Gather the chapter. An agent is the next-word model you already understand, given hands (tools it can ask to use) and run in a loop (observe, decide, act, repeat) over the same context-window board from last module. Give it a tool to look things up, and it can even reach past its frozen memory. That is genuinely new: a system that does not just talk about the world but reaches into it.

But you also felt the catch, by stepping the loop yourself. Each lap builds on the last, which is exactly why a single early slip can ride all the way to a confident, wrong ending. So power and risk arrive together, from the very same loop.

The next chapter takes that seriously: where agents genuinely shine, where they break, and how much you should trust one to run on its own.