Level 1.5 · Agents · Session 1 of 3

Meet the agent

What an agent is · the LLM engine · the agent loop — then build a working one.

Covers: Stations 1–3 · You'll need: the Session 1 Colab notebook + a free Gemini key

The shape of today

Learn the idea → build it

Three ideas, one build

What is an agent? (talks vs acts)
The engine underneath (a quick LLM recap)
The agent loop — then you build a real one in Colab

Idea 1

Chatbot vs. agent

Concept · Two kinds of AI

Talks vs. acts

Just talks

Chatbot

One question in, one answer out. If unsure, it guesses.

Talks + acts

Agent

Given a goal, it plans, uses tools, checks results, and loops until done.

🔑 The recipe

Agent = LLM + tools + a loop. Brain, hands, and a cycle that keeps going.

Idea 2 · The engine

What it runs on: an LLM

Reads text as tokens; the context window is how much it sees at once.
Temperature tunes creativity — low = reliable (good for tool use).
Open (run-your-own) vs closed (API) models — we'll use Gemini's free API.

Recap

This is Level 1 material — just enough to build agents. Deeper dive lives in Level 1 · Foundations.

Idea 3 · The loop

The agent loop

while not done:
    # Perceive  read the goal + any new info
    # Reason   decide the next step
    # Act      call a tool
    # Observe  read the result → repeat

🔑 Remember

Perceive → Reason → Act → Observe → repeat. This loop (a.k.a. ReAct) is every agent.

⚡ Energizer · 10–12 min · on your feet

Agent or Not? — The Sorting Line

Coach reads a task; students move to the AGENT or CHATBOT side.
Defend it: does it need multiple steps + tools + looping, or one answer?

Printable card in Coach HQ → Session 1 energizer

▶

Build it

Make an agent in Colab

Build · the whole idea

A tool + a loop

for step in range(6):
    resp = gemini(history, tools=[calculator, today])
    if resp.wants_tool:              # Reason → Act
        out = run(resp.tool, resp.args)
        history += observe(out)       # Observe
    else:
        print(resp.text); break       # goal reached

Goal in the notebook

"What's today's date, and 1440 × 365?" — watch it call today and calculator instead of guessing.

Session 1 · Wrap

You built an agent 🎉

Chatbot talks; an agent acts (LLM + tools + loop).
You ran Perceive → Reason → Act → Observe in real code.

Next session → Powering the agent: more tools, sharper prompts, and memory.