Agent Anxiety

We know multitasking is a myth. The New Atlantis piece and Stanford research via NPR have said this clearly for years: we don't do two things at once, we context-switch, and every switch costs us. It takes roughly twenty-five minutes to recover from an interruption and get back into deep work. The brain has a bottleneck. You can override it with adrenaline for a while, but adrenaline has an invoice, and it comes due (kinda like our credit cards).

I used to wrap up my day when my brain gave out. That's when I'd shift to the mundane: expense reports (ugh), a little mindless sorting, maybe washing my socks or my cat (should you have either). The natural ceiling was cognitive capacity. When the creative energy ran out, I dropped to lower-stakes tasks. It was a self-regulating system. Simple, effective, and honest about what I had left to give.

That system is broken now. Or more precisely, I broke it with AI.

I teach this. I reference books like Slack when talking about resilience and why organizations under constant pressure lose the ability to respond to the unexpected. I point out that boredom is generative, that creativity shows up in the gaps. I wrote a whole post about interstitial spaces because I believe the pauses between things are where synthesis actually happens. Take a walk, stare out a window, make space, and more will show up. I stake professional credibility on this.

And yet.

Lately I've been wrestling with something I've started calling "agent anxiety." It's the low-grade discomfort I feel when I look at my agent fleet and see idle capacity just sitting there. An agent that could be working isn't, because I haven't given it a task. That gap feels like waste. Like opportunity missed. Like a meter running with nobody in the cab. And I feel pressed into action.

The behaviors this produces are not healthy. I find myself checking agents constantly: is this one still running? Is that one waiting on me for a decision? Am I leaving something valuable idle that I should be orchestrating right now? I struggle emotionally to regulate the feeling that I should be doing something about it, even as I know intellectually that I should not. The irony of checking on agents to reduce my own anxiety, while knowing the checking is creating more anxiety, is not lost on me.

The feeling is not rational. I know this. An idle agent costs nothing meaningful when it's not running. Giving up focused, high-quality time working so that I can context-switch and give an agent work is counterproductive. But knowing something is irrational doesn't make it go away. It just makes me a person who feels an irrational thing and knows it, which is somehow worse.

There's a version of this I wrote about a while back, from a different angle. Taking a day off and doing absolutely nothing can make you more anxious, not less. The intuition says rest recharges you. For people wired for effort, the reality is that unstructured ease amplifies restlessness. Agent anxiety runs the same loop in a different direction. Instead of anxious rest making me seek work, I've got anxious abundance making me seek more tasks to delegate. And being willing to break out of flow to capture that abundance is the very thing I rail against...and, yet, find myself doing. The capacity is the trigger either way.

Here's what I think is actually happening. Compute used to be the constraint. Human attention was the only thing you could marshal on demand, so we built instincts and systems around it: protect focus, batch your distractions, guard your deep-work hours. The Slack framework exists because organizations under too much pressure lose the resilience to respond to the unexpected. The same applies at the individual level. Boredom was productive because our brains, left to wander, do their best integration work.

Now compute is on demand. Human attention is still the scarce resource, but the ratio has shifted. In many cases, I have more processing capacity than well-formed tasks to feed it. That's genuinely new. And my brain, which evolved for scarcity, reads "available capacity I'm not using" as a problem to solve. So I start scanning, checking, monitoring. Am I being efficient? Is anything waiting on me? Could I be orchestrating something right now? I wrote about how work and rest aren't opposites, and that recovery and integration are legitimate parts of productive output. But I'm not actually resting when I'm doing this checking. I'm fragmenting my own attention while telling myself I'm being strategic. Twenty-five minutes of recovery, gone, because I needed to know if an agent had a status update.

The practical cost is real. I'm less focused. The monitoring is becoming a habit. And habits like this are self-reinforcing: the more I check, the more anxious I feel when I'm not checking, which makes me check more. That's the same feedback loop that multitasking creates, and I know how that one ends. It's not pretty.

Time is still the constraint. It always was. The agents didn't change that. What they changed is what competes for my attention inside that constraint and how easy it is to access the distraction of directing agents. I've added a new category of interruption and dressed it up as leverage. That's on me.

The human instinct that made us productive in a world of scarcity may be what makes us less productive in a world of abundance. I don't have the answer yet. I'm still inside the experiment. But I'm curious whether you've felt it too. What do you do when the anxiety of having too much capacity is doing the same damage as the anxiety of having too little? And more importantly: are you actually doing something about it, or just scheduling more tasks to soothe the feeling?