Finding balance with context switching

I've done a lot of building with AI coding agents over the last couple weeks — standing up new projects, pushing tools like Claude harder than I ever have. Going into it, I had assumed that the people who could context switch the most in this new world were the ones set up best for success. And that I — someone who craves deep, focused work — was not built for this new world.

So I put that to the test.

The old way: minimize context switching

If I go back a decade to my Rails days, when I was doing deep development daily, I was most productive when I was in a hole and shut the world out. I couldn't do it all day, but for 3-4 hours at a time, that's how I did my best work. One thing at a time. Crank.

Writing code was most productive when I had dedicated blocks of time to minimize context switching. I think many developers would agree with that.

The new theory: maximize context switching

With AI doing the coding — and seeing people spin up whole applications virtually overnight — I wondered if we might be living in an opposite world. I wondered if working with AI was actually all about context switching, because in the end, all we're doing is providing context so that the AI agent can do its work.

I spent the better part of two full days absolutely hammering on Claude — further than I've ever pushed it. I was running multiple terminal sessions, bouncing between them, trying to build three or four applications at the same time.

Three streams is my max

What I found was that four concurrent sessions was too many. It was just too many threads of thought, and there was always some interaction I'd need with the first project before I could get to the fourth.

This was while actively standing up applications from scratch — not just adding one little feature, but doing the heavy lifting of building something new. I felt too slow when I was working on only one, and an agent was always waiting when I was working on four. So, depending on the length of feedback cycles (which is somewhat dependent on what I'm building and where I am in the process), having two to three concurrent projects has felt good.

Staying open to evolution

I tested this theory, and I was wrong — at least partially. There is a new balance for me, and it really does make me feel ultra productive. But that doesn't mean this is the way I'll always work.

One thing I found throughout this process is that human thought is the key to making all of this work. Even though coding agents can work for longer and longer in isolation, they still need that human input. But how I work, when and where that input comes from — that's going to keep changing.

Maybe these two to three synchronous projects won't be the case in a month. Maybe as some of these projects I've built mature, I start working on more, but smaller features. I don't really know what the future holds, but I know I have to be open to adaptation. Things are changing so fast right now, and I need to be able and willing to change my working style to keep up.

The balance I found this week made me feel more productive than I have in a long time. But I'm not holding onto it too tightly — I'd rather stay curious about what comes next and enjoy the build.