop site builders, to make rudimentary sites about whatever interested me. Before long, that curiosity led me to learn some basic coding skills so I could customize my MySpace profile or add a theme to my Invision board. What drew me in was the sense that the internet offered an infrastructure for self-expression and connection. As AI changes what kinds of skills seem valuable, I'm sure many of us are reassessing how we spend our time. For me, it's made something clear: some of the tasks AI handles well aren't the parts of creating that I care most about.
The internet is starting to feel like a hollow facsimile of what it once was. Part of its appeal was that creatives and tinkerers could own a little corner of it. Before so much of our activity moved into a handful of walled gardens, you could build something that felt genuinely yours. Sharing mattered, too. Social media offered a place for creatives to put their work in front of other people. Now that creative work can be absorbed into training datasets and imitated so quickly, it's fair to wonder what sharing online will look like in the future. And when you start asking who benefits most from these technologies, the ambivalence many people feel toward AI makes more sense.
That ambivalence has kept me from going too deep with these tools beyond mundane tasks like proofreading, troubleshooting code, or using AI as a glorified Google Search (and quite frankly due to the questionable benefits, I still feel that way). I'm also put off by parts of the surrounding culture and by the questionable ethics and risks that come with the technology. Even when AI handles a task reasonably well, I can get frustrated wading through lengthy responses or seeing how easily an answer shifts depending on how I phrase a question. It can sound like a living thing with an opinion, even though its responses aren't expressions of personal experience or conviction. Sometimes doing the work myself is not only less frustrating, it also brings me closer to what I'm making.
What Makes Creating Worthwhile?
After the initial wow factor wore off, the fun part of building the sequencer was simply playing with it. And since I already own several fully fledged digital audio workstations, it's hard to imagine choosing my vibe-coded app over them. As impressive as these creations can be, I find myself less interested in saying, “Wow, look what the computer can do,” than in saying, “Wow, look what this person can do with it.” Human ingenuity is impressive precisely because people find unexpected ways to work within, and push beyond, their limitations. Given the enormous resources poured into training AI systems, their capabilities can sometimes feel less surprising. What else would you expect? I'd much rather watch a person exceed expectations or accomplish something they weren't necessarily equipped to do.
My feelings are still evolving. These tools are getting freakishly good, and the conversation is moving at light speed, sometimes so fast that I can barely find my footing, let alone settle on what I think. But one thing keeps coming back to me: having an idea is only a tiny part of creating something. The real reward is seeing it through, including the effort and frustration along the way. Passion, after all, means to endure. For now, I think I'll continue using AI sparingly.