My fix for AI agent wait time was running more sessions. I just ended up with more waits.
I use Claude Code all day, and I kept catching myself drifting during every wait. Submit a prompt, the agent runs for two or three minutes, and somewhere in there I'd open shorts and surface twenty minutes later, annoyed at myself.
My first fix was the obvious one: run more sessions in parallel, so there's always something to do. It didn't work. Now I had three terminals working and I was still watching shorts, except I was also losing track of which session had finished and which one was sitting there blocked on a permission prompt, waiting for me while I wasn't looking.
That wasted time bothered me enough that I built Unwait: a Mac menu bar app that shows one flashcard (vocab, dev concepts, news) while a session runs, and when any session finishes or gets blocked, it tells me exactly which one and one click takes me back to that terminal.
Honest report after weeks of daily use: it did not cure me. I still drift sometimes. But the per-session notifications changed the economics of drifting. Before, a distraction cost me the distraction plus however long the agent sat idle after finishing. Now the idle half is gone, because I know the moment something needs me and which window it is. The flashcards fill the short waits; the notifications protect the long ones.
Every developer friend I showed it to had the same pain point, so I cleaned it up and launched it: 14-day trial, no signup, $1.99/mo after. macOS only for now.
Curious how other people here handle this. If you run coding agents daily, what do you actually do during the waits? And has anyone else found that parallel sessions made the distraction worse, not better?
