The zoning out got better for me when I stopped treating the run as waiting time and started using it as the slot where I write the check, new prompt, new tasks.
While it works, I review code, check new GUI, communicate on socials about my product. Mostly code review, prompting, because I run 3-4 agents in parallel on different coding tasks. They finish at a different time, so no free time for a rest.
Solo here. So pairing was never my baseline, but the thing agents changed for me is the review rather than the writing. I used to read my own diff once and ship it. Now I read it like it came from someone else, because it did, and the failure mode is different.
So from my own experience two agents are better. If setup properly they will collaborate and argue about code. For speedup use two or more agents if solo
Yep, geolocation is so creepy. And claude code will read your personal data, rc, .env, other projects files if you forgot to restrict it. So I have global settings for that.