◆ LearnAILevel 1.5 · Agents ← Roadmap

Station 1 · AI Agents 101

What is an AI agent?

A chatbot answers your question and stops. An agent takes a goal and keeps working — planning, using tools, and checking its own results — until the job is actually done.

1 of 10 · ~6 min

What you'll walk away with

Concept 1

Chatbot vs. agent

You've used a chatbot: you type, it replies, done. That's one turn. An agent is what you get when you let the model take actions and keep going toward a goal.

Just talks

Chatbot / plain LLM

One question in, one answer out. It can only use what it already "knows." If it's unsure, it guesses.

Talks + acts

AI agent

Given a goal, it plans steps, uses tools to get real information, checks the result, and loops until it's done.

🧑‍🍳 Analogy: recipe reader vs. chef

A chatbot is someone who can recite a recipe. An agent is a chef: they read the goal ("make dinner"), grab tools (knife, stove), taste as they go, fix the seasoning, and don't stop until the meal is on the plate. Same knowledge — but one of them actually cooks.

Concept 2

The thing that makes it an agent: tools

On its own, an LLM can only produce text. A tool is an ability you hand it — a function it can choose to call — that lets it reach outside itself.

🔑 The core idea

An agent = an LLM + tools + a loop. The LLM is the brain, tools are the hands, and the loop is what keeps it going until the goal is met.

Concept 3

Watch an agent think

Here's the same request handled two ways. Notice the agent doesn't guess — it uses a tool to find out.

chatbot vs agent
# You ask: "What's the weather in Austin right now, and should I bike?"

# CHATBOT (no tools):
"I can't know live weather, but Austin is usually warm..."   # a guess

# AGENT (has a weather tool):
Thought  → "I need live weather. Call the tool."
Action  → get_weather("Austin, TX")
Observe → 41°F, raining
Answer  → "It's 41°F and raining — maybe skip the bike today."
the differenceThe chatbot hedges. The agent acts, reads a real result, and answers with confidence — because it actually checked.

Concept 4

What agents are used for

Straight from the real-world map — agents show up wherever a task needs more than one step:

🗒️ Personal assistants

Book things, draft replies, manage a to-do list across apps.

💻 Code generation

Read a codebase, write & edit files, run tests, fix errors (Cursor, Claude Code).

📊 Data analysis

Load a dataset, run queries, make charts, explain findings.

🕸️ Web research

Search, read pages, and summarize across many sources.

Every one of these is the same recipe — an LLM using tools in a loop. The rest of this roadmap unpacks each piece: the loop (Station 3), tools (Station 4), memory (Station 6), and more.

Checkpoint

A friend says "ChatGPT is an AI agent." When is that most accurate?

Always — any chatbot is an agent. Only when it can use tools and take actions in a loop, not just reply. Never — agents aren't real yet.

A plain chat turn is just an LLM answering. It becomes an agent when it's given tools and a loop so it can act, observe, and keep going toward a goal. (Modern ChatGPT can do this when it browses or runs code — that's the agent mode.)

See it yourself — 5 minutes

Spot the moment a chatbot becomes an agent

Open any AI chatbot that can browse the web or run code (ChatGPT, Gemini, or Claude) and try this:

  1. Ask something it can't know from memory: "What's the top headline on a news site right now?" Watch it use a tool (browse) instead of guessing.
  2. Ask it to "calculate 4839 × 2917 and show your steps." Notice if it runs code vs. estimates.
  3. Each time it goes and does something before answering — that's the agent loop firing.

Recap. A chatbot talks; an agent acts. The magic ingredient is tools — abilities the LLM can call — run inside a loop until the goal is met. Next, let's look at the engine all of this runs on: the LLM itself.