Course Catalog
Agent-based Modeling and Agentic Technology
94-815
Units: 6
Description
No pre-requisites
Agentic AI has crossed a threshold from clever demos to systems that can act: tool-using agents that plan and execute workflows, computer-use agents that operate real interfaces, coding agents that write, run, and verify code, and multi-agent teams that coordinate to solve larger tasks. In the past year, the field has matured into production-grade agent stacks with interoperability standards, evaluation pipelines, and operations practices that look more like modern software engineering than prompt tinkering. This course is built around that shift.
Agentic Technologies bridges classical agent-based modeling (ABM) with the 2026 agent engineering toolkit. ABM is the course’s “systems lens,” it gives you durable ways to reason about emergence, feedback loops, heterogeneity, and systemic risk, exactly the phenomena that show up when LLM-powered agents interact at scale. Using that lens, we study and apply practical design patterns for single-agent and multi-agent architectures, governed tool surfaces and interoperability (including MCP-style tool ecosystems), memory and RAG as an engineering discipline, trace-based debugging using real failure-mode taxonomies, and evaluation science that measures success alongside cost, latency, and uncertainty.
Agentic Technologies is deliberately hands-on and practice-driven, while remaining accessible to students coming from strategy, policy, and management as well as AI engineering and data science. You will learn to frame real agentic AI opportunities, translate them into concrete system requirements, and make design choices that balance capability, cost, reliability, and risk. You will define success criteria up front, build and test agents in realistic scenarios, and use traces and evaluations to diagnose failures, improve performance, and justify decisions. Along the way, you will develop the professional artifacts that matter in the real world, architectures, evaluation plans, governance controls, and deployment-ready documentation, so you can confidently lead or contribute to agentic AI efforts in organizations and public-sector settings.
Learning Outcomes
Upon completion of this course, students will be able to:
- ABM as the systems lens: Explain and model how heterogeneity, interaction rules, and feedback loops produce emergent behavior, and use ABM concepts to reason about agentic AI at system scale.
- From opportunity to requirements: Translate real agentic AI use cases into clear objectives, constraints, and success metrics, balancing capability, cost, reliability, safety, and stakeholder accountability.
- Design agent architectures and protocols: Design single-agent and multi-agent architectures using established patterns, specify coordination and termination protocols, and justify architectural choices using anticipated failure signatures.
- Engineer governed tool and memory surfaces: Design least-privilege tool interfaces and memory/RAG pipelines, define access controls and retention policies, and mitigate common failure modes such as injection, leakage, and stale retrieval.
- Diagnose failures with traces and taxonomies: Instrument agents to produce auditable traces, classify failures using a structured taxonomy, and propose targeted mitigations that improve robustness without adding unnecessary complexity.
- Evaluate agents with statistical discipline: Construct evaluation harnesses that measure task success alongside latency and cost, quantify variance with repeated trials and confidence intervals, and distinguish real gains from noisy ties.
- Operationalize agents with safety and AgentOps: Specify deployment-ready controls, monitoring, incident response, and rollback criteria, and produce professional artifacts (architecture, eval plans, governance controls, evidence logs) suitable for enterprise and public-sector review.
Prerequisites Description
No pre-requisite courses. Introduction to analytics, data science, and generative AI would be useful.