Building Multi-Agent Systems with OpenAI Agents SDK

I led a 2.5-hour hands-on technical workshop on building multi-agent AI systems for the Code with Ahsan community. The session focused on designing agent-based architectures where multiple specialized agents collaborate, delegate responsibilities, and interact with external tools to produce reliable, real-world results.

The workshop started with core fundamentals, what an agent is, why a single agent is often insufficient, and how breaking responsibilities into multiple agents leads to more scalable and maintainable systems. From there, we built a complete multi-agent system live through step-by-step coding.

Key concepts covered included:

• Creating and executing agents
• Using a classification agent as an orchestration layer
• Agent handoffs and responsibility delegation
• Tool definition and controlled tool usage
• Input and output guardrails for safer, more predictable behavior

To apply these concepts, we built a practical application that solves a common real-life problem: choosing what to watch when time and preferences are limited. The system processes user input, classifies intent, routes tasks through the appropriate agents, and returns recommendations based on a structured dataset.

Participants were provided with a complete repository containing a guided, step-by-step implementation with TODOs and reference solutions, enabling them to reuse the architecture for multiple real-world use cases beyond the workshop.