A few years ago I gave a talk on protecting your focus: every interruption costs ~23 minutes, busy ≠ productive, guard the block. Then I spent a year building AI products with agents in the loop - and watched some of my own advice fail in real time. I still lose focus to context-switching, but now half the switches are between me and my own agents. Nobody sends them, and they feel like work. And "context" itself split in two: the one in my head, and the one in the model's window I keep feeding at the first one's expense. So I put my old advice on trial: what still holds, what quietly died, and what's brand new - me, a prompt, a wait, four half-finished agent runs. I'll show where deep work went (verifying the agent's work is the real thing now) and the moves that got my focus back: batch the agents and say good night, think first and hand off second, treat review as a real focus block. The habits that still earn their place when your pair-programmer never sleeps.
CTO at Idea to Product