Module 03 · Agent

Tool use and building an agent

Give a model hands. Write schemas it can read, build the ReAct loop from scratch, answer a whole batch of parallel calls including the ones that failed, and end every run with a reason you can act on.

5 lessons · 4 graded exercises·5.5 hours·intermediate·Prerequisite: Prompt engineering

What makes an agent an agent

25 min

A loop, the ability to act, and a reason to stop. Most of what gets called an agent is further down that list than the marketing suggests, and knowing where you are is the useful part.

Describing tools so the model picks the right one

35 min · exercise

The schema is a prompt. The model sees a name, a description and your parameter descriptions, and nothing else, so most wrong-tool bugs are writing bugs.

The ReAct loop, from scratch

60 min · exercise

Thought, action, observation, repeat. Forty lines, exact message bookkeeping, and a stopping condition that raises rather than shrugging.

When one turn asks for four tools

40 min · exercise

A single assistant message can carry several calls. You owe the model a result for every one of them, including the two that failed, and the batch has to stay in order while it runs out of order.

When tools fail

45 min · exercise

Four failures, and only three of them belong in the conversation. The fourth belongs in a retry, and the run has to end with something a caller can act on.