LLMs & Agent Basics

A chat call is stateless: prompt in, text out. An agent wraps the model in a loop that plans, calls tools such as market-data or filing-search functions, reads the results, and decides the next step, with memory carrying state between iterations. That loop separates asking a question from delegating a task. On a desk, tools turn a language model into an analyst that can look up a filing or run a screen rather than confabulate from parametric memory. The craft shifts toward prompts, tool schemas, and context: prompt context becomes the new codebase, versioned, reviewed, and tested like source, because small wording changes silently alter behavior downstream.