04-801-W2 Agentic AI: Foundations and Applications
Location: Africa
Units: 6
Semester Offered: Fall
Course description
Agentic AI systems, built on large language models that plan, use tools, retrieve information, and coordinate with other agents, are redefining how intelligent software is built. This seven-week mini-course takes students beyond prompt engineering to the engineering of goal-directed agents: reasoning-and-acting loops, robust tool integration, retrieval, and multi-agent orchestration with current frameworks such as LangChain, LangGraph, and AutoGen. Equal weight goes to making agents trustworthy: graceful degradation, systematic evaluation, hallucination control, and responsible deployment. Each week pairs one core concept and live demo with a hands-on lab, and every lab contributes a building block to a team capstone. The course is anchored in an applied domain of regional relevance (Fall 2026: agentic decision support for smallholder agriculture in East Africa, built on real open data) and culminates in a working multi-agent system, a live technical defence, and a documented design rationale.
Learning objectives
In this course, students will:
- Explain what distinguishes agentic systems from single-call LLM applications and judge when an agentic design is warranted
- Implement reasoning-and-acting agents with structured, machine-checkable outputs
- Build robust tools that connect agents to real-world APIs and retrieval components
- Design and justify multi-agent architectures, selecting the simplest orchestration that meets the requirements
- Evaluate agent behavior systematically and diagnose failure modes such as hallucination
- Assess the responsible-deployment implications of agentic systems in resource-constrained settings
- Document and defend architectural decisions and trade-offs
Outcomes
After completing this course, students will be able to:
- Build and debug tool-using, multi-step LLM agents in hands-on labs
- Deliver, in a team, a working agentic system on real-world data, including an evaluation harness
- Present, defend, and critically discuss architecture, design rationale, evaluation evidence, and ethical impact
Content details
Week 1: Introduction to agentic AI. Why agents now; from prompt engineering to agentic architectures; live demo; first tool-using agent.
Week 2: Reasoning and planning with LLMs. Chain-of-thought, ReAct, structured outputs; capstone launch and team formation.
Week 3: Tool use and retrieval. Anatomy of a robust tool, authentication and error handling, graceful degradation; retrieval and vector stores; LangChain.
Week 4: Multi-agent orchestration. Role specialization, task routing, coordination patterns (LangGraph, AutoGen, CrewAI); team architecture lab.
Week 5: Evaluation and responsible deployment. Measuring agent performance, LLM-as-judge and its pitfalls, hallucination mitigation, reflection loops; logging and cost monitoring of agent runs; deployment ethics; capstone sprint.
Week 6: Graded team presentations: live demo, architecture, evaluation, design rationale.
Week 7: Debrief, futures, and societal impact of agentic AI.