PART IV PRIMER

What is AI, an AI agent and agentic AI?

Three terms that are often mixed up, explained from the ground up.

Supplementary — not from the source guide

1. What is AI?

Artificial intelligence is software that learns patterns from data instead of following only hand-written rules. A model is trained on many examples, then used to make predictions on new input: a label, a score, a number or — for large language models (LLMs) — the next words of a reply.

• Narrow AI does one task: spam filters, recommendations, fraud scores.
• Generative AI creates content: text (LLMs), images, audio, code.
• A plain model is reactive: one input in, one output out. It doesn't act or remember on its own.

2. What is an AI agent?

An AI agent wraps a model in a loop so it can pursue a goal. It reasons about the next step, acts by calling tools (search, APIs, code), observes the result and repeats until done. Your code runs the tools; the model only decides which to call.

• Brain: the LLM that plans each step.
• Tools: functions described by name and schema that the model may call.
• Memory: short-term (the conversation) and long-term (a database or vector store).
• Guardrails: step and cost limits, permissions, and human approval for risky actions.

Example: “Book me a table for 4 tonight.” The agent searches restaurants, checks availability, asks you to confirm, then books.

3. What is agentic AI?

Agentic AI is the wider system design approach where one or more agents work with real autonomy on long, multi-step goals. Usually a supervisor agent breaks the goal into tasks and hands them to specialist agents, each with their own tools, then combines and checks their work.

• Autonomy: decides its own steps over minutes or hours.
• Collaboration: several agents with isolated context pass condensed results.
• Durability: state saved after each step so long tasks can resume after failure.
• Cost: often several times the tokens of one agent — use only when one agent isn't enough.

Side-by-side