An agent is a loop where a model proposes actions and tools carry them out with checks. This part covers the loop, structured tool use, planning with memory and budgets, and when simpler scripts fit.
Topic | Lessons | Description | Time |
|---|---|---|---|
5 | An agent is a loop where a model picks actions and tools carry them out. | 40 min | |
5 | How models use tools through proposed calls, schemas, validation, and safe execution. | 40 min | |
4 | Decompose tasks, keep memory outside the model, and stop loops on budget. | 30 min | |
4 | Planner, worker and reviewer patterns, hand-off cost, error compounding, and when a script wins. | 30 min |
When the steps are known and repeatable: a script follows the same path each time while an agent adds planning and hand-off cost without benefit. The When a Script Wins lesson and the Script, Workflow or Agent lesson cover the choice.