Module 21: AI Agents: Tool Use, Planning & Reasoning

Chapter Overview

AI agents represent a fundamental shift in how we use large language models. Rather than generating a single response to a single prompt, agents operate in loops: they perceive their environment, reason about what to do next, take actions using tools, and observe the results before deciding on their next step. This perception-reasoning-action cycle transforms LLMs from passive text generators into active problem solvers that can browse the web, write and execute code, query databases, and orchestrate complex multi-step workflows.

This module covers the complete landscape of AI agents. It begins with the foundational concepts that distinguish agents from simple chains and workflows, introducing the four core agentic patterns: reflection, tool use, planning, and multi-agent collaboration. It then dives deep into tool use and function calling across major providers, covering everything from basic schema definitions to the Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication standards. The module explores planning and reasoning strategies, from simple ReAct loops to sophisticated tree search algorithms. Finally, it addresses code generation and execution agents that can write, run, debug, and iterate on code autonomously.

By the end of this module, you will be able to build agents that use tools effectively, plan multi-step solutions, generate and execute code safely, and handle the practical challenges of token budgets, error recovery, and human oversight.

Learning Objectives

Explain the perception-reasoning-action loop and how it distinguishes agents from chains and workflows
Implement function calling and tool use with OpenAI, Anthropic, and open-source models
Design tool schemas with clear descriptions, parameter validation, and error handling
Build ReAct agents that interleave reasoning and action in a structured loop
Apply planning strategies including plan-and-execute, reflection loops, and tree search
Implement human-in-the-loop checkpoints for high-stakes agent decisions
Build code generation agents with sandboxed execution environments
Manage token budgets, state persistence, and graceful degradation in long-running agent tasks
Evaluate agent performance using task completion, efficiency, and safety metrics

Prerequisites

Module 09: LLM APIs (chat completions, message formatting, system prompts)
Module 10: Prompt Engineering (few-shot prompting, chain-of-thought, structured outputs)
Module 20: Conversational AI (multi-turn dialogue, memory management)
Familiarity with Python async programming and JSON schema definitions
Basic understanding of REST APIs and command-line tools

Chapter Overview

Learning Objectives

Prerequisites

Sections