Track 10: AI Agents

What actually makes a system "an agent" rather than a chatbot — and you build one by hand here, with no framework, so that track 11's abstractions are recognizable rather than magic. Covers the ReAct loop, tool schemas and execution, planning, memory, reflection, and the failure modes (runaway loops, cost blowups) that bite every real deployment.

Modules

1. What Makes Something an Agent
2. Agents vs. Chatbots vs. Workflows
3. The ReAct Loop
4. Building a ReAct Agent From Scratch
5. Tools: Definition and Schemas
6. Tool Calling With the OpenAI API
7. Tool Execution and Error Handling
8. The Agent Loop: Termination and Limits
9. Planning: Task Decomposition
10. Planning: Plan-and-Execute
11. Memory: Short-Term and Scratchpad
12. Memory: Long-Term and Vector-Backed
13. Reflection and Self-Correction
14. Human-in-the-Loop Patterns
15. An Agent Design Pattern Catalog
16. Agent Failure Modes and Debugging
17. Cost and Latency Control in Agents
18. Capstone Project