Report a bug

What went wrong?

AI Agents & Tool Use — AI Academy

AI Academy/Courses/AI Agents & Tool Use

⌘K

UK Sign in Sign up

AdvancedИнженерия

AI Agents & Tool Use

Complete course on building AI agents: from basic tool use via Claude and OpenAI to complex multi-agent systems. Covers function calling, MCP protocol, agent frameworks (LangChain, AutoGen, CrewAI), memory patterns, planning, human-in-the-loop, and production agent deployment.

4modules

16lessons

525 mintotal time

Developers building autonomous AI systemsaudience

Progress

0%0 / 16 lessons

Sign in to save your progress.

Module 1

Tool Use Fundamentals

What tool use is, how the model decides to call a tool vs answer directly, the tool schema, and the differences between Claude, OpenAI and Gemini.

What tool use is

Tool use lets the model call your functions instead of hallucinating an answer. We cover JSON Schema, the call loop, and Claude / OpenAI / Gemini differences.

Function calling in Claude

The full tool_use loop in Claude: tool_use and tool_result blocks, parallel calls, and forced tool selection via tool_choice.

Function calling in OpenAI

OpenAI's tools array, tool_calls and tool_choice, parallel calls in GPT-4o, and working with tools during streaming.

4

Designing good tools

Single responsibility, clear descriptions, typed parameters, the 'tool vs system-prompt knowledge' boundary, and tool error handling.

30 min

Module 2

Agent Architectures

ReAct, planning, memory, and the MCP protocol — the foundational patterns for building a single agent.

The ReAct pattern from scratch

Implement the Reason–Act–Observe loop without a framework: thought, action, observation, stopping conditions, max iterations, and tool failure handling.

Plan-and-Execute and task decomposition

The plan-and-execute pattern: task breakdown, a dependency graph, replanning on failures, and structured output for the plan.

In-context memory and history management, external memory via a vector store, episodic vs semantic memory, and summarization.

Model Context Protocol (MCP)

What MCP is, the client/server architecture, Claude Desktop + MCP servers, a simple MCP server, and tools / resources / prompts.

Module 3

Multi-Agent Systems

Orchestrator + subagents and three popular frameworks: LangChain, AutoGen, CrewAI.

Orchestrator and subagents

The orchestrator + subagents pattern, when multiple agents are justified, task routing, and result aggregation.

Agents in LangChain

AgentExecutor, create_tool_calling_agent, custom tools via @tool, and LangSmith tracing for agents.

AutoGen: ConversableAgent and GroupChat

ConversableAgent, AssistantAgent and UserProxyAgent, GroupChat and nested chats, a code-execution agent.

CrewAI: Crew, Agent, Task, Process

CrewAI role-based agents, Task with expected output, sequential vs hierarchical process, and passing context between tasks.

Module 4

Production Agents

Human-in-the-loop, observability, security, and deploying agents to production.

Human-in-the-loop

When to pause for human approval, interrupt patterns, approval gates, and asynchronous agent workflows.

Agent observability

LangSmith and Langfuse, tracing tool calls, per-agent cost tracking, debugging failed calls, and replay.

Prompt injection in tool inputs, tool result poisoning, sandboxed code execution, rate limiting agent actions, and scope limitation.

Deploying agents

Stateless vs stateful agents, state persistence (Redis, Postgres), long-running and webhook agents, deploying on Modal and Fly.io.

AAI AcademyEN Start learningCatalog

Overview

AI Academy Progress Achievements Leaderboard Insights Notes Notifications Team API Pricing Settings

Learning

Tracks8 Courses51

Libraries

Prompts30 Workflows12 Tools22 Compare Blog

Updated · May 14, 2026