Polle de Maagt

The brain fry is real.

By Polle de Maagt · First written · 3 min read

Supervising AI agents is real work, and it drains you in a way that ordinary task switching does not.

A home office late in the evening, two screens glowing with many windows, a half-finished coffee, the chair pushed back
Two screens, many windows, cold coffee. Image made with Midjourney.

The fog.

I work with a stack of agents. One watches trends, one keeps my books, one prepares a meeting. Most days that is a joy. Some days it is not.

On those days I am not working, I am switching. I check one agent, wait for it, switch to the next, wait again, then come back to the first to see if it finished. By the evening I have read a lot and done little. Some weekday evenings I was simply exhausted. Mentally drained, a foggy head. And frustrated, because nothing was moving forward.

The name for it is brain fry. Alexander Klopping describes the mechanism better than I can: the cause is not the work itself, it is the context switching between agents.

It is not only me. In March 2026, researchers from BCG and UC Riverside reported in HBR that 14 per cent of workers who use AI experience brain fry, rising to 26 per cent in marketing. Those who do report 33 per cent more decision fatigue and 39 per cent more major errors.

There is a second drain I underestimated: keeping the setup itself running. Things break. Things could be better. There are updates. Constantly optimising your AI structure and your data structure takes real time, on top of the work it is meant to save. That wore me out too.

What I changed.

Four things came out of one question I asked myself in my quarterly review: what can run without me watching it?

  1. More autonomy. I stopped looking for what an agent can help me with, and started looking for what it can finish on its own. Waiting is the tax.
  2. Two machines. My iMac and my laptop are both open, each with its own work, so I stop hopping between agents on one screen.
  3. Fewer switches. Mornings are for deep work: one document, full focus, no switching. I check the agents in batches after that, not continuously.
  4. Deep blocks, and more project work. I plan my week in long blocks of deep thinking. The blocks work, so I rebuilt part of the week around them and work in projects instead of a stream of small tasks.

The lesson I take from it: the fix is less supervision, not fewer agents.

Like a busy wedding after which you lie in bed with tinnitus. Alexander Klopping

Try this

Pick the agent task you check most often, give it a finish line it can reach without you, and leave it alone until it is done.

Sources: Alexander Klopping on LinkedIn, from a linked podcast episode, July 2026. When Using AI Leads to Brain Fry, Harvard Business Review, March 2026.

Your version of this

Tried it, or thinking about it?

Let me know how it went.

Let's talk →How I work
Polle de Maagt

Keep reading

All thoughts →