What makes an AI an "agent"
What you’ll learn
Section titled “What you’ll learn”This is the opening lesson of Track 20 (AI Agents and Tool Use) and the first lesson of Phase 1, What an agent actually is. It answers the question the marketing skips: what is an AI agent, precisely? The answer is one idea you will use for the rest of the track. An agent is a model wrapped in a loop: it looks at a goal, decides whether a tool would help, calls the tool, reads the result, and repeats until the goal is met. The lesson contrasts that with a plain chatbot (one shot, sealed off from the world), traces the loop through a multi-step task, and takes apart the four parts that make any agent work.
The track structurally mirrors Microsoft’s “AI Agents for Beginners” (MIT-licensed), with the Berkeley CS294 LLM Agents course as a depth reference. Full attribution is in this lesson’s references.
Where this fits
Section titled “Where this fits”This is lesson 1 of the track, so nothing comes before it. It establishes the vocabulary (the loop, the four parts, agency-versus-intelligence) that every later lesson builds on. The next lesson, How tool use turns a model into an agent, opens up the single move that powers the loop: how the model emits a tool call, the loop runs it, and the result comes back. After that the track builds out frameworks, memory, retrieval, planning, multiple agents, self-checking, and finally trust and security.
Before you start
Section titled “Before you start”Prerequisites: none beyond having used an AI chat tool at least once. You do not need to code, and there is no math in this lesson. If you have ever typed a question into a chatbot and gotten an answer, you have the only background this lesson assumes.
By the end, you’ll be able to
Section titled “By the end, you’ll be able to”- Define an agent by its perceive-decide-act loop and distinguish it from a single model call
- Identify the four parts that make a system agentic (model, system prompt, tools, loop)
- Explain why agency comes from the scaffolding around the model, not from the model’s strength
- Trace the agent loop through a simple multi-step task
- Judge when a task warrants an agent versus a plain model call
Time and difficulty
Section titled “Time and difficulty”- Read time: about 11 minutes
- Practice time: about 15 minutes (two applied exercises plus flashcards)
- Difficulty: intro