Agents

Every "agent" you will ever hear about is: a context window, loaded with an instruction file, given reach into tools, run on a loop or a schedule. That is the entire trick. There is no additional being in there.

Which is why the word shouldn't intimidate you — you already understand every part. The instructions are steering. The reach is connectors or the command line. The loop is just software handing the window its own next prompt: perceive, decide, act, repeat.

When someone says "we deployed an AI agent," you now know exactly what to ask: What's in its window? What can it touch? Where does its truth live? And who reviews its work before it counts? Those four questions are the difference between an agent you can trust and a story you've been told.

Say it back in one line: what is an agent, mechanically — without the word "magic"?