Cognitive Overload

June 2026

Anyone who's spent serious time working with agents has probably noticed: the level of exhaustion at the end of the day has spiked dramatically. It has for me.

We all became managers overnight — without learning how to set goals properly first. Some actual managers never quite figured that out either, so the rest of us are in good company. I'd put myself in the "not great at it" camp, even though I wrote a whole essay arguing goal ownership is the scarce skill and still hit the wall by Friday.

But goal-setting isn't the only problem. We're now managing not people, but an entire fleet of agents. And that fleet's availability triggers something primal in my inner resource manager — an irresistible urge to assign it every task in existence, because the resource pool feels infinite. When agents aren't running, a little voice says: they're on the bench — paid for, better keep them busy. Agents may not be the sharpest tools in the shed, but they are extraordinarily diligent and obedient virtual counterparts, and their "development plan" gets implemented instantly.

There's something else that grinds you down. This fleet is different from people in one critical way: the feedback loop is nearly instant. And unlike people, agents don't take smoke breaks — that moment when your colleague suddenly realizes they made a mistake, or gets a better idea in the stairwell. No smoke break for them means no smoke break for you either.

The bigger the work-package, the more an error cascades down the chain. And everything they produce needs to be read and checked. In theory. Ideally, not just patched point-by-point, but traced back: where was the goal wrong? Why didn't you get what you wanted?

In practice, everyone has suddenly become a senior manager getting bombarded from all sides — deliverables, decisions, documents of questionable quality, occasionally good ones on the first try. If you think it's different with humans and the problem is the models — oh sweet summer child. You just saw the problem. It was always there.

Honestly, it makes a decent test for a manager: don't let anyone manage people until they've gotten an agent to do a medium-complexity task on the first try — and can show you exactly how they organized a team of agents to pull it off. But the test isn't enough. Because with infinite resources, any fool can manage. Try it with people. They get tired, they sleep for some reason, they wander off for tea, they disagree with you — or they just don't do what you asked at all.

This is essentially the moment of transition into management. I've seen it happen more than once: someone tries to move into management and hits a glass ceiling because of the sudden explosion of context. They just can't process it all. Trained correctly, thinking right — but not making it through. Many stepped back. But some adapted to the new load, and after a while stopped treating it as anything special. It just became part of the routine.

Our relationship with agents will get there too. We'll adapt, tune, adjust. Throughout human history, some fundamental technologies multiplied speed, and others created tools to let people actually use that speed for their own purposes. Nothing new — just a shorter cycle.

But for now, if you're ending the week with brutal cognitive overload — I'm in the boat with you.