Mastering Agents with OpenAI Agents SDK & OpenAI Codex

Master the OpenAI Agents SDK and build production-ready AI agents from the ground up, then build every one of them hands-on with OpenAI Codex as your AI pair programmer. This comprehensive course takes you from your very first agent to deploying realtime, voice, and sandboxed multi-agent systems that integrate with real tools and external services.

Whether you're a Python developer stepping into agentic AI, or an experienced engineer who wants to ship reliable agents to production, this course gives you everything you need: the core SDK primitives, the patterns that make agents safe and observable, and a Codex-driven workflow that lets you build faster than ever.

What makes this course different: you don't just watch, you build. Every lab is constructed with OpenAI Codex, so you learn both the OpenAI Agents SDK and how to drive an AI coding agent to scaffold, implement, test, and ship real applications.

What You'll Learn, Core SDK & Agent Design:

• Foundations – Understand what agents are, the agent loop, and how the OpenAI Agents SDK fits together

• Prompts & Structured Output – Write effective instructions and get typed, structured results you can rely on

• Model Settings – Configure models, providers, and transport for the behavior and cost profile you need

• RunContext & Dependency Injection – Pass typed context and dependencies cleanly into your agents

• Run Loop, RunResult & the REPL – Run agents, stream output, inspect results, and iterate fast

• Hosted & Function Tools – Extend agents with built-in tools and your own Python functions

• Agents as Tools – Compose specialized agents by calling one agent from another

• Guardrails & Human Review – Add input/output guardrails and human-in-the-loop control for safety

• Sessions & Memory – Persist conversation state, and move beyond SQLite to production session storage

• Handoffs & Multi-Agent Orchestration – Route work across multiple agents to solve complex tasks

• MCP & Connectors – Connect agents to external systems through the Model Context Protocol

• Monitoring & Tracing – Trace, monitor, and debug agent runs for full production visibility

• Realtime & Voice Agents – Build realtime voice agents and chain multi-step voice workflows

• Sandbox Agents – Master SandboxAgents: manifests, capabilities, providers, mounts, credentials, memory, and state composition

• Bonus: AWS Bedrock AgentCore – Deploy an OpenAI agent to production on AWS

Hands-On Learning: 6 Production Labs Built with Codex

This isn't just lectures. You'll build real applications through comprehensive, step-by-step labs, each driven with OpenAI Codex:

• Lab 0: Setup – Set up your environment and your Codex workflow so you can build along from day one

• Lab 1: Basic Agents – Build your first working agents with instructions, structured output, and context

• Lab 2: Multi-Agent Handoffs & Agents as Tools – Orchestrate multiple agents that hand off work and call each other as tools

• Lab 3: SandboxAgents – Build a SandboxAgent with manifests, capabilities, and secure mounts

• Lab 4: SQL Analyzer Agent – Build a complete capstone agent that combines function tools, structured output, and sessions to analyze data

• Lab 5: Voice Agents – Build a realtime voice agent and a chained voice workflow

By the end, you'll be able to design, build, orchestrate, and ship production-grade OpenAI agents, and you'll have a repeatable Codex workflow to keep building long after the course ends.