one side effect of treating LLMs like humans is people keep using them ineffectively you can't break a task up into 5 pieces and work on all of them in parallel and non-linearly but the agent can. you don't even have to tell it to, if it has the right environment it will do it