Parallelism, the Responses API, and the App Server

So far the loop has run one tool at a time. But a model often wants several things at once: read three files, run two independent checks. Doing those sequentially wastes time. This chapter covers three connected upgrades: running independent tool calls in parallel, the API shape that was designed for agents (OpenAI's Responses API), and the protocol that lets a single harness power a terminal, an IDE, and a website (the App Server). The first is something you build; the other two are mostly things you understand so you make good choices.