I was used to working on one task, one screen, in one application window.

Arguments in favor that keep spinning in my head:

• Boris Cherny (Head of Claude Code) works across 5 parallel instances.
• Peter Steinberger (creator of OpenClaw) runs 3-8 Codex terminals at once.

It makes sense on paper – why watch AI "think" for 5 minutes when you could be moving another task forward?

So last week I tried it. Two important tasks, parallel terminals, switching between them, checking status. 90 minutes later – both done. Great output. Top performance.

Then I hit a wall.

Brain completely drained. Mental energy at zero. Good thing lunch came – I wouldn’t have made it through the afternoon.

The only parallel setup that actually worked for me: running one low-priority task in the background. Something that didn’t need my attention. Could wait hours before I checked the result.

In a world obsessed with multitasking, singletasking might be the real superpower.

Maybe the bottleneck isn’t the AI. It’s my cognitive switching cost.