AI Agentic Design Patterns with AutoGen (DeepLearning.AI)

Course Overview

This short course teaches AutoGen, an open-source framework for building applications where several LLM-powered agents talk to each other. Each agent gets a role. The agents then work through a task together, with or without a human stepping in. The instructors are Chi Wang and Qingyun Wu, who helped create the framework.

Instead of a long theory tour, the course is organized around four agentic design patterns: reflection, tool use, planning, and multi-agent collaboration. Each pattern gets a small, concrete project. By the end you will have built several working agent setups, from a comedy dialogue to a stock-analysis report. You should also be able to choose a pattern that suits a workflow of your own.

What You Will Learn

• Create agents with distinct roles using AutoGen's ConversableAgent class and make them hold a conversation.
• Chain several agent conversations in sequence, using a customer-onboarding scenario.
• Set up a "critic" agent that contains nested reviewer agents, so a draft blog post gets improved through feedback loops.
• Give agents tools they can call, shown through a chess game where both players must make legal moves.
• Have agents write and run code, including plotting stock gains and using functions you supply yourself.
• Add human feedback to agent workflows, with code generated by an LLM or code you provide.
• Run a custom group chat with a planning agent and control how the conversation passes between agents.
• Connect AutoGen to a model through an API or run it locally.

Course Structure

1. Introduction (4 min)
2. Multi-Agent Conversation and Stand-up Comedy (12 min, with code)
3. Sequential Chats and Customer Onboarding (8 min, with code)
4. Reflection and Blogpost Writing (10 min, with code)
5. Tool Use and Conversational Chess (15 min, with code)
6. Coding and Financial Analysis (17 min, with code)
7. Planning and Stock Report Generation (15 min, with code)
8. Conclusion (1 min)
9. Graded quiz (10 min)

Who Is This Course For?

It suits people who can already write basic Python and want a fast, hands-on look at agent frameworks. That includes developers, analysts, and data practitioners who are curious about automating multi-step work with LLMs.

Look elsewhere if you have never written code, because the lessons assume you can read and run Python notebooks. It is also a weaker fit if you want a broad comparison of agent frameworks. This course teaches one framework and does not survey the others. Engineers who already ship agent systems will probably find the material introductory.

Format & Time Commitment

The course is video lessons paired with code examples you can follow along with, and it totals about 1 hour 35 minutes. You can finish it in one sitting or spread it over a couple of evenings. The page lists no fixed schedule or deadlines. Add extra time if you want to experiment beyond the provided notebooks.

Pros and Cons

Pros

• Taught by two of the people behind AutoGen, so the explanations come from the source.
• Each design pattern gets its own project, which makes the ideas easier to remember than abstract description.
• The projects cover varied scenarios (conversation, writing, games, finance), showing how flexible the framework is.
• Works with different models, via an API or locally.
• Short enough to finish in a weekend.

Cons

• At under two hours, each pattern is covered at an introductory level. Expect a starting point, not production-ready design guidance.
• You need basic Python to follow along. Complete beginners to programming will struggle.
• It teaches only AutoGen. Skills transfer conceptually, but the code will not carry over directly to other agent frameworks.
• The formal accomplishment requires the PRO plan, so learners who stay on the free tier get the knowledge but not that credential.
• You will need access to an LLM, through an API or a local setup, to try your own experiments.

FAQ

How much does it cost? The page describes course access as free for a limited time during the platform's beta. The accomplishment is tied to the PRO plan.

Do I need prior AI experience? No. The course is labeled beginner and expects basic Python, not machine learning knowledge.

Is the credential worth much? Treat it as a modest résumé line showing you completed a short course. Your portfolio and the agents you build will say more than the accomplishment itself.

What is a sensible next step afterward? Try rebuilding one of the examples with your own data or task. If you want to compare approaches, look at other agent-focused short courses covering different frameworks.

If a quick, project-based introduction to multi-agent systems sounds useful, the full syllabus and current access options are on the DeepLearning.AI course page.