This is project week. You will build, test, and demo a real working AI agent. No coding required. Choose your tool based on what you want to build.
Claude.ai
claude.ai
Best for research, writing, and conversational agents. Set a detailed system prompt. Free tier is enough.
ChatGPT
chatgpt.com
Great for agents that need web browsing. Custom GPTs let you set a system prompt and share with others.
Zapier AI Agents
zapier.com/ai
Best for agents that connect to real apps — email, calendar, Sheets. No code, visual workflow builder.
Make.com
make.com
More powerful than Zapier for complex pipelines with branching logic. Steeper learning curve.
Part 2
The system prompt: the most important thing you will write
The system prompt is the background instruction that defines your agent's identity, rules, and goals. Get this right and your agent works. Get it wrong and no amount of tweaking fixes it.
Weak system prompt
"You are a helpful assistant. Answer questions."
Strong system prompt — Research Agent example
"You are a research assistant for students. When given a topic you: (1) identify the 3 most important sub-questions to answer, (2) answer each one clearly with evidence, (3) note areas where experts disagree, (4) end with 3 follow-up questions worth exploring. Use plain language. Never fabricate sources. If you are unsure about a fact, say so explicitly."
Specific role, specific process, specific output format, explicit rules about what not to do. A great system prompt is like a job description — detailed, measurable, impossible to misinterpret.
Part 3
Testing: break it on purpose
Once you have built your agent, the most important thing you can do is try to make it fail. This is how real AI engineers test — not by hoping it works, but by deliberately looking for how it breaks.
1
Test the happy path
Give it a perfect typical input. Does it do exactly what you designed?
2
Test edge cases
What if the input is empty? Very long? In a different language? Completely off-topic?
3
Try to confuse it
Ask it to do something it should not. Give contradictory instructions. See if it breaks character.
4
Fix failures in the system prompt
Almost every failure can be fixed by adding a rule, clarifying a boundary, or giving an example.
🚀 Activity — Build Challenge + Demo Day
Build your first agent. Then show it off.
Choose one of three agent types. Build it, test it using the break-it-on-purpose method, then fill in the build worksheet. You will demo it in 2-3 minutes.
🔍
Option A: Research Agent
Give it a topic → it researches, summarizes key findings, and lists open questions.
📚
Option B: Study Buddy
Tell it your subject and weak areas → it creates a personalized study plan with practice questions.
✍
Option C: Creative Agent
Give it a genre and theme → it writes a short story then critiques its own writing.
🌍
Option D: Community Impact Agent ★
Pick a real problem in your school or community → build an agent that helps, plus a one-page agent policy. The recommended capstone / portfolio track — you demo the build and the policy together.
🛡 Demo Day — The Three Accountability Questions
Every builder answers these three at their demo — they’re part of the grade. This is where “Who owns this agent’s decisions?” from Week 1 pays off, and it’s exactly the mindset real AI teams hire for.
Go Deeper — Optional
Week 5 Python Notebook
Run in Google Colab — no setup, no install. Interactive Python exercises that reinforce every concept from this lesson.