Objective
Get the whole room to feel the line between a plain chatbot and a real agent. A task is "agent work" when it needs several steps, a tool, and a loop to finish — not just a single reply. By the end, students can hear a task and instantly know which one it is.
The setup
- Clear a path down the middle of the room. Label one wall CHATBOT ("answers in one shot") and the opposite wall AGENT ("needs steps + tools + a loop").
- Have your list of task cards ready (below). Everyone stands in the middle.
The rules — read aloud
Round 1 · "Easy calls"
- "I'll read a task. Walk to the side you think it is — CHATBOT or AGENT."
- "I'll pick one person to defend their choice in one sentence: does it need tools and a loop, or just one answer?"
- Try: "Define the word photosynthesis" (chatbot) · "Book me the cheapest flight to NYC next Friday" (agent) · "Translate 'hello' to Spanish" (chatbot) · "Track this package and text me when it's delivered" (agent).
Round 2 · "Tricky ones — expect a split"
- "Same game, but these are sneaky. If the room splits, argue it out for 20 seconds."
- Try: "Summarize this paragraph I paste" (chatbot — one step) vs "Read the latest news and summarize what's trending" (agent — must search first) · "What's 12 × 47?" (chatbot can try, but an agent would use a calculator tool) · "Plan and order groceries for a week of dinners" (agent).
The process
- Round 1 (~4 min): quick calls, one-sentence defenses. Keep the pace fast and physical.
- Name it: "Chatbot = one answer. Agent = it has to go do things first."
- Round 2 (~5 min): the tricky cards. Let the splits spark a short debate — the disagreement is the learning.
- Land it: "The tell for an agent: it needs tools and more than one step, run in a loop."
The debrief
- What word in a task usually signals "agent"? (verbs like book, track, monitor, find-then-do.)
- Could a chatbot become an agent for the same task? (Yes — the moment you give it tools + a loop.)
- Finish this: "A chatbot answers. An agent ______." (plans · uses tools · loops · finishes the job)
WORKSHOP TIE-IN: In today's Colab build you'll take a plain LLM (chatbot) and turn it into an agent by giving it two tools (a calculator + today's date) and running the loop — the exact upgrade you just sorted for.
WATCH-OUT: Don't let students over-think Round 1 — go fast and keep bodies moving; the energy is the point. Save the debate for the tricky Round 2 cards.