Agentic AI

Course Description

This course builds from first principles to the infrastructure frontier of AI agents, exploring what AI agents are, what makes a system truly "agentic," and how autonomous agents are scheduled, secured, orchestrated, and managed at the systems level. Students develop a multi-tool, memory-equipped planning agent and progressively extend it into a multi-agent, protocol-aware, proof-of-concept (POC) system. The course culminates in a "toy agent kernel" project that applies operating-system concepts to agent orchestration, giving students practical insight into the architecture and engineering principles behind increasingly autonomous AI systems.

Learning Outcomes At the conclusion of the course, you should be able to

• Differentiate agentic systems from chatbots, workflows, and prompt chains, and explain the foundations of agentic systems.
• Implement LLM agents with tools, memory, and multi-step planning using APIs and frameworks.
• Architect multi-agent systems and design agent communication using protocols such as Model Context Protocol (MCP) and Agent2Agent (A2A).
• Evaluate agentic systems for performance, safety, security, and governance.
• Design and build a POC-grade agent architecture incorporating scheduling, memory, and access control.

Prerequisites/Skills Needed: Python, Jupyter Notebooks, GCP, REST APIs, JSON, basic git workflows and basic Project Management skills.