As someone who also loves the act of programming, I find the idea of transitioning to using AI agents difficult. Not because they are better or worse at the job, but because it shifts me into a role of writing specs and shepherding robot monkeys with typewriters. I hope I'm wrong.
I also love programming. Out of curiosity (and possibly job stability) I've spent the past solid week building an app from scratch using Claude code.
I'm overall impressed with the result. There are things I might have approached differently, and there are things I would have gotten done much faster, but the result is more thorough than I might have done it.
I think what's most interesting is that I've never written specs to this level of detail before. I now have this series of project specs that hold every decision and consideration of this project written in plain english that's incredibly easy for any human to understand regardless of their ability develop software.
Whether or not I keep coding this way, I think this tool is incredible for figuring out exactly what to work on and how to approach it
I love programming and I love building things and these are two separate things.
I've been making a lot of stuff vibe coding, where I'm only talking to the agent about what I want to build, features and such. That's really cool but doesn't scratch the programming itch at all.
However you CAN talk to the agent about code. This is a lot slower to make things than talking about features. But if your goal is to make a codebase that's easy to work with and that you fully understand and have ownership of, you can talk to the agent about code instead of about features and then work with the output rather than just accept the output.
since i actually love the act of solving real-world problems by building programs (and not the act of programming itself), writing specs and shepherding robot monkeys with typewriters is an acceptable means to an end
