A shift in agency
Coding with AI started for me as code completion and then I had some chats about some API or the shape of a class or a feature from time to time. I then started to enjoy AI coding when I managed to use Claude Code to basically do pair-programming. I navigated, while Claude Code executed, and we explored the best options together.
That approach of pair programming with claude code was short-lived for me, as agentic coding has become the default. I shifted to more agentic coding to try it out, but now it feels like Claude Code won’t really let me use anything but an agentic approach. I can set the thinking effort to low and after a couple of questions, a few comments of mine, Claude wants to run off and implement it. More and more often I actually stop Claude to say: “Hey, I am not done talking, I want to discuss this more.”
It is impressive what AI can do, it is quite thorough, thinks about many angles, I don’t even ask it to consider. But there is definitely a shift in agency when I use Claude Code nowadays. Claude tries to be more autonomous, be more proactive. It goes off spinning up agents and it’s letting me wait, as If I was there for its benefit. It’s harder to just keep a conversation going with a quick back and forth. I feel like I have to rein Claude in and give boundaries. In the past it was more that you told it what steps to take, what to do. Now it makes assumptions, goes off and does its own thing.
Claude feels like an overeager teenager, trying too hard to prove themselves. Only that it’s not my child, I don’t educate it. Some company does. And there are a million versions of that teenager out there. So who’s in control?
Because now we start seeing things like these:
Let’s keep the agency to ourselves please.
