These days LLM agents such as Claude Code/Codex are getting more and more popular, In my daily work, I leverage them to do much portion of the development work, but I am wondering how people manage time efficiently when LLM agent is doing the work.

Take me as a example, when LLM agent doing the work especially I turn on the auto accept mode. I just do the following things:

1. thinking about designs of new task or read some document related of the project I am working on 2. surfing on hacker news:)

but it seems the context switching bother me. when I jump back to check how the LLM agent do the work it seems slowly to get into the context.

Maybe this is the new norm we need to adapt to?