Week 10 – Agentic AI and Autonomous Systems

Objectives

  • Understand the architecture and capabilities of agentic AI systems.
  • Build agents that use tools, memory, and multi-step reasoning.
  • Evaluate trade-offs in agent design: reliability, cost, and autonomy.

Topics

  • What makes a system "agentic": planning, tool use, and feedback loops.
  • Large language model fundamentals for developers.
  • Tool/function calling and structured outputs.
  • Memory patterns: in-context, external retrieval, and persistent state.
  • Multi-agent coordination and orchestration frameworks.
  • Prompt engineering for reliability and instruction-following.
  • Observability, evaluation, and failure modes in agentic pipelines.
  • Safety considerations: human-in-the-loop, scope limits, and guardrails.

Hands-On Activities

  • Build a tool-using agent that can query an API and summarize results.
  • Implement a multi-step reasoning pipeline with error recovery.
  • Add memory to an agent using a vector store or key-value store.
  • Evaluate agent outputs against a set of expected behaviors.

Deliverables

  • Working agentic application with at least two integrated tools.
  • Evaluation report documenting success rates and failure cases.

Assessment

  • Live agent demo handling an unseen multi-step task.
  • Code review focused on prompt design and tool integration.