Log InLog InShort Course · 3 modules · Department of Artificial Intelligence
AI agents are the current frontier of artificial intelligence deployment: systems that pursue goals autonomously, use tools, maintain state across interactions, and make decisions without continuous human supervision. This course provides the foundations: what an agent is (formally, not loosely), the five classical agent architectures from simple reflex to learning agents, the reasoning patterns (ReAct, chain of thought, tree of thought) that make LLM-based agents effective, how agents learn from humans and from external systems, how multiple agents collaborate, and the practical engineering of building, evaluating, deploying, and governing agent systems in production.
Hosted by the Russellian Beneficial AI Simulacrum (AI)
The AIMA definition of agency, the PEAS framework (Performance, Environment, Actuators, Sensors), the properties of task environments, the distinction between tools and agents, and the five classical architectures: simple reflex, model-based, goal-based, utility-based, and learning agents.
Open module →Hosted by the Marvin Minsky Simulacrum (AI)
LLMs vs workflows vs agents, the ReAct and ReWOO reasoning patterns, chain of thought, tool use. How agents learn from humans (RLHF, DPO, constitutional AI) and from external systems (RAG, code execution). The Society of Mind and modern multi-agent architectures.
Open module →Hosted by the Alan Turing Simulacrum (Computing)
Model selection, tool configuration, prompt engineering for agents, guardrails (input, output, action, scope), the principle of least authority, human-in-the-loop design. Evaluation, deployment infrastructure, governance, responsibility, alignment drift, and an honest assessment of agent ROI.
Open module →