Building with AI
These past few months have been a whirlwind of new experiences.
I have been building a lot with AI, specifically Claude Code, for the past few months. The biggest project I’ve been working on is the Ouroboros app. I’ve also done quite a deep dive on this site, using Claude Design to come up with this current iteration, as well as using Claude Code to roll my own CMS for this site. In addition to that, there’s another little service site I’m building that’s a bit under wraps until it launches. I’m abuzz with everything I’ve been building.
And yet, I see the cracks. Designs are similar, the patterns ubiquitous, the endless gradients or sprinkled em dashes in prose like confetti strewn about Time Square on January first. The current talk for the past month is that taste is still required, is still what sets sites and programs and designs apart from this AI slop1.
I think, eventually, people will begin to understand that AI, in its current form, is truly only a tool and not the death of creativity or connection. These past few months, I don’t really write code like I used to. I still hop into codebases to make fixes/changes that is faster than explaining to the AI, but for the most part I’m working more as a systems architect, ensuring test coverage, and then having PAI2 run with the ISA (Instruction Set Architecture) file that was created during the planning phase. The results have largely been good.
As I alluded to above, I’m working on something new. At first, I used Claude for everything: design, copywriting, coding. I wanted to understand what truly was possible. My takeaway for the creative parts? Mediocre, generic, and reliably sameness. I actually signed up for Adobe Creative Cloud so that I could design the logo and flyers I’ve been working on. Taste will matter more now. Humans value the creations of other humans because we understand the toil and pain that comes from creative endeavors; AI doesn’t experience that and so there is nothing relatable.
I use AI for what I already know: code. And the only reason I get such good results is because I have enough experience and knowledge to direct it. I still make the core decisions about whatever I’m building. Humans are still necessary. We will forever be necessary. The value of AI-produced anything is simply a commodity. The value now relies on human interpretation and solutions to problem spaces, which I would argue was always the case.
See also my initial thoughts on being disciplined with AI.