AI-500: Designing and Implementing Multi-Agent AI Solutions

About this course

This course contains the use of artificial intelligence

Most teams run one AI agent. This course teaches you to design and build a whole system of agents, and to run it in production.

You join Larkspur Mutual, an insurer whose claims-intake agent is doing well, until the business asks for more than one agent can carry. Across seventeen modules you follow one company and one claim platform: you design stateful agent loops, coordinate teams of agents, connect them to tools and data, secure and govern them, measure them, control their cost, and handle the night something goes wrong.

What you build

Design: stateful agentic loops with checkpoints and resume, orchestration patterns (hub and spoke, hierarchical, parallel), task decomposition and agent collaboration, and agent-to-agent communication with A2A.Build: production prompting for agents, tool ecosystems with MCP, advanced retrieval with Azure AI Search, agent memory in Azure Cosmos DB, and CI/CD with GitHub Actions.

Secure and govern: zero-trust security for agents, responsible AI governance, and agent lifecycle governance from first version to retirement.

Operate: observability across agents, evaluation of whole claims, staying within budget, human approval in the loop, and incident response at three forty-seven in the morning.

How every module worksA short Larkspur chapter sets up the problem. The concept builds up on screen as a diagram, with the design decision and its trade-offs. Then you see real Python code with the current SDK names. Hands-on lab videos follow, where every step is run on camera, and portal demos show the same ideas in the Azure portal and Microsoft Foundry.

Read the full description on Udemy on Udemy (opens in a new tab)

Prices, discounts and availability are set by Udemy. We may earn a commission when you purchase through links on this site.