I’ve been building this weekend. It felt really good. Mostly just exploring some ideas in Replit, Codex, and Claude. Yes, I’ve been using all three. There are some overlaps, some distinct quirks, and some ease of use aspects. I am by no means an expert in any of the tools, and so most of this exploration in how to break an idea down, and utilize their respective capabilities to benefit my aims. One of the obvious conclusions I’ve learned is that not everything is an app.
Yeah, totally obvious. For someone who has almost no formal technical background, it’s a groundbreaking level of understanding. And while this could turn into an indictment of every “hello world” tutorial I’ve done and then subsequently quit, there were some truly simple levels of understanding that came together in the moments of playing around with the various agents. I started to see the pros and cons of deterministic vs nondeterministic code. I started to understand where a Python or Bash script could be more useful than Javascript with a slick UI.
And then the technology starts to become clearer. Where there’s hype – AI will NOT solve everything and it definitely has downsides. And where it’s just like other technologies – it will change the way we work – in some of the ways we assume and some that we don’t see coming. It feels a lot like cars. Horses weren’t outright replaced (try driving the Grand Canyon) and cars weren’t perfect overnight: here, the infrastructure piece is critical. AI will need roads…and given how much is being spent on data centers and the like, maybe we’ll see a bicycle effect: because roads exist, biking becomes more adoptable.
So maybe there’s a question there: what’s the equivalent of biking when it comes to all this new technology? Something cheap, efficient, mildly counterculture but also sustainable and human?