Agentic AI Problem Set
20 Problems to Test Your Knowledge
How many can you get right?
I developed this problem set for my students to study the 20 key terms for Agentic AI that became popular in 2025.
In the spirit of AI by Hand ✍️, here is the design philosophy behind this problem set:
Problem-Driven. Every term begins with a problem statement—what motivated this idea in the first place.
Think-First: Instead of giving away the term immediately, you first get a multiple-choice question so you can think about which idea (term) best solves that problem.
Solution: On the next page, I reveal the correct term and explain how it addresses the problem.
Industry Insights: As a university professor, I don’t have industry experience. I’m fortunate to collaborate with Ofer Mendelevitch (Vectara), who contributes real-world industry perspectives.
Hope you enjoy this problem set! 😊
~ Prof. Tom Yeh
Terms
LLM Agent
Tool Use
Function Calling
Reason-and-Act (ReAct)
Chain of Thought
Agent Loop
Reflection
Critic
Plan-and-Execute
Episodic Memory
Semantic Memory
Context Selection
RAG (Retrieval-Augmented Generation)
Tree of Thought
Graph of Thought
Graph RAG
Delegation
Orchestration
Model Control Protocol (MCP)
Safety Guardrails
Really useful refresher. I’ve been working with agent-style systems since before “agentic AI” became the term, and the problem-first approach here is exactly what helps uncover blind spots once you think you already know this stuff. Worth revisiting even for experienced builders.