Course Catalog
Agent-based Modeling and Agentic Technology
94-815
Units: 6
Description
Agentic Technologies teaches the fundamentals of how AI agents work, why they fail, and how to evaluate, secure, and govern them. The course starts with agent-based modeling as a vocabulary for emergence, feedback, and systemic risk in populations of interacting agents. Students then learn how a language model becomes an agent, the architecture and design patterns of single agents, the tools, skills, and memory an agent uses, how multi-agent systems are connected through shared protocols, and the loops that let agents work for long stretches without a person directing each step. The final part covers how to evaluate agents for reliability and cost, what coding agents, AI scientists, and computer-use agents can and cannot do, and an overview of how to secure, govern, and operate agents that act with delegated authority. Sessions mix lectures and in-class exercises. Students complete an individual project, a productivity agent for their own work, and a team project, a multi-agent system for an application such as policy analysis or gaming. Both projects must demonstrate the value the system delivers, and students take one individual quiz. The course is offered in a programming track and a no-code track and is restricted to students in the AIM program.
Learning Outcomes
- Students will be able to explain how heterogeneity, interaction rules, and feedback loops produce emergent behavior, and apply agent-based modeling concepts to analyze populations of interacting AI agents.
- Students will be able to explain how a language model becomes an agent through tool calling and an iterative loop, and how context, tools, skills, and memory are assembled at each step.
- Students will be able to choose single-agent and multi-agent design patterns and protocols, including MCP and A2A, only where a task needs them, and state what each costs in latency, money, opacity, and governance work.
- Students will be able to specify a loop with a verifiable definition of done, checker agents, memory, triggers, and budgets, and explain how each element limits failure over long runs.
- Students will be able to design evaluations that combine code, model, and human graders, measure consistency across repeated runs along with cost and latency, and use benchmark and field evidence to judge what coding agents, AI scientists, and computer-use agents can and cannot do.
- Students will be able to identify the main security risks for agents, including prompt injection, and outline the identity, governance, and operational controls needed to deploy an agent with delegated authority.
- Students will be able to build an agentic system, demonstrate the value it delivers, and present it to technical, managerial, and policy audiences.
Prerequisites Description
Restricted to students in the Master of Science in Artificial Intelligence Systems Management (AIM) program.