Underneath every agent framework is the same small loop: show the model the situation, let it think, maybe call a tool, feed the result back, repeat. You could write it in an afternoon, and for some jobs you should. What a framework actually sells is an opinion about who owns the next step: a graph you drew, a crew you cast, the model itself, a full kit with a deploy path, or an agent that edits its own tomorrow. This lab builds the same support-ticket agent behind each opinion, so the differences stop being brand names and start being trade-offs, and it ends by untangling the one family of names everyone mixes up.