AI is coming for the rubber ducks

My copilot went from code monkey to the one I think out loud with.

In software development there’s a concept called rubber ducking. It’s when, while trying to find out what went wrong in your code, you talk to someone imaginary and explain what you’re doing. The really devout put an actual rubber duck next to their desk and start talking to it (hence the name).

Besides reinforcing the stigma about how weird programmers are, it’s a process that’s generally been shown to help with solving problems. Sometimes in your head you skip things “because they’re trivial”, and just saying them out loud is enough to see what the problem is.

Why am I telling you this? Because lately I’ve picked up a habit of “talking” to Copilot, or to whatever AI tool I’m working with at the moment. During development I keep writing to it what I want to do now, why I’m doing it and how I think it should be implemented.

A lot of the time it’s just there with me, saying “sounds good bro, go for it”. But it also often asks me more questions along the way, offers another angle, or puts the information aside and brings it up later.

That’s very different from how I used it at first, when it was simply my executor: I told it what to do, down to which function and how. Slowly its role is becoming less code monkey and more of a partner through the whole process.

It’s also a good habit. It helps me not miss things, and on the other side it gives the tool more context and therefore better results. (Although sometimes that context isn’t relevant, like when and where I’m going to give the talk we’re working on slides for. But why not.)

In short, the first industry AI is going to kill is rubber ducks.

This post started as a LinkedIn post.