Build production AI agents with tools & MCP, multi-agent workflows, realtime voice, and eval-driven deployment
AI agents can do far more than return a block of text. They can use tools, retrieve live information, collaborate with specialist agents, remember context, ask for human approval, and communicate through voice.
What you’ll learn
- Build and run text, streaming, voice, and realtime AI agents in Python with the OpenAI Agents SDK..
- Extend agents with function tools, hosted tools, local runtime tools, web search, and MCP servers..
- Design multi-agent workflows using specialist agents, agents as tools, handoffs, memory, and shared context..
- Apply prompt engineering, context engineering, structured outputs, and model settings to improve agent behavior..
- Add input and output guardrails, human approval, and configurable safety controls to agent workflows..
- Build a production AI Travel Assistant with Chainlit, live progress updates, rich artifacts, and realtime voice..
- Trace, evaluate, red-team, deploy, and monitor agents using Promptfoo, Hugging Face Spaces, and Langfuse..
- Turn production failures into repeatable offline evaluation and regression test cases..
- Build ChatKit applications and equip agents with sandboxed workspaces and reusable skills on demand..
Course Content
- Start Here: Course Roadmap & Resources –> 1 lecture • 2min.
- OpenAI Agents SDK Fundamentals –> 5 lectures • 45min.
- Tools & External Capabilities –> 4 lectures • 34min.
- Multi-Agent Workflows –> 2 lectures • 16min.
- Guardrails & Agent Safety –> 2 lectures • 18min.
- Voice & Realtime Agents –> 2 lectures • 19min.
- Build a Production AI Travel Assistant I: Core Chat Experience –> 4 lectures • 31min.
- Build a Production AI Travel Assistant II: Prompting, Context & Human Control –> 5 lectures • 52min.
- Build a Production AI Travel Assistant III: Multi-Agent Architecture –> 3 lectures • 38min.
- Build a Production AI Travel Assistant IV: Product Experience & Safety –> 4 lectures • 52min.
- Productionize Your AI Agent: Evals, Deployment & Monitoring –> 4 lectures • 39min.
- Extensions: ChatKit, Sandbox Agents & Skills –> 3 lectures • 24min.
- Your Next Steps –> 1 lecture • 1min.

Requirements
AI agents can do far more than return a block of text. They can use tools, retrieve live information, collaborate with specialist agents, remember context, ask for human approval, and communicate through voice.
In this hands-on course, you will learn how to build these systems in Python using the OpenAI Agents SDK. We begin with focused lessons covering the core SDK before combining everything into a production-style AI Travel Assistant.
You will learn how to:
- Build, configure, run, and trace OpenAI agents
- Stream responses and return validated structured outputs
- Add function tools, hosted tools, local runtime tools, and MCP servers
- Create multi-agent workflows using agents as tools and handoffs
- Manage memory, prompts, user context, and self-updating profiles
- Add input guardrails, output guardrails, and human approval
- Build voice pipelines and continuous realtime voice conversations
- Evaluate, red-team, deploy, and monitor an agent application
The main project is a browser-based AI Travel Assistant built with Chainlit. It searches for current weather and flights through MCP, finds activities through web search, remembers user preferences, and delegates work to Weather, Flights, and Places specialists.
You will improve the user experience with live tool-progress updates, customer-support handoffs, reasoning summaries, a rich trip-plan artifact, and realtime voice. You will then protect the application with configurable guardrails and human approval for sensitive actions.
Finally, you will create offline evaluations with Promptfoo, red-team the agent with adversarial inputs, deploy it to Hugging Face Spaces, and monitor production conversations with Langfuse.
The advanced section introduces ChatKit, Sandbox Agents, isolated workspaces, and reusable skills loaded on demand.
This course is designed for developers with basic programming experience. No previous experience with the OpenAI Agents SDK, MCP, multi-agent systems, voice AI, or agent evaluations is required.
By the end, you will have a complete, deployed AI agent project and the practical foundation to design, build, test, and ship your own text and voice agents.