Agentic AI 

• Overview: A hands-on technical course covering the full lifecycle of agentic AI systems, from foundational concepts through multi-agent orchestration, production deployment, and enterprise governance. Participants build working agents throughout the course, culminating in a team capstone.

• Objectives:

  • Explain the architecture and reasoning patterns underlying agentic AI systems

  • Design and implement single and multi-agent solutions using current frameworks

  • Apply memory, tool use, and orchestration patterns appropriate to the problem

  • Evaluate, instrument, and secure an agent for production deployment

  • Address governance, compliance, and accountability requirements for enterprise adoption

• Length: 3 days

• Target Audience: Software engineers, data engineers, and solution architects with Python proficiency and working knowledge of REST APIs. Familiarity with LLMs helpful but not required.

Module 1: Agentic AI Concepts and Architecture

• What makes a system agentic

• The perception–reasoning–action loop

• Agents vs. assistants vs. pipelines

• Anatomy of an agent: model, tools, memory, state

Module 2: LLM Capabilities and Limitations for Agents

• Reasoning modes: CoT, ReAct, plan-and-execute

• Context window management

• Hallucination and reliability patterns

• Calibrating when to trust model output

Module 3: Tool Use and Function Calling (Lab)

• OpenAI/Anthropic tool schemas

• Defining tools and handling responses

• Error routing and retry strategies

Module 4: Memory Architectures

• In-context vs. external vs. semantic memory

• Vector stores and RAG in an agentic context

• Episodic and procedural memory patterns

• When to use each memory type

Module 5: Orchestration Frameworks Deep Dive (Lab)

• LangGraph architecture: nodes, edges, state machines

• Graph construction and conditional routing

• Cycles and looping behavior

Module 6: Multi-Agent Patterns (Lab)

• Supervisor–worker, peer-to-peer, hierarchical delegation

• Agent handoffs and shared state

• CrewAI vs. AutoGen vs. LangGraph for multi-agent

Module 7: Long-Horizon Task Execution

• Task decomposition strategies

• Checkpointing and resumption

• Human-in-the-loop patterns

• Handling failure mid-chain

Module 8: Prompt Engineering for Agents

• System prompt design for reliability

• Role and persona framing

• Output formatting constraints

• Preventing prompt injection attacks

Module 9: Evaluation and Observability (Lab)

• Task completion, faithfulness, and tool efficiency metrics

• Tracing with LangSmith/Arize

• Cost tracking and token budgeting

Module 10: Security, Safety, and Guardrails

• Prompt injection and adversarial inputs

• Least-privilege tool design

• Output validation and content filters

• Blast radius containment for autonomous actions

Module 11: Deployment and Production Patterns

• Containerizing agents

• Async execution and queuing

• Stateful vs. stateless deployment

• Latency vs. cost trade-offs

Module 12: Governance, Ethics, and Enterprise Readiness

• Accountability in autonomous systems

• Audit trails and explainability requirements

• EU AI Act and emerging compliance frameworks

• Building an AI governance charter

Module 13: Capstone — End-to-End Agent Build (Capstone)

• Teams design and build a multi-tool agentic solution

• Scored on architecture, reliability, and observability