Exam-style
A support team runs three systems on the same foundation model. The first drafts a reply to a customer email, which a support agent edits and sends. The second answers staff questions from the policy library, citing its sources. The third reads an email, looks up the order, applies the refund rules, issues the refund and writes back, its own output selecting which steps a given email needs. Which is agentic AI, and why?
Reveal the answer
C. The third, because its own output selects the sequence of steps and it carries them out against external systems rather than producing text for a person to act on
AWS defines agentic AI as an autonomous system that acts on its own to reach a goal, working through tools it can call, memory that carries across steps, and a plan assembled as it runs. Only the third system fits. It reads the email, queries the order system, applies the refund rules, moves money and replies, with nobody signing off in between, and systems outside the model hold different data afterwards. The first generates and stops: the draft sits in an inbox until a support agent edits it and presses send, so the person acts. The second is retrieval-augmented generation. A search over the policy library returns passages, those passages go into the prompt, and the answer cites where each part came from, which makes it traceable. Retrieval can be a tool an agent invokes, as Amazon Bedrock Knowledge Bases supports, and either way the output lands in front of a person. Agent count is a separate matter: AWS describes single-agent and multi-agent systems alike as agentic. Building all three on foundation models makes them all generative AI applications, the wider label.
Q. One system drafts a reply for a support agent to send. One answers questions from a document library and cites its sources. One reads an email, checks the order, issues the refund and writes back, its output selecting which steps run. Which is agentic AI?
A. The third. Agentic AI is a generative model given tools, memory and the autonomy to take multi-step action against external systems, instead of producing text for a person to act on.
Why? Artificial intelligence (AI) is the outer label, generative AI (GenAI) is the part that produces new content, and a large language model (LLM) sits at the centre of all three. The shared machinery is why they look alike. Sort them by what happens to the output. The drafter stops at text, and a person sends it. The document-library system is retrieval-augmented generation, which grounds the answer in retrieved passages so it can be cited and checked. Whether retrieval runs as a fixed pipeline step or as a tool an agent invokes, the answer still lands in front of a person. Only the refund system’s output selects the next step and runs it against systems outside the model. One agent on its own is enough; a second is not part of the definition.