1
From Design to Build
5:00 – 5:15 PM 15 min
--:--
Key Talking Points
  • Session 6 was design on paper. Session 7 was knowledge. Today we BUILD a multi-agent system.
  • The org chart analogy: Why do companies have departments? Because one person can't be an expert at everything.
  • Your pipeline today: Researcher → Analyst → Editor. Three specialists, one workflow.
  • Key question to plant: "When does adding another agent help vs. just adding complexity?"
Demo Cues
OPEN: Lecture Guide — Sections 1-2 (Org Chart + Three Patterns)
Ask the Class
“Think about your job or internship. What happens when one person tries to do research, analysis, AND writing all at once? How does that compare to splitting it across a team?”
Transition
Let's look at HOW agents actually talk to each other...
2
Agent Communication Patterns
5:15 – 5:40 PM 25 min
--:--
Key Talking Points
  • Three communication patterns: Message Passing, Shared Context, Orchestration.
  • Message passing is the simplest — Agent A's output becomes Agent B's input. That's literally it.
  • Shared context = all agents read/write a shared workspace. More flexible but harder to debug.
  • Orchestration = a manager agent decides who works on what. This is what Claude Code's team feature uses.
  • Key insight: Structured JSON handoffs > free text. Show why with Pipeline Demo.
Demo Cues
OPEN: Pipeline Demo — Researcher → Analyst → Editor trace
Ask the Class
Which pattern would you use for customer support? Why?
Transition
Now it's YOUR turn to build one...
3
Hands-On Build
5:40 – 6:25 PM 45 min
--:--
Key Talking Points
  • Phase 1 (first 15 min): Walk through building the Researcher agent TOGETHER. Show the system prompt, output format, test it.
  • Phase 2 (next 15 min): Students build the Analyst agent. Walk the room. Common issue: analysts just rephrase instead of adding value.
  • Phase 3 (last 15 min): Add Editor, wire pipeline, run end-to-end. Celebrate first successful pipelines!
Demo Cues
SWITCH TO: Google Colab OPEN: workshop-timer.html
Tips for Walking the Room
  • Check that Researcher output is STRUCTURED (not free text)
  • Make sure Analyst actually ADDS patterns, doesn't just summarize
  • Editor should CITE specific facts when fact-checking
Transition
Take a 10-minute break. You've earned it — you just built an AI team!
4
Break
6:25 – 6:35 PM 10 min
--:--
Key Talking Points
  • While they're on break, check completion: How many got the full pipeline running?
  • Tease what's coming: "After break, we're going to break your pipeline on purpose."
  • The telephone game — information gets distorted as it passes between agents.
Quick Check
Hands up if your pipeline is producing output!
Transition
Welcome back. Your pipeline works. Now let's see what happens when information passes through multiple agents...
5
The Telephone Game
6:35 – 7:05 PM 30 min
--:--
Key Talking Points
  • Ethics moment: "Run the same question 3 times. Do you get the same answer? Why not?"
  • The telephone game analogy: facts change subtly at each handoff.
  • Four distortion types: Generalization, Amplification, Omission, Fabrication.
  • Accountability question: "If Agent C gives a wrong answer based on Agent A's research, who's responsible?"
  • Three safeguards: Citation chains, confidence propagation, human review gates.
Demo Cues
OPEN: Lecture Guide — Section 6 (Telephone Game)
Ask the Class
In a hospital, would you trust a multi-agent system for diagnosis? Why or why not?
Who is liable when an AI pipeline produces harmful output?
Transition
So when IS multi-agent worth the complexity tax?
6
When Multi-Agent Is Worth It
7:05 – 7:20 PM 15 min
--:--
Key Talking Points
  • The complexity tax: each agent adds latency, cost, and failure points.
  • Decision framework: Use multi-agent when tasks need different expertise, quality review is critical, or adversarial checking is needed.
  • Stick with single-agent for straightforward tasks, when speed matters, or when the pipeline is linear and simple.
  • Real examples: Content creation (multi wins), simple Q&A (single wins), code review (debate wins).
Demo Cues
OPEN: Lecture Guide — Section 7 (Decision Matrix)
Ask the Class
“For your final project industry — would a multi-agent system help? What agents would you need? Or would a single agent with good tools be enough?”
Transition
Let's zoom out and see how far we've come...
7
Wrap-Up + Homework
7:20 – 7:30 PM 10 min
--:--
Key Talking Points
  • The stack: Tools (S4) → Real APIs (S5) → ReAct (S6) → RAG (S7) → Multi-Agent (TODAY).
  • You started by teaching Claude to use a calculator. Now you've built a team of specialized agents.
  • Session 9 = YOUR industry. Pick your department, build your agent.
Homework Preview
  • Extend: Add a 4th agent (Fact-Checker, Summarizer, or Domain Expert)
  • Analyze: Run pipeline 5×, document distortion
  • Design: Multi-agent system for YOUR industry
  • Read: Claude Code's team feature
Demo Cues
OPEN: homework-guide.html
Closing
Next week, everything you've built comes together for YOUR career.