Is context switching solved?

As I write this, five other things are happening on my computer. One is fixing a bug, one is reading through logs I did not want to read, and one is building a small page I will probably throw away. They are all AI agents, and they are all working while I sit here typing this. Not long ago that would have been impossible for me.

Every engineer knows the tax. You are deep in something, holding a whole fragile structure in your head: this function calls that one, this variable is wrong for some reason, the fix is probably around here. Someone asks you a quick question, you answer it, and when you come back the structure has fallen over. Gloria Mark, who studies attention at UC Irvine, puts it at about 23 minutes to get back to a task after an interruption. I believe it, and some days it felt like more. So the dream was always to leave and come back without losing anything.

Then it kind of happened. With agents, the context does not live in my head anymore. It lives in the session. I can walk away from a task in the middle of a thought, come back an hour later, and everything is there: what we tried, what failed, what the plan was. It is like having a colleague who never forgets anything and never sighs when you ask “wait, where were we?” So I did the obvious thing and started running more things at once. Two, then three, then five on a good day, or what I thought was a good day. On paper, it worked. I got more done than ever.

At the end of those days I could not tell you what I had done. I was tired the way you are after a day of meetings, scattered, like I had read six books one page at a time. It took me a while to see that the switching had changed shape. I don’t have to reload the code into my head anymore. But I do have to reload the why. Why did I start this one? What did “done” mean here? Is this answer right, or does it only sound confident? An agent will happily hand you something that looks finished. Deciding whether it is finished is still my job, and it needs the real me, fully loaded, every single time. So with five things open, I pay a smaller switching tax much more often, and it adds up to more than before.

I run fewer things at once now: three, mostly, and one when it matters. Before I start something, I write down what done looks like, in one sentence. Then when I come back to it, I check that one sentence instead of reloading a vague feeling. And I try to finish things and close the tab, because an open task is a small weight even when nothing is happening in it.

For the computer, context switching is mostly solved. It holds the context better than I ever did. For me, it is not. The machine can hold where we were. It cannot hold whether I care about the answer, or whether the answer is good. That part still lives in one head, and there is only one of me.