The correct option is ReAct approach.

This pattern fits because the agent explicitly reasons about the subtasks it must perform, decides to call external tools to fetch live weather and look up a fact, executes those tool calls, and then synthesizes the final answer. That think and act loop is the defining characteristic of this approach.

Chain-of-Thought prompting is not correct because it focuses on guiding the model to show its reasoning steps, yet it does not require the model to invoke tools or perform actions to gather fresh facts.

Retrieval-augmented generation is not correct because RAG emphasizes retrieving documents or knowledge snippets to ground the answer, which is usually a single retrieval step that informs generation. The scenario highlights iterative reasoning and multiple tool calls rather than only retrieval to ground the response.

Few-shot prompting is not correct because few-shot relies on exemplars in the prompt to shape behavior. It does not inherently involve planning subtasks or calling external tools.

Full AWS Practitioner Certification Question