The most common programmatic interaction with Large Language Models (LLMs) and LLM APIs (ChatGPT, Claude) is to give it some natural language instructions and get a shaped, specific output back. For example you might ask it to summarize a news article for you, and have it respond only with the summary, for storage and further processing later. More advanced applications might have the LLM acting as an agent with tooling that needs to be invoked, so it outputs (in JSON) a tool name with some a...
