Running Many Agents in Parallel is a Bad Idea

May 13, 2026 Rens Jaspers AI Reflections Workflow

Previously I wrote that my brain no longer gets tired from coding, and multitasking felt fine now.

I was too quick with my assumption about multitasking. Multitasking is still not fine.

Waiting for one agent gets dull, so over time I've been adding more and more parallel sessions to my workflow.

But sadly, constantly switching between multiple agents turns out to be mentally exhausting. The energy saved by letting agents write my code does not make up for it at all. Context switching is the most expensive thing I do with my brain.

Other developers are complaining about this as well.

Tech entrepreneur and vibe coding expert Alexander Klöpping has an interesting fix for his AI-induced "brain fry": an AI chief of staff that steers other agents and keeps him from multitasking past his limit.

Multitask mode in a single Cursor thread is a smaller step in that direction. You have one conversation where the agent tracks parallel work, so you don't have to watch every background tab yourself.

This is an improvement, but the real solution is to resist the urge to multitask altogether. That's hard, since we're used to never being bored thanks to the entertainment machines in our pockets.

I'm currently reading Het grote ontprikkelboek by Brankele Frank, a Dutch book about calming an overstimulated brain. It describes two thinking modes: focused mode and default mode—the mind-wandering you do when you're not actively thinking about anything. Healthy brains balance both. Activating your default mode network will also often allow you to find interesting solutions to problems that you weren't able to find while in focused mode.

The book makes it clear that the human brain wasn't built to multitask at all. Research shows that people who consider themselves "good multitaskers" perform worse.

From now on, I'll be doing more monotasking and mind wandering. Will report back in the future.