Contact Us
Artificial Intelligence

Developing Agents: an Introduction

Group Training
Course Outline

Overview

This course provides a practical, code-first understanding of modern agentic systems, AI systems capable of reasoning, planning, and acting through structured workflows and tool use. Participants begin by distinguishing LLMs, agents, and full agentic systems, then explore how agency emerges through multi-step workflows including prompt chaining, routing, parallelization, reflection, and orchestration-worker patterns. The course then focuses on tool use, covering tool calling formats, execution models, and the Model Context
Protocol (MCP) for secure and interoperable tool integration. Finally, participants learn to design and evaluate multi-step planning agents using patterns such as ReACT, Reflexion, ReWOO, and tree-search-based planning strategies. Hands-on examples and guided labs ensure learners can apply these patterns to build reliable, transparent, and extensible agentic systems.

Duration

1 day

Who Should Take This Course

Audience

ML/AI Developers, Software & Data Engineers, IT & QA Staff, Technical Managers.

Prerequisites

Participants should have basic skills in Python, experience with machine learning and an understanding of LLMs.

Why You Should Take This Course

Developing Agents: an Introduction ensures participants are able to:

  • Explain how agency emerges from planning, memory, and structured workflows
  • Build grounded practical agents
  • Design and implement agents with multi-step workflow patterns
  • Integrate and manage tools within agent systems by using tool calling, structured I/O schemas, and
    MCP

Course Outline

Developing Agents: an Introduction

1. Agents and Agentic Systems

  • Distinguish LLMs, agents, and agentic systems
  • Levels of agency: single-step, multi-step, autonomous planning
  • Agent memory types: scratchpad, episodic, long-term / vector
  • Execution models: looped reasoning, event-triggered, goal-driven
  • Evaluation principles: correctness, robustness, alignment, traceability
  • LAB: Building a grounded Agent with web search

2. Workflow Patterns

  • Prompt chaining for structured multi-stage reasoning
  • Routing patterns (rule-based, embedding-based, and model-driven)
  • Parallelization strategies (branching, voting, map-reduce for reasoning)
  • Reflection loops for error detection and quality improvement
  • Orchestration vs. worker roles (controller – sub-agent execution)
  • LAB: Workflow Patterns for Agentic Systems

3. Tools

  • Tool calling formats
  • Designing reliable tool interfaces (inputs, outputs, error handling)
  • External system control: APIs, databases, and execution environments
  • MCP (Model Context Protocol)
  • Safety considerations: sandboxing, rate limiting, observability
  • LAB: Integrating Tools

4. Multi-Step & Planning Agents

  • ReACT pattern
  • Reflexion, ReWOO, and self-correcting agent loops
  • Planner–executor pattern for delegated task execution
  • Tree-search reasoning
  • Performance tuning: hallucination reduction, cost control, latency optimization
  • LAB: Multi-Step & Planning Agents
Search UMBC Training Centers