@jxnlco
subagent are great when you need to spin a subprocess to burn a lot of agents that would otherwise pollute the main context, Here we created a sub-agent whose job is to simply read the entire site map of my website to propose cross-links. Even a burn, a lot of tokens reading other blog posts to make sure that these crossings are done correctly, but then will ultimately give me a plan to actually make the insertions. This way I can kick up this task without setting up another tab, but then continue my context of actually working on the blog post, which is what the purpose of the main thread is.